Main

Chronology lies at the heart of archaeology16. Radiocarbon dating by accelerator mass spectrometry (AMS) is the most widely used method for providing calendrical chronologies for human activities over the past 50,000 years17, and is most commonly performed on samples of charred plant remains and bone17. Radiocarbon dates can be used alongside relative sequences, such as those derived from stratigraphy or the typological analysis or seriation of artefact types, to build chronological models. Applying Bayes’ theorem enables radiocarbon dating to provide calendar age estimates with uncertainties as low as a few decades18.

The invention of pottery in the late Pleistocene epoch was probably a critical driver for developments in food processing19,20. Pottery vessels can often be placed in robust relative chronological sequences using typology and seriation, although obtaining precise and accurate radiocarbon dates from pottery is challenging2,3,21. All sources of carbon associated with pottery vessels have been considered for dating2,3,4, including organic temper, which occasionally survives firing, and surficial food crusts, although these are rare and prone to contamination owing to their exposed nature22. By contrast, the lipidic components of food residues absorbed into—and protected by—the clay matrix during cooking occur very commonly8, often in high concentrations (milligrams per gram of clay fabric). These offer an untapped resource for radiocarbon dating. The most common absorbed residues correspond to degraded animal fats characterized by their high abundances of C16:0 and C18:0 fatty acids7,8. The possibility of using preparative capillary gas chromatography (pcGC) to isolate chemically pure fatty acids from such residues for compound-specific radiocarbon analysis (CSRA) was recognized more than 20 years ago21,23,24. Although initial attempts to date pottery vessels were promising, the accuracy and precision demanded by archaeology could not be achieved owing to unidentified technical difficulties, leading to highly variable results21,23.

We have brought together the latest technologies for radiocarbon measurements, including automated graphitization and MICADAS compact AMS, in conjunction with high-field 700-MHz NMR, to undertake systematic investigations of the pcGC protocol5,6. Rigorous assessment of contamination in compounds purified by pcGC was undertaken, leading to our invention of a solventless pcGC trap and implementation of cleaning procedures to avoid between-run carryover5,6. These advances reduce the exogenous contamination of fatty acids that has previously been associated with pcGC to below concentrations that would significantly affect measured radiocarbon ages. For archaeological animal fats, it has previously been demonstrated that two fatty acids isolated from the same matrix generate the same radiocarbon age (that is, statistically consistent at the 95% significance level), providing an internal quality control for archaeological dating6. In this study, we aim to extend this method to archaeological pot lipids. We selected pottery vessels that were rich in animal fats from our database of lipid residues that we accumulated over the last three decades. Pottery vessels from chronologically well-characterized settings and different burial environments were analysed and the compatibility of pot lipid dates with these existing chronologies was evaluated by statistical comparison of posterior density estimates for the key parameters and the use of indices of agreement with inclusion in these known frameworks (Fig. 1, Extended Data Table 1 and Supplementary Information 1).

Fig. 1: Site location map, partial gas chromatograms and stable isotope determination of compound-specific radiocarbon-dated lipid residues preserved in Neolithic pottery vessels.
figure 1

a, Map of the location of the archaeological sites for which CSRA was used in this study. Scale bar, 1,000 km. CUI, Cuiry-lès-Chaudardes; ENS, Ensisheim; GEL, Geleen–Janskamperveld; KAR, Karwowo 1; KON, Königshoven 14; LDW, Ludwinowo 7; PPL, Principal Place, London; ROS, Rosheim; SW, Sweet Track; TAK, Takarkori; TP, Çatalhöyük East. b, Partial gas chromatograms of a selection of potsherds showing C16:0 and C18:0 fatty acid abundances. c, Scatter plots of Δ13C ( = δ13C18:0 − δ13C16:0) values plotted against δ13C16:0 values (mean of 2 measurements) for all of the sherds dated (n = 31), ranges on the left denote the mean ± 1 s.d. of modern reference fats, as reported in ref. 28.

