SCIENCE ADVANCES | RESEARCH ARTICLE ANTHROPOLOGY Copyright © 2021 The Authors, some High-resolution dietary reconstruction of victims rights reserved; exclusive licensee of the 79 CE Vesuvius eruption at Herculaneum by American Association for the Advancement compound-specific isotope analysis of Science. No claim to original U.S. Government Works. Distributed Silvia Soncin1*, Helen M. Talbot1, Ricardo Fernandes2,3,4, Alison Harris1,5, Matthew von Tersch1, under a Creative Harry K. Robson1, Jan K. Bakker1,6, Kristine K. Richter1,2†, Michelle Alexander1, Steven Ellis7, Commons Attribution Gill Thompson8, Valeria Amoretti9, Massimo Osanna9, Marina Caso10, Francesco Sirano10, License 4.0 (CC BY). Luciano Fattore11, Andre C. Colonese1,12, Peter Garnsey13, Luca Bondioli14,15, Oliver E. Craig1* The remains of those who perished at Herculaneum in 79 CE offer a unique opportunity to examine lifeways across an ancient community who lived and died together. Historical sources often allude to differential access to foodstuffs across Roman society but provide no direct or quantitative information. By determining the stable isotope values of amino acids from bone collagen and deploying Bayesian models that incorporate knowledge of protein synthesis, we were able to reconstruct the diets of 17 adults from Herculaneum with unprecedented resolution. Significant differences in the proportions of marine and terrestrial foods consumed were observed between males and females, implying that access to food was differentiated according to gender. The approach also provided dietary data of sufficient precision for comparison with assessments of food supply to modern populations, opening up the possibility of benchmarking ancient diets against contemporary settings where the consequences for health are better understood. INTRODUCTION biases toward any particular social class, although we know from The human remains found at Herculaneum represent a sample of a other evidence, namely, the so-called Album of Herculaneum, that “living” population who died trying to escape from the eruption of freedmen and slaves made up a high proportion of the residents of the Vesuvius volcano in 79 CE. In total, 340 individuals have been the town (3, 4). excavated from the beach and from nine adjacent fornici (stone Here, we sought to reconstruct the diets of 17 individuals from vaults) that run parallel to the seashore, where they sought shelter this catastrophic death assemblage through compound-specific (Fig. 1) (1). This remarkable assemblage of victims of a natural stable isotope analysis (CSIA) of amino acids (AAs) directly ob- catastrophe not only is of huge public interest but also offers an tained from bone collagen. The aim of this study was to quantify opportunity to substantially advance our knowledge of Roman society through the application of bioarchaeological approaches. The skeletal sample at Herculaneum is not constrained by the biases usually faced by osteoarchaeologists when dealing with attritional cemetery assemblages, such as selective mortality and burial; rather, it provides a “snapshot” of an ancient population rarely afforded in archaeology. Although some selectivity between the few who failed to evacuate the town of ca. 3000 to 4000 inhabitants and the major- ity who escaped may be expected, males, females, the old, and young are all well represented (1, 2). No evidence has emerged as yet of 1 BioArCh, Department of Archaeology, University of York, York, UK. 2Department of Archaeology, Max Planck Institute for the Science of Human History, Jena, Germany. 3School of Archaeology, University of Oxford, Oxford, UK. 4Faculty of Arts, Masaryk University, Brno, Czech Republic. 5Department of Archaeology and Classical Studies, Stockholm University, 10691 Stockholm, Sweden. 6ACASA, Uni- versity of Amsterdam, Amsterdam, Netherlands. 7Department of Classics, University of Cincinnati, Cincinnati, OH 45221, USA. 8School of Archaeological and Forensic Sciences, University of Bradford, Bradford, UK. 9Parco Archeologico di Pompei, Naples, Italy. 10Parco Archeologico di Ercolano, Naples, Italy. 11Dipartimento di Biologia Ambientale, Sapienza Università di Roma, Rome, Italy. 12Department of Prehistory and Institute of Environmental Science and Technology (ICTA), Universitat Autònoma de Barcelona, Bellaterra, Spain. 13Faculty of History, University of Cambridge, Cambridge, UK. 14Servizio di Bioarcheologia, Museo delle Civiltà, Rome, Italy. 15Dipartimento dei Beni Culturali, Università di Padova, Padua, Italy. *Corresponding author. Email: silvia.soncin@york.ac.uk (S.S.); oliver.craig@york. ac.uk (O.E.C.) †Present address: Department of Anthropology, Harvard University, 11 Divinity Ave, Fig. 1. View of skeletal remains in one of the vaulted chambers (fornici) during Cambridge, MA 02138, USA. excavation. Photo credit: L. Fattore, Sapienza Università di Roma. Soncin et al., Sci. Adv. 2021; 7 : eabg5791 25 August 2021 1 of 9 SCIENCE ADVANCES | RESEARCH ARTICLE and examine dietary variability within this unique sample of Roman A range of controlled studies and feeding experiments have been society at much higher resolution than has previously been achievable undertaken to understand both the degree of fractionation and the (5–7), particularly by deploying a Bayesian model that incorporates extent of AA routing. It has more recently emerged that the latter is prior knowledge of AA metabolism. We examine the capacity of this likely to be itself dependent on dietary composition (19), further method to provide nutritional information regarding major food reducing the reliability of dietary reconstructions based on bulk classes that were available and compare their consumption across the carbon and nitrogen isotopic values. Moreover, the degree of sample. In doing so, we hoped to create an approach for dietary re- fractionation between food and consumer tissues has also been found construction that would be suitable for much wider application. to be variable in animal feeding experiments and controlled dietary Despite its importance for assessing health and well-being, studies of humans (20). To overcome these sources of uncertainty, quantitative data regarding food supply and diet are rarely available isotope ecologists and archaeological scientists are turning to to historians, leaving only impressionistic accounts of consumption. measurements of the isotopic signatures of individual AAs (18, 21–23), Literature, epigraphy, and other documentary evidence, including which can be more easily traced to specific dietary sources. Such papyri, can be a useful source of information for social and economic CSIA approaches are beginning to reveal additional dietary infor- historians, but they are often anecdotal, difficult to quantify, and far mation that is often obscured in bulk stable isotope datasets, allow- from complete, and even the most detailed accounts of consumption ing population-level dietary patterns to be tracked through time practices usually only refer to a narrow stratum of society (2). Faunal and space at much greater resolution (22). However, here, we focus and botanical remains recovered from archaeological excavations on the utility of CSIA to explore intrapopulation dietary differences. provide detailed evidence of the range of the foods available, and Rather than using