Full size image

We initially focused on Neolithic Carinated Bowl pottery from the Sweet Track (Fig. 2a), an elevated wooden trackway discovered in a wetland area of the Somerset Levels9,10,25 in the United Kingdom (Supplementary Information 2). This site is critical because its construction has been precisely dated by dendrochronology to the winter–early spring of 3807–3806 bc and the trackway was used and maintained for approximately 10 years10. Lipids from pots that were found alongside the trackway, and were probably contemporaneous to its construction and use, have previously been dated, but the measured dates were a century later than the construction of the trackway23. Re-analysis of the two vessels (Fig. 2b) using our new approach produced uncalibrated radiocarbon ages of 5,110 ± 25 years before present (bp; taken as ad 1950) (SW1) and 5,092 ± 26 bp (SW2), which are statistically indistinguishable (T′ = 9.0, T′(5%) = 9.5, ν = 4) from the measurements of the tree rings included in the IntCal13 calibration curve for the relevant decade26 (Fig. 2c). The calibrated dates of these ages are clearly compatible with the tree-ring dates for the construction of the trackway.

Fig. 2: Sweet Track timbers, a pottery vessel and calibrated radiocarbon dates.
figure 2

a, Photograph of Sweet Track timbers. b, Photograph of a Carinated Bowl (SW2) that was recovered alongside the Sweet Track. Scale bar, 5 cm. c, Probability distributions of dates from pots deposited next to the Sweet Track (green) and from oak trees (black) included in IntCal1326 that include the date of the Sweet Track construction in 3807–3806 bc. Each distribution represents the relative probability that an event occurs at a particular time. For each of the dates, two distributions have been plotted: one in outline, which is the simple radiocarbon calibration, and a solid distribution, based on the model used. The square bracket down the left side along with the OxCal keywords define the overall model exactly (provided in Supplementary Information 2). A, Acomb and An are the individual agreement indices, the combination agreement indices and the acceptable threshold to combine n radiocarbon dates, respectively. The photographs were provided by S.M. and are reproduced with permission from the Somerset Levels Project.

Full size image

Extending our approach to Anatolia, the Neolithic tell of Çatalhöyük East was a locus for the emergence and development of pottery production. A 21-m-deep stratigraphic sequence provides strong archaeological prior information for a Bayesian chronological model that covers the upper parts of the mound (TP area)11. The sequence of houses, middens and burial structures has been combined with 50 radiocarbon dates, revealing a Neolithic sequence of occupation from the mid-sixty-fourth to the mid-sixtieth centuries calibrated (cal.) bc11. Our compound-specific radiocarbon ages on adipose lipids27 from four pottery vessels from four different contexts (TP.M17, 7,382 ± 31 bp; TP.N10, 7,348 ± 25 bp; TP.O23, 7,340 ± 27 bp; and TP.P13, 7,364 ± 25 bp) were incorporated into the Bayesian chronological model for this part of the site (Extended Data Figs. 1, 2 and Supplementary Information 3). The revised model for the Neolithic deposits in the TP area shows posterior distributions for the key parameters that are almost identical to those from the original model11. Their median values vary by an average of 4 years and a maximum of 10 years, confirming the compatibility of the radiocarbon ages determined using fatty acids with the site stratigraphy. On the basis of sensitivity analyses (Supplementary Information 3), this well-constrained model is at least as sensitive as measurements on paired materials to detect inaccuracies. In this case, the CSRA dates not only provide direct dating for the importance of ruminant carcass products (Fig. 1c) to the inhabitants of Çatalhöyük at this time (derived from δ13C values of preserved fats), but also provide direct dating evidence for the climatic changes associated with the global event of 8.2 thousand years ago (derived from compound-specific deuterium isotope analyses using the same fats)27.

The next analysis tests the accuracy of our dating approach using a classic pottery seriation study related to Neolithic ceramics from Lower Alsace (France) that spans the second quarter of the fifth millennium cal. bc12 (Supplementary Information 4). The regional correspondence analysis clearly separates the Hinkelstein, Grossgartach, Planig-Friedberg and Rössen Middle Neolithic ceramic groups. We focused on vessels from three pits, all of which can be assigned to the Grossgartach phase (Fig. 3a, b). The sequence of ceramic phases was combined with the existing assemblage of 95 radiocarbon dates, which were largely measured on articulated bones, along with four CSRA dates on fatty acids (ROS-C-4596, 5,804 ± 25 bp; ROS-C-4600, 5,904 ± 28 bp; ROS-C-4644, 5,931 ± 26 bp; and ROS-C-4657, 5,912 ± 28 bp) from the Grossgartach sherds in a model using Bayesian statistics. The phase boundaries in this revised model are very similar to those produced by the original analysis12, as median values differ by an average of 6 years and a maximum of 15 years (Fig. 3c). The sensitivity analyses (Supplementary Information 4) demonstrate that the model is particularly sensitive to small biases, and probably more sensitive than measurements on paired materials. The CSRA dates are clearly compatible with the attribution of these pottery vessels to the Grossgartach ceramic phase based on their decorative motifs, and with the other radiocarbon dates for this group.