AA isotope proxies to distinguish dietary groups, quantitative analysis can reveal major economic changes through we use previous knowledge of the AA metabolic pathways, their time (8, 9), but both are subject to sample and taphonomic biases dietary isotope values, and their dietary concentrations to quantify and only rarely they can be reconciled with specific household individual diets using probabilistic models (24). We aimed to exam- activities [e.g., (10)], let alone individual diets. These gaps in our ine whether the differences between individuals at Herculaneum, knowledge limit our ability to meaningfully compare diets either as shown from bulk SIA (7), could be refined and quantified at through time or by geographical location. In addition, we have only higher precision. limited knowledge of how diets may have varied within an ancient society, for example, by social standing, gender, or between households, villages, or towns or over the course of an individual’s life. Without RESULTS accurate quantification, we are unable to make fruitful comparisons We extracted collagen and measured the 13C and 15N values of among ancient populations or with modern societies, where more AAs by gas chromatography–combustion–isotope ratio mass spec- robust and detailed nutritional data are available. Such comparisons trometry (GC-C-IRMS) from the ribs and one tarsal bone (individual are essential for studying the long-term relationship between diet, F10i22) of 11 adult males and 6 adult females whose remains were health, disease, environmental change, and social inequality and the found within the vaulted chambers (fornici) next to the Herculaneum origin and changing nature of food cultures. beachfront (Materials and Methods and table S1). We considered Following its first application over four decades ago (11), stable three potential food groups (C3 cereals, terrestrial animals, and isotope analysis (SIA) of bone collagen offered a way to circumvent marine fish) as the most likely dietary sources for people living in these problems by providing dietary estimates that can be compared 79 CE Herculaneum, based on archaeological finds from the site across time and space. The approach has penetrated all aspects of (10) and historical records (25). We obtained baseline 13CAA and archaeology and anthropology, offering dietary information regarding 15NAA values from the collagen of terrestrial animals (omnivores specific individuals, from Neanderthals to historical figures (12, 13), and herbivores) and marine fish bones, the majority from first and insight into differential access to foodstuffs within populations century CE contexts at Herculaneum and Pompeii (table S2). As (14). The carbon and nitrogen in adult bone collagen derive from endogenous AAs cannot be reliably extracted from archaeological foods typically consumed over a period of at least 10 years before plant remains, which are often charred, an alternative strategy was death (15), and their respective collagen isotope ratios, expressed as used. Bulk and AA stable isotope values were first measured in 13C and 15N values, are related to those in the foodstuffs consumed modern grains to derive an offset for each AA. AA stable isotope over this period. Atoms in collagen are derived from AAs either in- values of archaeological cereal grains were then predicted by applying corporated directly from dietary proteins (source AAs) or synthe- the offsets to bulk measurements of cereal grains from Herculaneum sized de novo (trophic AAs), the latter using additional carbon from and previously reported values from comparable Roman contexts proteins, carbohydrates, and lipids (16), and nitrogen from trans- (Materials and Methods and table S3) (26, 27). Last, Bayesian mixing amination reactions with the metabolic pool of amino nitrogen (17). models were applied to explore the data considering uncertainties This integrated bulk isotopic signal is immensely powerful at pro- in the isotope measurements and the concentration of AAs and viding long-term dietary records, but the approach relies on knowl- macronutrients in the different potential foodstuffs. edge of the proportion of AAs routed to collagen directly from the In model 1, we considered only nitrogen and carbon isotope val- diet against those synthesized de novo by the body. While source AAs ues of source AAs [leucine (Leu), valine (Val), isoleucine (Ile), and undergo negligible isotopic fractionation, trophic AAs are synthe- phenylalanine (Phe) for 13C and Phe and lysine (Lys) for 15N] that sized by a series of transamination and deamination reactions, leading we were able to reliably measure in ancient bone collagen (fig. S1) to significant isotopic fractionations (18). Understanding the mag- and modern cereals (fig. S2). As these AAs show negligible (<1‰) nitude of these isotopic changes under different dietary scenarios is isotopic fractionation between diet and consumer and are derived a major challenge still outstanding in this field, severely limiting the only from dietary protein (Fig. 2), they offer the most robust approach accuracy of the approach. for estimating the composition of ancient human diets because the Soncin et al., Sci. Adv. 2021; 7 : eabg5791 25 August 2021 2 of 9 SCIENCE ADVANCES | RESEARCH ARTICLE Carbohydrates Lipids Proteins C C C N Carbon pool Val M1 His Nitrogen pool LeuM1 Thr Pro Val Arg IleM1 Met Asx Lys PheM1 LysM1 Leu GlxM2 Tyr 3-P - Tyr glycerate Gly Ser Ile Ala PheM1 Free NH3 TCA Pyruvate cycle Arg AlaM2 GlxM2 Met Ser Pro AsxM2 His Hyp Thr Gly Fig. 2. Rationale for metabolic model parameters. Carbohydrates, lipids, and proteins all contribute to the “metabolic carbon pool”: carbon in alanine, serine, and glycine has a glycolytic origin, which is directly linked to carbohydrate digestion; glutamic acid and aspartic acid are synthesized via transamination through the TCA cycle from all macronutrients (33). Dietary protein is considered to be the only source of nitrogen, with glutamic acid as the source of nitrogen for other trophic AAs (17). “Source” AAs incorporated directly from diet with negligible isotopic fractionation are indicated by dashed circles. Isotope values for AAs labeled M1 and M2 are used in model 1 and model 2, respectively. Ala, alanine; Gly, glycine; Val, valine; Leu, leucine; Ile, isoleucine; Thr, threonine; Ser, serine; Pro, proline; Asx, aspartic acid/asparagine; Glx, glutamic acid/glutamine; Phe, phenylalanine; Lys, lysine; Tyr, tyrosine; His, histidine; Arg, arginine. major assumptions regarding fractionation and routing are negated. isotope datasets that males had greater access to marine fish at Using this approach, we were able to easily discriminate the three Herculaneum (7) and more broadly in Roman Italy (29, 30). Males different food groups at Herculaneum, implying fundamental iso- were more likely to be directly engaged in fishing and maritime topic differences in the AAs of primary producers in their respective activities; they generally occupied more privileged positions in society food sources (i.e., cereals, animal forage, and marine phytoplankton). and were freed from slavery at an earlier age, providing greater The estimates obtained from model 1 (Fig. 3B) represent % compo- access to expensive commodities, such as fresh fish (7, 31). However, nent contribution to total dietary