Fig. 3: Drawings, correspondence analysis and radiocarbon dates of Neolithic vessels from Alsace (France) modelled using Bayesian statistics.
figure 3

a, Drawings of decorated pottery vessels from the Grossgartach group from pits 122 (1, 2), 50 (4, 5) and 63 (3, 6, 7) from which the undecorated potsherds dated in the model were recovered. Scale bar, 5 cm. b, Revised correspondence analysis on the cultural assemblages (axis 1) and horizontal decorative motifs (axis 2), including features that contained the dated sherds from the Alsatian Neolithic groups. c, Revised statistical model of the Middle Neolithic with radiocarbon dates on pot lipids included in grey. The data are shown as described in Fig. 2c. Amodel is the model agreement index.

Full size image

We then explored the introduction of a new food product—that is, milk—into Neolithic Europe by undertaking radiocarbon dating of animal fat residues, including dairy fats, that were recovered from early farming settlements with Linearbandkeramik (LBK) pottery (Fig. 1). These communities settled in central Europe from the early fifty-fourth century bc13. Animal fats in 12 potsherds from the earliest LBK contexts at 6 sites, in Poland, France, Germany and the Netherlands, produced radiocarbon dates that were modelled and shown to be compatible with the currency of LBK ceramics in northern and western Europe12,13 (Extended Data Figs. 3, 4 and Supplementary Information 5). Sensitivity analyses (Extended Data Fig. 4 and Supplementary Information 5) demonstrate that this model is more sensitive to older biases as we focused on early settlements, illustrating the direct dating of a new food commodity. The radiocarbon dates on the earliest dairying residues suggest that the practice began in 5385–5225 cal. bc (95% probability; start LBK lipid; Extended Data Fig. 3) and probably arrived with the earliest farmers in these areas. Thus, the linking of fatty acid structures with compound-specific carbon isotope values and CSRA dates provides a powerful means of directly dating prehistoric foodways and their introduction.

We next investigated pottery from the Sahara Desert to provide a test of the methodology for a region in which depositional conditions are very different from the temperate climes of northern Europe. The Takarkori rock shelter, located in the now hyper-arid area of the Acacus Mountains, southwest Libya, demonstrates evidence of animal exploitation based on rock art and archaeological finds14 (Extended Data Figs. 5, 6). Previous work revealed abundant adipose and dairy fat residues in fragments of the pottery vessels28. Stratigraphy and radiocarbon dating of a range of materials (bone collagen, charred plant remains, dung, skin and enamel bioapatite) placed deposits associated with Middle Pastoral pottery in the sixth–fifth millennia cal. bc14,28,29. The fatty acids from 5 potsherds, containing dairy fat (Extended Data Fig. 6b), produced uncalibrated radiocarbon ages of 5,993 ± 28 bp (TAK443), 5,979 ± 28 bp (TAK120), 5,493 ± 28 bp (TAK420), 5,348 ± 24 bp (TAK21) and 5,085 ± 24 bp (TAK1572). The CSRA dates were proven to be entirely compatible with the currency of Middle Pastoral Neolithic ceramics (Extended Data Fig. 6d and Supplementary Information 6), and the direct radiocarbon dating of dairy residues confirms that dairying in North Africa began as early as the end of the sixth millennium cal. bc14,28,29. Although the model sensitivity is weak based on the small number of reference dates that it includes (Extended Data Fig. 7 and Supplementary Information 6), it demonstrates the possibility of dating potsherds from extremely arid burial conditions. In addition, direct dating of pottery lipids represents a major contribution to ascertain the correct cultural attribution of materials found in loose sediments (organic sands), which are typical of desert environments and frequently found in highly disturbed sequences14.