protein (by dry weight). Using this here, we were able to quantify the gender gap more accurately within approach, we achieve far higher dietary resolution compared to the group, with males, on average, obtaining 1.6 times more dietary previous approaches that rely on bulk collagen stable isotope data protein from seafood compared with females (Fig. 4A). Males also alone (Fig. 3A) (5), with individual estimates of each food group obtained a higher proportion of protein from cereals compared typically ±10% at the 68% credible interval. We show that the bulk with their female contemporaries, whereas females obtained a greater isotope data underestimate the marine protein component of diet, proportion of protein from terrestrial animal products or locally leading to an erroneous interpretation of the importance of fish to the grown plant foods. Although these estimates do not reflect the inhabitants of this coastal town. When the AA data are considered, absolute quantities of protein consumed, which also may have varied the marine contribution is shown to be nonnegligible (mean = 26 ± 6%) considerably by gender, such a quantitative approach is likely to be for all individuals, in line with estimations based on 14C marine immensely useful for studying nutritional health in ancient societies, reservoir ages (6) and supported by other assessments of the economy especially when used in conjunction with historical sources. of the Bay of Naples during the first century CE (10). The estimated Next, we estimated the contribution of each source to the total marine protein consumption at Herculaneum is notably higher than diet by dry weight, broadly equivalent to the contribution to total the relative amounts of marine protein supplied to mid- and late-20th calorific value. To do so, we considered the additional contribution century Mediterranean populations (28), which are consistently of carbon from dietary carbohydrates and lipids. We adapted the below 10% (Fig. 3). concept of “metabolic pools” (17) from which carbon and nitrogen The source AAs also show significant sex-based dietary differences are drawn for AA synthesis. This model (model 2) additionally throughout the group for all food sources (Fig. 4 and table S5), with considers trophic AAs: alanine (Ala), glutamine/glutamic acid (Glx), females generally obtaining less of their total dietary protein from and asparagine/aspartic acid (Asx) as sources of carbon. The carbon fish and cereals than males but relatively more from terrestrial in Ala is considered to have a glycolytic origin and therefore to have animal products (i.e., meat, eggs, and dairy). This last category could been obtained from the digestion of carbohydrates via pyruvate also theoretically include protein from a broad range of locally (Fig. 2). Conversely, the carbon in Glx and Asx is derived from produced foods, including pulses, legumes, and nuts, as these food- intermediates of the tricarboxylic acid cycle (TCA) and therefore stuffs are likely to have had similar isotope values of source AAs to considered to have been derived from the pool of carbon from all animal forage. It has previously been demonstrated from bulk macronutrients including protein (Fig. 2). These proxies are confirmed Soncin et al., Sci. Adv. 2021; 7 : eabg5791 25 August 2021 3 of 9 SCIENCE ADVANCES | RESEARCH ARTICLE A B C D Female 100 100 Male Average 75 75 Cereals Cereals 50 50 25 25 0 0 100 100 Calorific contribution (%) Protein contribution (%) Terrestrial animals Terrestrial animals 75 75 50 50 25 25 0 0 100 100 75 75 Marine fish Marine fish 50 50 25 25 0 0 98 3 F1 i6 F1 6 8 F1 i3 8 i7 i10 i7 i23 i9 F1 3 F1 4 F1 7 F1 0 F1 2 3 F1 i6 F1 6 8 F1 i3 8 i7 i10 i7 i23 i9 F1 3 F1 4 F1 7 F1 0 F1 2 3 F1 i6 F1 6 8 F1 i3 8 i7 i10 i7 i23 i9 F1 3 F1 4 F1 7 F1 0 F1 2 3 F1 i6 F1 6 8 F1 i3 8 i7 i10 i7 i23 i9 F1 3 F1 4 F1 7 F1 0 F1 2 3 98 3 F1 1 F1 1 -00 -00 G G F1 1 F1 1 G G -6 0i1 0i1 0i2 2i2 i1 0i1 0i1 0i2 0i2 2i2 0i1 0i1 0i2 2i2 i1 0i1 0i1 0i2 0i2 2i2 0i1 0i1 0i2 2i2 i1 0i1 0i1 0i2 0i2 2i2 0i1 0i1 0i2 2i2 i1 0i1 0i1 0i2 0i2 2i2 -6 F8 2 F7 F8 F9 F8 2 F7 F8 F9 F8 2 F7 F8 F9 F8 2 F7 F8 F9 AV AV AV AV F1 F7 F8 F9 F1 F7 F8 F9 F1 F7 F8 F9 F1 F7 F8 F9 61 61 Fig. 3. Dietary estimates for 17 individuals from Herculaneum under different scenarios. Estimates were obtained using a concentration-dependent Bayesian mix- ing model. (A) Model 0p: SIA, protein routed model. (B) Model 1: CSIA, protein model. (C) Model 0wd: SIA, whole diet model. (D) Model 2: CSIA, whole diet model. Boxes represent a 68% credible interval (corresponding to the 16th and 84th percentiles), while the whiskers represent a 95% credible interval (corresponding to the 2.5th and 97.5th percentiles). The horizontal continuous line represents the estimated median (50th percentile). Orange, females; blue, males; gray, outcomes based on average AA isotopic values of the 17 individuals. Equivalent proportions of protein and calorie supplied to modern Mediterranean populations between 1961 and 1963 (gray circles) and 1998 and 2000 (black circles) are shown, with bars representing 1 SD (28). by the high correlations observed in 13C values between AAs and estimations of % marine carbon in diet based on their marine reservoir dietary macronutrients from controlled feeding experiments (16, 32–34). ages (fig. S4) (6, 7). The % dietary protein contribution estimated The 13C values of dietary protein, carbohydrates, and lipids are esti- from model 2 is also within the error of those from model 1, providing mated from the bulk 13C values of faunal collagen or plant remains further cross-validation (table S4). using previously established macronutrient “offsets” updated after The results of model 2 show that, on average, individuals at more recent studies (Supplementary Materials and Methods) (19, 35–37). Herculaneum obtained the majority of their energy from terrestrial Last, model 2 also considered Glx as an additional source of nitrogen. resources, i.e., cereals (49 ± 10%) and terrestrial animal products The difference in 15N of Glx and Phe has been used to study an (40 ± 10%). However, other high-energy products such as olive oil, organism’s trophic position (18), but alone, they fail to resolve more and potentially wine, are not considered as dietary sources and complex diets, as in this case, when there are multiple sources (fig. S3). therefore missing in the outputs provided in Fig. 3. Olive oil, for exam- As glutamic acid is involved in transamination of other AAs, its nitro- ple, contributes ca. 5% of the calories in contemporary Mediterranean gen is considered to derive from the total pool of nitrogen and there- populations (28). By using the 13C value of modern Mediterranean fore is estimated from the bulk 15N value of each protein source olive oils (38–40) corrected for the Suess effect (41), model 2 permits (Fig. 2) (17). The estimation of nitrogen isotopic fractionation asso- a contribution of 29 ± 17% to total diet, when it is included as an ciated with interchange of nitrogen between glutamic acid/glutamine additional source (Supplementary Materials and Methods and data and the nitrogen pool was obtained from studies of a range of consumers file S1). Although even the lowest estimation would be much higher and their food sources (Supplementary Materials and Methods). than most modern Mediterranean populations, this value is consistent Compared to model 1, model 2 introduces additional sources of with estimations of oil consumption in Rome during the first century uncertainty regarding the degree of trophic AA fractionation, energy CE [ca. 20 liters/year (42)], directly attesting to the importance of macronutrient source values, and the flux of both carbon and nitrogen the olive as one of the triads of the Roman Mediterranean diet, along from dietary pools to collagen AAs. Nevertheless, even by using with cereals and wine (25). conservative estimations of these errors (Supplementary Materials and Methods and data file S1), the output of model 2 shows much greater dietary resolution compared to using bulk data alone (Fig. 3C) DISCUSSION (5), with a nonnegligible contribution of marine foods to total calories By applying the CSIA approach to the Herculaneum sample, here, we for the majority of individuals and a statistical difference between are able to reconstruct the diets of people who lived contemporaneously sexes for all foodstuffs (Fig. 3 and table S5). The estimations of with unprecedented resolution compared to previous studies (5–7). calorific value provided by model 2 also correspond well with previous We show with much greater certainty that adult males and females Soncin et al., Sci. Adv. 2021; 7 : eabg5791 25 August 2021 4 of 9 SCIENCE ADVANCES | RESEARCH ARTICLE Fig. 4. “Raincloud” plots of dietary estimates for 17 individuals from Herculaneum grouped by sex. Estimates were obtained using a concentration-dependent Bayesian mixing model. (A) Model 1: CSIA, protein model. (B) Model 2: CSIA, whole diet model. The rainclouds show the raw outputs of each model alongside the means and SDs and the probability density of the distribution. Nonparametric Wilcoxon test (two-sided) shows statistical differences for all the food sources across sex when applied to both model 1 and model 2 (P < 0.05; table S5). drawn from the sample population had different diets during their to 20th century Mediterranean populations, while cereals were of lower lifetime. This must be attributable to differential access to foodstuffs, overall dietary significance compared to the typical “Mediterranean perhaps related to the different occupations held by men and women, diet,” as defined in the 1960s (28). Whether this pattern is reflected cultural prohibitions, or evidence of the uneven distribution of more broadly in ancient Mediterranean societies or is peculiar to coastal power that restricted certain foods to the latter (25). A clear distinction settlements, such as Herculaneum, remains to be determined. Such by sex, however, is not observed in all cases. The dietary estimates high-resolution data also open up the possibility of “benchmarking” from the male sample were more variable than the female, with ancient diets against modern records, where, for example, the nutritional some males consuming less cereal-based foods than the others consequences for health are better understood [e.g., (43)]. (Fig. 3, B and D), perhaps related to differences in their occupation More broadly, we show that CSIA of collagen AAs combined or social standing, aspects difficult to directly assess given the nature with probabilistic modeling, as presented above, offers a robust of the assemblage. It is significant that such subtle dietary differences approach for dietary reconstruction at unprecedented resolution. are not observable from the lower-resolution reconstructions based This is an important advance that is likely to transform paleodietary on the bulk isotope data alone (Fig. 3, A and B). research, not least by providing data that are of adequate quality to The paleodietary data obtained from CSIA are also of sufficient be of interest to the broader community of nutritional and environ- quality for comparison with records of food supplied to modern mental scientists. For example, quantification of seafood consumption populations. We found that proportionally more marine foods were by past communities could be used to study long-term anthropogenic consumed by the inhabitants of first century Herculaneum compared impacts on marine ecosystems (44) or help assess health inequalities Soncin et al., Sci. Adv. 2021; 7 : eabg5791 25 August 2021 5 of 9 SCIENCE ADVANCES | RESEARCH ARTICLE (45). Dietary accuracy is greatly enhanced by our knowledge of the average values and SD across the runs were as follows: IAEA-600 13CAA and 15NAA values of the main food groups under consideration, (n = 25), 13C raw = −27.69 ± 0.15‰ (13C true = −27.77 ± 0.04‰) and and so, obtaining these data from a broader range of non-osseous 15N raw = +0.99 ± 0.26‰ (15N true = 1 ± 0.2‰); IAEA-N-2 (n = 25), sources, such as legumes, nuts, fungi, and wild plant foods, would 15N raw = +20.32 ± 0.15‰ (15N true = 20.3 ± 0.2‰); and IA-CANE be a fruitful focus for future research. Last, we show that using bulk (n = 24), 13C raw = −11.67 ± 0.11‰ (13C true = −11.64 ± 0.03‰). stable isotope data alone to reconstruct an individual’s diet can lead The maximum uncertainty across all samples (n = 83) was ±0.41‰ to erroneous conclusions regarding the relative quantities of different for 13C and 0.32‰ for 15N. foodstuffs consumed and the extent of dietary variability within ancient populations. Preparation of AAs for GC-C-IRMS Collagen was hydrolyzed (6 M HCl, 200 l, 110°C, 24 hours) after addition of 250 l of an internal norleucine standard (Sigma-Aldrich) MATERIALS AND METHODS of known isotopic composition. The hydrolysates were centrifuged Experimental design (11,000g, 1 min) using Pall Nanosep filters (0.45 m) to remove the The ribs and one tarsal bone (individual F10i22) of 17 adult individuals remaining insoluble material. The hydrolysates were gently dried at were obtained from vaulted chambers (fornici) next to the Herculaneum room temperature under N2, redissolved in 0.1 M HCl (100 l), and beachfront. Nine had previously been subjected to radiocarbon dating, stored at −20°C until required for analysis. Samples were again and all had had a full osteological assessment (table S1). Further evaporated to dryness before derivatization. Amino acids were then samples that previously yielded the highest amounts of collagen (1) derivatized to form N-acetyl-i-propyl (NAIP) esters (48). were preferentially selected for analysis. Collagen was extracted and Briefly, isopropanol and acetyl chloride (1 ml; 4:1 v/v) were added, analyzed by elemental analysis IRMS (EA-IRMS) and prepared for and tubes were sealed and heated at 100°C (1 hour). After 1 hour, GC-C-IRMS following hydrolysis to release AAs. The same procedure sample mixtures were cooled (at −20°C), and the solution was dried was applied to faunal remains from the study area (table S2). Stable under a gentle stream of N2. Dichloromethane (DCM) was added carbon and nitrogen isotope measurements were obtained for at (2 × 0.5 ml) and blown down under a gentle stream of N2 to remove least nine individual AAs for each extract. Procedures were made excess reagents. Next, a mixture of acetic anhydride, triethylamine, for assuring quality control (Supplementary Materials and Methods). and acetone (1 ml; 1:2:5, v/v/v) was added to the tubes and heated at AA stable isotope values were also obtained from modern cereals, 60°C (10 min). The mixture was cooled and evaporated to dryness and these data were used to estimate the values for ancient cereals under a gentle stream of N2. NAIP esters were then dissolved in based on their bulk isotope values (table S3). Several mixing models ethyl acetate (EtAc; 2 ml), and a saturated NaCl solution (1 ml) was were constructed using the knowledge of the bulk and AA stable added to separate polar and/or inorganic components from the isotope values in the source foodstuffs, the concentrations of AAs in organic phase and transferred into a new culture tube. The phase the foodstuffs, and their associated uncertainties (Supplementary separation was repeated with additional EtAc (1 ml). Trace water Materials and Methods). The outputs of the models were used to was removed from the organic phase with molecular sieves (sodium create Figs. 3 and 4 and derive inferences. aluminum silicate, 0.3 nm; Merck KGaA, Darmstadt, Germany). The EtAc containing the NAIP esters was blown down under a Collagen extraction gentle stream of N2, and then DCM (1 ml) was added and dried to Collagen was extracted from bone fragments following the modified remove excess water. Samples were redissolved in known quantities Longin method (46). Briefly, small human and animal bone fragments of EtAc and stored at −20°C until required for analysis by GC-C-IRMS. (ca. 100 to 500 mg) were mechanically cleaned to remove exogenous The same derivatization procedure was used for preparing mixtures residues and demineralized at +4°C in 8 ml of 0.6 M HCl for at least of international reference standards (Indiana, USA and SHOKO 48 hours. A homogenized modern bovine bone sample was included Science, Japan) and standards purchased from Sigma-Aldrich (Sigma-­ with each batch of sample to serve as a control. More fragile fish el- Aldrich Company Ltd., UK). ements were demineralized with a more diluted HCl solution (0.1 M). Once completely demineralized, collagen was gelatinized at 80°C Gas chromatography–combustion–isotope for 48 hours in 0.001 M HCl. Gelatinized collagen was filtered (60 to ratio mass spectrometry 90 m; Ezee filters), ultrafiltered (Amicon Ultra-4 Millipore 30 kDa GC-C-IRMS measurements of the AAs were conducted using a of Ultracel membrane), and then freeze-dried. Delta V Plus IRMS (Thermo Fisher Scientific, Bremen, Germany) linked to a Trace Ultra gas chromatograph (Thermo Fisher Scientific, Elemental analysis isotope ratio mass spectrometry Bremen, Germany) with a GC IsoLink II interface fitted with a Collagen (0.9 to 1.1 mg) was analyzed in duplicate using a Sercon Cu/Ni combustion reactor maintained at 1000°C. Ultrahigh-purity–­ continuous flow 20-22 IRMS interfaced with a Universal Sercon gas solid grade helium with a flow rate of 1.4 ml min−1 was used as the carrier liquid elemental analyzer to determine the carbon and nitrogen isotopic gas, and parallel acquisition of flame ionization data was achieved values. The obtained values were corrected from the isotopic ratio of by diverting a small part of the flow to an integrated flame ioniza- the international standards, Vienna Pee Dee Belemnite (VPDB) for tion detector (Thermo Fisher Scientific). Ethyl acetate was used to carbon and air (AIR) for nitrogen, using the standard  (‰) notation. dilute the samples, and 1 l of each sample and 2 l of each standard Uncertainties on the measurements were calculated by combin- were injected at 240°C with a 3.5-spre-injection dwell time onto a ing the SDs of the sample replicates and those of reference material custom DB-35 fused silica column (60 m × 0.32 mm × 0.50 m; according to Kragten (47). Caffeine (IAEA-600), ammonium sulfate Agilent J&W Scientific Technologies, Folsom, CA, USA). All sam- (IAEA-N-2), and cane sugar (IA-Cane) international standards were ples were injected in triplicate. The oven temperature program used used as reference material in each analytical run. International standard for samples and standards was as follows: 40°C (hold 5 min) and Soncin et al., Sci. Adv. 2021; 7 : eabg5791 25 August 2021 6 of 9 SCIENCE ADVANCES | RESEARCH ARTICLE then increasing by 15°C min−1 up to 120°C, then by 3°C min−1 up to The standard 13C average correction factor values and SD 180°C, then by 1.5°C min−1 up to 210°C, then by 5°C min−1 up to (n = 154) across the runs were as follows: Ala, −40.46 ± 1.22‰; 280°C (hold 8 min). Gly, −39.73 ± 1.02‰; Val, −45.57 ± 1.39‰; Leu, −45.03 ± 2.10‰; A Nafion membrane removed water, and a cryogenic trap was Ile, −46.31 ± 1.81‰; Nle, −43.43 ± 1.63‰; Thr, −48.52 ± 1.25‰; used to remove CO2 from the oxidized and reduced sample when Ser, −46.56 ± 1.19‰; Pro, −42.27 ± 1.41‰; Asp, −37.27 ± operated in nitrogen mode. In carbon mode, eluted products were 1.09‰; Met, −41.82 ± 2.12‰; Glu, −36.73 ± 1.10‰; Hyp, −47.97 ± combusted to CO2 and ionized in a mass spectrometer by electron 1.13‰; Phe, −45.36 ± 1.46‰; Lys, −48.29 ± 2.29‰; Tyr, −48.71 ± 1.23‰. impact. Ion intensities of mass/charge ratio (m/z) 44, 45, and 46 were Correction factors induce a new source of error; therefore, the error monitored to automatically compute the 13C/12C ratio of each peak propagated for each AA was calculated according to the following in the samples. In nitrogen mode, ion intensities of m/z 28, 29, and equation (49) 30 were monitored to automatically compute the 15N/14N ratio of each peak in the samples. Computations were made with Isodat ​​​σ​​  2​ = ​σ​​  2​ ​​​ S​​ ​​(​​ ​ ─ ​n​  C ​​​​ )​​​​  ​ + ​σ​​  ​ ​​​ DS​​​(​​ ​ ─ ​​ )​​  + ​σ​​  ​ ​​​ DC​​​(​​ ​ ─ ​​ )​​​​ ​n​  S​​ 2 2 ​n​  S​​  + ​n​  D​​ 2 ​n​  C​​  + ​n​  D​​ (version 3.0; Thermo Fisher Scientific) and were based on comparisons ​n​  C ​​ ​n​  C ​​ with a repeatedly measured high-purity standard reference gas (CO2 or N2). The results from the analysis are reported in parts per mil (‰) relative to international standards using the  notation. where  is the SD, n is the number of carbon atoms, S represents the nonderivatized standard, DS is the derivatized standard, DC is 15N measurements of AAs the derivatized compound, C is the original compound, and D is the Each reported value is a mean of triplicate 15N measurements. An derivative group. AA international standard mixture of known isotopic composition was run after every three sample injections to monitor instrument Analysis of modern and archaeological cereals performance and drift. The AA standard mixture used for 15N Modern C3 cereals were collected from Italian organic productions. determinations comprised eight international standards (Indiana Three species were selected for the analysis: barley (Hordeum vulgare), and SHOKO Science) and l-norleucine (Sigma-Aldrich). 15N true einkorn wheat or farro (Triticum monococcum), and durum wheat values of l-norleucine were determined in-house by EA-IRMS. (Triticum durum). Grains were homogeneously powdered, washed International standard average raw values and SD (n = 124) across three times with deionized water, and freeze-dried. Around 2 mg the runs were as follows: Ala, 42.22 ± 3.07‰ (true: +43.25 ± 0.07‰); was weighed out in duplicate and analyzed by EA-IRMS to measure Gly, +1.09 ± 2.02‰ (true: +1.76 ± 0.06‰); Val, −4.07 ± 1.73‰ bulk carbon and nitrogen isotopic values following the approach (true: −5.21 ± 0.05‰); Leu, +6.39 ± 1.27‰ (true: +6.22‰); described above. A portion of the original powdered material was Nle, +14.46 ± 1.42‰ (true: +14.31 ± 0.23‰); Asp, +33 ± 1.50‰ prepared for compound-specific analysis following a slightly modi- (true: 35.2‰); Glu, −3.52 ± 1.10‰ (true: −4.52 ± 0.06‰); Hyp, −8.19 ± fied protocol from Styring et al. (50, 51). Lipids were first ex- 1.08‰ (true: −9.17‰); Phe, +1.73 ± 0.69‰ (true: +1.70 ± 0.06‰). tracted from the powdered samples with DCM/methanol (2:1 v/v, Sample 15N raw values were corrected by the calibration curve and 10 ml) by ultrasonication, and the extracts were stored at −20°C the l-norleucine internal standard true value. until required for analysis. Around 40 mg of dry lipid extracted residues was hydrolyzed (6 M HCl, 2 ml, 110°C, 24 hours). A known 13C measurements of AAs quantity of internal standard was added at this stage (norleucine, Each reported sample value is a mean of triplicate 13C measurements. Sigma-Aldrich). The hydrolyzed samples were centrifuged (11,000g, Amino acids in the samples were first corrected for the isotopic differ- 1 min) twice using Nanosep filters to remove the insoluble matter ence between l-norleucine in the standard mixture and l-norleucine left. The hydrolyzed samples were blown to dryness under N2, in the sample. 13C AA measurements were then corrected by specific redissolved in 0.1 M HCl, and stored at −20°C until required correction factors to account for the derivatizing carbon and the for analysis. kinetic isotope effect (49), according to the following equation Four charred cereals (ca. 300 mg) from excavations at Herculaneum (table S3) were sampled for EA-IRMS analysis, including H. vulgare (archive #1703/76981), Triticum sp. (#1703/76981 and #723/76000), (​n​  DC​​ ​​​  13​ ​C​ DC​​  ) − (​n​  C​​ ​​​  13​ ​C​ C​​) and Triticum dicoccum (#1895/77175). The samples were treated ​​​​  13​ ​C​ CORR​​ = ​​​  13​ ​C​ D​​ = ​───────────────       ​n ​  D​​ ​​ with 0.5 M HCl for 20 min at room temperature to remove external carbonates and then rinsed three times with deionized water. The where n is the number of carbon atoms, DC indicates the derivat- samples were then frozen and lyophilized, grounded, and weighed ized compound, C is the original compound, and D is the deriva- into tin capsules for EA-IRMS analysis of both carbon and nitrogen tive group. stable isotopes, as described above. A standard AA mixture was run after every three sample injec- The offset in the 13C and 15N values of each AA and the bulk tions, and the average correction factors from the standard mixture value was calculated for each of the three modern C3 cereal samples were used for the correction of the samples (Sigma-Aldrich, UK). (table S3). All AAs in plants are synthesized de novo by following spe- True  13 C values of standards were measured by EA-IRMS: cific metabolic reactions (52). Therefore, we assumed that the degree Ala, −19.31 ± 0.02‰; Gly, −33.31 ± 0.02‰; Val, −10.89 ± 0.02‰; of fractionation of nitrogen and carbon in C3 cereal AAs can be pre- Leu, −13.78 ± 0.06‰; Ile, −24.89 ± 0.07‰; Nle, −27.59 ± 0.02‰; dicted relative to their total nitrogen and carbon. Our 15NAA-bulk Thr, −10.46 ± 0.01‰; Ser, −12.54 ± 0.09‰; Pro, −12.33 ± 0.02‰; Asp, −27.52 ± values were observed to be similar to those of barley and bread 0.12‰; Met, −29.88 ± 0.14‰; Glu, −28.57 ± 0.09‰; Hyp, −12.52 ± 0.03‰; wheat (only grains) published by Styring et al. (52) and to those of Phe, −11.52 ± 0.05‰; Lys, −13.7 ± 0.11‰; Tyr, −24.85 ± 0.02‰. bread wheat published by Paolini et al. (53). Soncin et al., Sci. Adv. 2021; 7 : eabg5791 25 August 2021 7 of 9 SCIENCE ADVANCES | RESEARCH ARTICLE Next, we predicted AA 13C and 15N values by applying the 10. E. Rowan, Sewers, archaeobotany, and diet at Pompeii and Herculaneum, in The Economy measured 15NAA-bulk offsets to the bulk 13C and 15N values from of Pompeii (Oxford Univ. Press, 2017), pp. 111–134. 11. N. J. van der Merwe, J. C. Vogel, 13C Content of human collagen as a measure four samples of C3 cereals from 79 CE Herculaneum, a barley of prehistoric diet in woodland North America. Nature 276, 815–816 (1978). sample from 79 CE Pompeii (27), and four second century CE cereal 12. M. P. Richards, E. Trinkaus, Out of Africa: Modern human origins special feature: Isotopic samples from the Imperial Roman harbor, Portus Romae (26). The evidence for the diets of European Neanderthals and early modern humans. Proc. Natl. bulk values of the archaeological grains were corrected for charring Acad. Sci. U.S.A. 106, 16034–16039 (2009). after (54). From this, we obtained an average value for each AA with 13. A. L. Lamb, J. E. Evans, R. Buckley, J. Appleby, Multi-isotope analysis demonstrates significant lifestyle changes in King Richard III. J. Archaeol. Sci. 50, 559–565 (2014). an associated uncertainty derived from propagating all errors from 14. K. L. Privat, T. C. O'Connell, M. P. Richards, Stable isotope analysis of human and faunal the measurements made on charred archaeological cereal grains remains from the Anglo-Saxon cemetery at Berinsfield, Oxfordshire: Dietary and social and the errors associated with the 15NAA-bulk offset (table S3). implications. J. Archaeol. Sci. 29, 779–790 (2002). 15. R. E. M. Hedges, J. G. Clement, C. D. L. Thomas, T. C. O'Connell, Collagen turnover Statistical analysis in the adult femoral mid-shaft: Modeled from anthropogenic radiocarbon tracer measurements. Am. J. Phys. Anthropol. 133, 808–816 (2007). Bayesian mixing models were performed using FRUITS version 3.0 16. S. Jim, V. Jones, S. H. Ambrose, R. P. Evershed, Quantifying dietary macronutrient sources beta (available at http://sourceforge.net/projects/fruits/). Markov of carbon for bone collagen biosynthesis using natural abundance stable carbon isotope chains were obtained in FRUITS using the Markov chain Monte analysis. Br. J. Nutr. 95, 1055–1062 (2006). Carlo method with the BUGS software (https://www.mrc-bsu.cam. 17. T. C. O’Connell, “Trophic” and “source” amino acids in trophic estimation: A likely metabolic explanation. Oecologia 184, 317–326 (2017). ac.uk/software/bugs/). The BUGS software applies the Metropolis-­ 18. N. Ohkouchi, Y. Chikaraishi, H. G. Close, B. Fry, T. Larsen, D. J. Madigan, M. D. McCarthy, Hastings algorithm and automatically discards the first 5000 iterations K. W. McMahon, T. Nagata, Y. I. Naito, N. O. Ogawa, B. N. Popp, S. Steffan, Y. Takano, of the Markov chains and then additionally runs them for 10000 I. Tayasu, A. S. J. Wyatt, Y. T. Yamaguchi, Y. Yokoyama, Advances in the application iterations. Convergence was assessed by examining the trace of amino acid nitrogen isotopic analysis in ecological and biogeochemical studies. autocorrelation plots generated. Last, the model outputs (Markov Org. Geochem. 113, 150–174 (2017). 19. E. C. Webb, J. Lewis, A. Shain, E. Kastrisianaki-Guyton, N. V. Honch, A. Stewart, B. Miller, chains) were summarized, plotted, and statistically analyzed in R J. Tarlton, R. P. Evershed, The influence of varying proportions of terrestrial and marine (version 4.0.3) using ggplot2 and the raincloud plot function dietary protein on the stable carbon-isotope compositions of pig tissues (https://github.com/RainCloudPlots/RainCloudPlots) (55). Param- from a controlled feeding experiment. STAR: Sci. Technol. Archaeol. Res. 3, 28–44 (2017). eters and the rationale for the four models deployed (model 0p, 20. T. C. O'Connell, C. J. Kneale, N. Tasevska, G. G. C. Kuhnle, The diet-body offset in human model 0wd, model 1, and model 2) are described in Supplementary nitrogen isotopic values: A controlled dietary study. Am. J. Phys. Anthropol. 149, 426–434 (2012). Materials and Methods, and the FRUITS files used to generate the 21. K. Jaouen, M. P. Richards, A. Le Cabec, F. Welker, W. Rendu, J.-J. Hublin, M. Soressi, outputs are also provided. A nonparametric two-sided Wilcoxon test S. Talamo, Exceptionally high 15N values in collagen single amino acids confirm Neandertals was used to test whether distribution of median predicted contribu- as high-trophic level carnivores. Proc. Natl. Acad. Sci. U.S.A. 116, 4928–4933 (2019). tions differed between sexes for each food group at the 0.05 signifi- 22. Y. Ma, V. Grimes, G. Van Biesen, L. Shi, K. Chen, M. A. Mannino, B. T. Fuller, Aminoisoscapes cance level. This test was used because of the low sample of and palaeodiet reconstruction: New perspectives on millet-based diets in China using amino acid 13C values. J. Archaeol. Sci. 125, 105289 (2021). independent observations (17 individuals). 23. A. S. Commendador, B. P. Finney, B. T. Fuller, M. Tromp, J. V. Dudgeon, Multiproxy isotopic analyses of human skeletal material from Rapa Nui: Evaluating the evidence from carbonates, bulk collagen, and amino acids. Am. J. Phys. Anthropol. 169, 714–729 (2019). SUPPLEMENTARY MATERIALS Supplementary material for this article is available at http://advances.sciencemag.org/cgi/ 24. R. Fernandes, A. R. Millard, M. Brabec, M.-J. Nadeau, P. Grootes, Food reconstruction using content/full/7/35/eabg5791/DC1 isotopic transferred signals (FRUITS): A Bayesian model for diet reconstruction. PLOS ONE 9, e87436 (2014). View/request a protocol for this paper from Bio-protocol. 25. P. Garnsey, Food and Society in Classical Antiquity (Cambridge Univ. Press, 1999), 175 pp. 26. T. C. O’Connell, R. M. Ballantyne, S. Hamilton-Dyer, E. Margaritis, S. Oxford, W. Pantano, REFERENCES AND NOTES M. Millett, S. J. Keay, Living and dying at the Portus Romae. Antiquity 93, 719–734 (2019). 1. R. Martyn, O. E. Craig, S. T. D. Ellingham, M. Islam, L. Fattore, A. Sperduti, L. Bondioli, 27. F. D. Pate, R. J. Henneberg, M. Henneberg, Stable carbon and nitrogen isotope evidence for T. Thompson, A re-evaluation of manner of death at Roman Herculaneum following dietary variability at ancient Pompeii, Italy. Mediter. Archaeol. Archaeom. 16, 127–133 (2016). the AD 79 eruption of Vesuvius. Antiquity 94, 76–91 (2020). 28. R. Balanza, P. García-Lorda, C. Pérez-Rodrigo, J. Aranceta, M. B. Bonet, J. Salas-Salvadó, 2. A. Sperduti, L. Bondioli, O. E. Craig, T. Prowse, P. Garnsey, Bones, teeth, and history, in Trends in food availability determined by the Food and Agriculture Organization’s food The Science of Roman History: Biology, Climate, and the Future of the Past, W. Scheidel, Ed. balance sheets in Mediterranean Europe in comparison with other European areas. (Princeton Univ. Press, 2018), pp. 123–173. Public Health Nutr. 10, 168–176 (2007). 3. L. de Ligt, P. Garnsey, The Album of Herculaneum and a model of the town’s demography. 29. O. E. Craig, M. Biazzo, T. C. O'Connell, P. Garnsey, C. Martinez-Labarga, R. Lelli, L. Salvadei, J. Rom. Archaeol. 25, 69–94 (2012). G. Tartaglia, A. Nava, L. Reno, A. Fiammenghi, O. Rickards, L. Bondioli, Stable isotopic 4. L. de Ligt, P. Garnsey, The Album of Herculaneum revisited, in Popolazione, Risorse e evidence for diet at the Imperial Roman coastal site of Velia (1st and 2nd centuries AD) Urbanizzazione nella Campania Antica, M. M. Balbo, Ed. (Pragmateiai, Edipuglia, 2019), in Southern Italy. Am. J. Phys. Anthropol. 139, 572–583 (2009). vol. 31, pp. 197–209. 30. T. L. Prowse, H. P. Schwarcz, S. R. Saunders, R. Macchiarelli, L. Bondioli, Isotopic evidence 5. R. Fernandes, A simple(R) model to predict the source of dietary carbon in individual for age-related variation in diet from Isola Sacra, Italy. Am. J. Phys. Anthropol. 128, 2–13 (2005). consumers. Archaeometry 58, 500–512 (2016). 31. F. Crowe, A. Sperduti, T. C. O'Connell, O. E. Craig, K. Kirsanow, P. Germoni, R. Macchiarelli, 6. O. E. Craig, L. Bondioli, L. Fattore, T. Higham, R. Hedges, Evaluating marine diets through P. Garnsey, L. Bondioli, Water-related occupations and diet in two Roman coastal radiocarbon dating and stable isotope analysis of victims of the AD79 eruption communities (Italy, first to third century AD): Correlation between stable carbon of Vesuvius. Am. J. Phys. Anthropol. 152, 345–352 (2013). and nitrogen isotope values and auricular exostosis prevalence. Am. J. Phys. Anthropol. 7. R. E. V. Martyn, P. Garnsey, L. Fattore, P. Petrone, A. Sperduti, L. Bondioli, O. E. Craig, 142, 355–366 (2010). Capturing Roman dietary variability in the catastrophic death assemblage at 32. V. Jones, “Investigating the routing and synthesis of amino acids between diet and bone Herculaneum. J. Archaeol. Sci. Rep. 19, 1023–1029 (2018). collagen via feeding experiments and applications to palaeodietary reconstruction,” 8. A. Trentacoste, A. Nieto-Espinet, S. Guimarães, B. Wilkens, G. Petrucci, S. Valenzuela-Lamas, thesis, University of Bristol (2002). New trajectories or accelerating change? Zooarchaeological evidence for Roman transformation 33. M. R. Howland, L. T. Corr, S. M. M. Young, V. Jones, S. Jim, N. J. Van Der Merwe, of animal husbandry in Northern Italy. Archaeol. Anthropol. Sci. 13, 25 (2021). A. D. Mitchell, R. P. Evershed, Expression of the dietary isotope signal in the compound- 9. A. King, Diet in the Roman world: A regional inter-site comparison of the mammal bones. specific 13C values of pig bone lipids and amino acids. Int. J. Osteoarchaeol. 13, 54–65 J. Rom. Archaeol. 12, 168–202 (1999). (2003). Soncin et al., Sci. Adv. 2021; 7 : eabg5791 25 August 2021 8 of 9 SCIENCE ADVANCES | RESEARCH ARTICLE 34. S. D. Newsome, M. L. Fogel, L. Kelly, C. M. del Rio, Contributions of direct incorporation 56. E. Gasteiger, C. Hoogland, A. Gattiker, S. e. Duvaud, M. R. Wilkins, R. D. Appel, A. Bairoch, from diet and microbial amino acids to protein synthesis in Nile tilapia. Funct. Ecol. 25, Protein identification and analysis tools on the ExPASy server, in The Proteomics Protocols 1051–1062 (2011). Handbook, J. M. Walker, Ed. (Humana Press, 2005), pp. 571–607. 35. R. Fernandes, P. Grootes, M.-J. Nadeau, O. Nehlich, Quantitative diet reconstruction 57. J. M. Bland, D. G. Altman, Applying the right statistics: Analyses of measurement studies. of a Neolithic population using a Bayesian mixing model (FRUITS): The case study Ultrasound Obstet. Gynecol. 22, 85–93 (2003). of Ostorf (Germany). Am. J. Phys. Anthropol. 158, 325–340 (2015). 58. T. C. O'Connell, M. J. Collins, Comment on “Ecological niche of Neanderthals from Spy 36. E. C. Webb, A. Stewart, B. Miller, J. Tarlton, R. P. Evershed, Age effects and the influence Cave revealed by nitrogen isotopes of individual amino acids in collagen” [J. Hum. Evol. of varying proportions of terrestrial and marine dietary protein on the stable nitrogen- 93 (2016) 82-90]. J. Hum. Evol. 53–55 (2018). isotope compositions of pig bone collagen and soft tissues from a controlled feeding 59. P. Edgar Hare, M. L. Fogel, T. W. Stafford, A. D. Mitchell, T. C. Hoering, The isotopic experiment. STAR: Sci. Technol. Archaeol. Res. 2, 54–66 (2016). composition of carbon and nitrogen in individual amino acids isolated from modern 37. J. M. Bownes, P. L. Ascough, G. T. Cook, I. Murray, C. Bonsall, Using stable isotopes and fossil proteins. J. Archaeol. Sci. 18, 277–292 (1991). and a bayesian mixing model (FRUITS) to investigate diet at the early neolithic site 60. K. W. McMahon, M. L. Fogel, T. S. Elsdon, S. R. Thorrold, Carbon isotope fractionation of Carding Mill Bay, Scotland. Radiocarbon 59, 1275–1294 (2017). of amino acids in fish muscle reflects biosynthesis and isotopic routing from dietary 38. S. N. Dudd, “Molecular and isotopic characterisation of animal fats in archaeological protein. J. Anim. Ecol. 79, 1132–1141 (2010). pottery,” thesis, University of Bristol (1999). 61. K. W. McMahon, M. J. Polito, S. Abel, M. D. McCarthy, S. R. Thorrold, Carbon and nitrogen 39. V. J. Steele, B. Stern, A. W. Stott, Olive oil or lard?: Distinguishing plant oils from animal isotope fractionation of amino acids in an avian marine predator, the gentoo penguin fats in the archeological record of the eastern Mediterranean using gas chromatography/ (Pygoscelis papua). Ecol. Evol. 5, 1278–1290 (2015). combustion/isotope ratio mass spectrometry. Rapid Commun. Mass Spectrom. 24, 62. K. W. McMahon, M. D. McCarthy, Embracing variability in amino acid 15N fractionation: 3478–3484 (2010). Mechanisms, implications, and applications for trophic ecology. Ecosphere 7, e01511 40. J. E. Spangenberg, N. Ogrinc, Authentication of vegetable oils by bulk and molecular (2016). carbon isotope analyses with emphasis on olive oil and pumpkin seed oil. J. Agric. Food Chem. 63. I. P. Kendall, M. R. F. Lee, R. P. Evershed, The effect of trophic level on individual amino 49, 1534–1540 (2001). acid 15N values in a terrestrial ruminant food web. STAR: Sci. Technol. Archaeol. Res. 3, 41. H. Hellevang, P. Aagaard, Constraints on natural global atmospheric CO2 fluxes from 1860 135–145 (2017). to 2010 using a simplified explicit forward model. Sci. Rep. 5, 17352 (2015). 64. B. T. Fuller, K. J. Petzke, The dietary protein paradox and threonine 15N-depletion: 42. E. C. De Sena, An assessment of wine and oil production in Rome’s hinterland: Ceramic, Pyridoxal-5′-phosphate enzyme activity as a mechanism for the 15N trophic level effect. literary, art historical and modern evidence. Arheologija I Prirodne Nauke 6, 25–46 (2010). Rapid Commun. Mass Spectrom. 31, 705–718 (2017). 43. D. Kromhout, E. B. Bosschieter, C. de Lezenne Coulander, The inverse relation between 65. L. R. Germain, P. L. Koch, J. Harvey, M. D. McCarthy, Nitrogen isotope fractionation fish consumption and 20-year mortality from coronary heart disease. N. Engl. J. Med. 312, in amino acids from harbor seals: Implications for compound-specific trophic position 1205–1209 (1985). calculations. Mar. Ecol. Prog. Ser. 482, 265–277 (2013). 44. D. Pauly, D. Zeller, Catch reconstructions reveal that global marine fisheries catches are 66. D. J. Morrison, B. Dodson, C. Slater, T. Preston, 13C natural abundance in the British higher than reported and declining. Nat. Commun. 7, 10244 (2016). diet: Implications for 13C breath tests. Rapid Commun. Mass Spectrom. 14, 1321–1324 45. E. J. Brunner, P. J. S. Jones, S. Friel, M. Bartley, Fish, human health and marine ecosystem (2000). health: Policies in collision. Int. J. Epidemiol. 38, 93–100 (2009). 67. L. L. Tieszen, Natural variations in the carbon isotope values of plants: Implications 46. T. A. Brown, D. E. Nelson, J. S. Vogel, J. R. Southon, Improved collagen extraction by for archaeology, ecology, and paleoecology. J. Archaeol. Sci. 18, 227–248 (1991). modified longin method. Radiocarbon 30, 171–177 (1988). 47. J. Kragten, Tutorial review. Calculating standard deviations and confidence intervals with a universally applicable spreadsheet technique. Analyst 119, 2161–2165 (1994). Acknowledgments: We thank A. Trentacoste for her feedback on the species identification of 48. M. Philben, S. A. Billings, K. A. Edwards, F. A. Podrebarac, G. van Biesen, S. E. Ziegler, terrestrial animal samples from Porta Stabia. Funding: This work was supported by AHRC Amino acid 15N indicates lack of N isotope fractionation during soil organic nitrogen White Rose College of the Arts & Humanities scholarship AH/L503848/1 (to S.S.), European decomposition. Biogeochemistry 138, 69–83 (2018). Research Advance Grant Advanced Grant SICTRANSIT (ERC-ADG-2015 no. 693600) (to O.E.C. 49. G. Docherty, V. Jones, R. P. Evershed, Practical and theoretical considerations in the gas and M.A.), Anglo-American Project (AAPP) in Pompeii (to G.T.), and Pompeii Archaeological chromatography/combustion/isotope ratio mass spectrometry 13C analysis of small Research Project: Porta Stabia (PARP:PS) (to S.E.). Author contributions: Conceptualization: polyfunctional compounds. Rapid Commun. Mass Spectrom. 15, 730–738 (2001). S.S., O.E.C., and L.B.; methodology (isotopes): S.S., H.M.T., A.H., and M.v.T.; methodology (animal 50. A. K. Styring, A. Kuhl, T. D. J. Knowles, R. A. Fraser, A. Bogaard, R. P. Evershed, Practical bone ID): H.K.R. and J.K.B.; sample access and contextualization: K.K.R., M.A., S.E., G.T., V.A., considerations in the determination of compound-specific amino acid 15N values M.O., L.F., M.C., and F.S.; data analysis: S.S. and O.E.C. with support from R.F.; supervision: O.E.C., in animal and plant tissues by gas chromatography-combustion-isotope ratio mass A.C.C., and M.A.; writing (original draft): O.E.C. and S.S.; writing (review and editing): S.S., O.E.C., spectrometry, following derivatisation to their N-acetylisopropyl esters. Rapid Commun. A.C.C., P.G., and L.B. Competing interests: The authors declare that they have no competing Mass Spectrom. 26, 2328–2334 (2012). interests. Data and materials availability: All data needed to evaluate the conclusions in the 51. A. K. Styring, R. A. Fraser, A. Bogaard, R. P. Evershed, The effect of manuring on cereal paper are present in the paper and/or the Supplementary Materials. and pulse amino acid 15N values. Phytochemistry 102, 40–45 (2014). 52. A. K. Styring, R. A. Fraser, A. Bogaard, R. P. Evershed, Cereal grain, rachis and pulse seed Submitted 15 January 2021 amino acid 15N values as indicators of plant nitrogen metabolism. Phytochemistry 97, Accepted 1 July 2021 20–29 (2014). Published 25 August 2021 53. M. Paolini, L. Ziller, K. H. Laursen, S. Husted, F. Camin, Compound-specific 15N and 13C 10.1126/sciadv.abg5791 analyses of amino acids for potential discrimination between organically and conventionally grown wheat. J. Agric. Food Chem. 63, 5841–5850 (2015). Citation: S. Soncin, H. M. Talbot, R. Fernandes, A. Harris, M. von Tersch, H. K. Robson, J. K. Bakker, 54. E. K. Nitsch, M. Charles, A. Bogaard, Calculating a statistically robust 13C and 15N offset K. K. Richter, M. Alexander, S. Ellis, G. Thompson, V. Amoretti, M. Osanna, M. Caso, F. Sirano, L. Fattore, for charred cereal and pulse seeds. STAR: Sci. Technol. Archaeol. Res. 1, 1–8 (2015). A. C. Colonese, P. Garnsey, L. Bondioli, O. E. Craig, High-resolution dietary reconstruction of victims 55. M. Allen, D. Poggiali, K. Whitaker, T. R. Marshall, R. A. Kievit, Raincloud plots: A multi-platform of the 79 CE Vesuvius eruption at Herculaneum by compound-specific isotope analysis. tool for robust data visualization. Wellcome Open Res 4, 63 (2019). Sci. Adv. 7, eabg5791 (2021). Soncin et al., Sci. Adv. 2021; 7 : eabg5791 25 August 2021 9 of 9