Finally, archaeological excavations of several pits by the Museum of London Archaeology in advance of building works at Principal Place, London (PPL11) revealed one of the largest assemblages of Neolithic pottery recovered so far from the City of London or its immediate environs. Notably, the only finds other than pottery recovered from this deposit (lithics, bones and charred plant remains) were compromised by later disturbance and truncation. The assemblage comprised Neolithic plain and decorated bowls, consisting of thin-walled medium-sized open/neutral bowls, together with several smaller open bowls/cups (J.C. et al., manuscript in preparation). Similar material has been found elsewhere in the Thames Valley and beyond. Lipid-residue analyses revealed high concentrations of degraded animal fats in several sherds, which were shown by compound-specific δ13C values to derive from dairy and adipose fats (Fig. 1c). Radiocarbon measurements of fatty acids from four plain sherds yielded uncalibrated ages of 4,911 ± 27 bp (PPL012), 4,742 ± 22 bp (PPL015), 4,652 ± 26 bp (PPL020) and 4,733 ± 22 bp (PPL021). A statistical model confirms that the pottery dates fit well within the currency of Plain Bowls in southern Britain15 (Extended Data Fig. 8 and Supplementary Information 7). The sensitivity analyses (Extended Data Fig. 9 and Supplementary Information 7) are weaker in this case, but demonstrates the value of dating absorbed lipid residues in situations in which no other datable material exists. Our ability to undertake accurate radiocarbon dating of compound-specific fatty acids from pottery was invaluable in affording a temporal insight into some of the earliest traces of human activity in what is now the City of London.

In summary, the radiocarbon determinations of lipid residues from Neolithic pottery vessels presented above, modelled against site chronologies, establish the CSRA of fatty acids as a robust tool for archaeological dating. Importantly, our method and findings bring pottery vessels within the range of other archaeological materials that are routinely dated by radiocarbon. The importance of this advance to the archaeological community cannot be overstated. Pottery typology is the most widely used dating technique in the discipline, and thus the opportunity to anchor different kinds of pottery securely to the calendrical timescale will be of utmost practical importance. Notably, pottery often survives in circumstances in which other organic materials often do not and, therefore, archaeological questions relating to chronology that are currently intractable will come within the scope of our technologies.

Methods

Lipid extraction and isolation

Potsherds were selected on the basis of the presence of terrestrial animal fats (dairy and ruminant carcass fats) in the lipid residue to avoid any possible reservoir effect caused by the processing of aquatic products in pots. A piece of 1–10 g of the potsherd was sampled, according to the lipid concentration. The sherds were extracted in a glass culture tube using H2SO4/MeOH (4% v/v, 3× 8 ml, 70 °C, 1 h). The supernatants were centrifuged (2,500 rpm, 10 min) and combined into new culture tubes containing double-distilled water (5 ml). The lipids were extracted with n-hexane (4× 5 ml), transferred into 3.5-ml vials and blown to dryness at room temperature under a gentle nitrogen stream. Subsequently, around 180 μl of n-hexane was added to obtain a concentration of fatty acid methyl esters (FAMEs) at 5 μg of C μl−1 before transfer to an autosampler vial for isolation by pcGC.

The pcGC consisted of a Hewlett Packard 5890 series II gas chromatograph coupled to a Gerstel Preparative Fraction Collector by a heated transfer line. The pcGC was equipped with a column with a 100% poly(dimethylsiloxane) stationary phase (Rxi-1ms, 30 m × 0.53 mm inner diameter, 1.5 μm film thickness, Restek). Helium was used as the carrier gas at a constant pressure of 10 psi. The GC temperature programme started with an isothermal hold at 50 °C for 2 min, the temperature was increased to 200 °C at 40 °C min−1, to 270 °C at 10 °C min−1 and finally increased to 300 °C at 20 °C min−1 and held for 8.75 min. The C16:0 and C18:0 FAMEs were injected (1 μl per run), separated and trapped 40 times per trapping sequence. Of the GC column effluent, 1% flows to the flame ionization detector, while the remaining 99% passes through a transfer line into the fraction collector, both of which were heated to 300 °C. Compounds were isolated based on their retention times6. The stationary phase degradation of the pcGC column and other sources of exogenous carbonaceous contamination were monitored on a Brucker Avance III HD 700 MHz NMR instrument following a previously published procedure5,6.

Radiocarbon determinations and statistical analysis

The pcGC isolated compounds were transferred into Al capsules, after which they were combusted and graphitized in a Vario Microcube Elemental Analyser linked to an Automated Graphitisation System (AGE 3, IonPlus). All of the radiocarbon measurements were performed by the Bristol Accelerator Mass Spectrometer (BRAMS) facility at the University of Bristol. Data reduction was performed using the software BATS30 (v.4.07). Radiocarbon dates obtained for FAMEs were corrected for the presence of added methyl carbon using a mass balance approach5,6,21 and reported as the conventional radiocarbon ages31 (Supplementary Information 1).

Two contemporaneous compounds (C16:0 and C18:0 fatty acids) were dated and every pair of statistically indistinguishable measurements (at the 95% significance level)32 was combined as a weighted average before Bayesian chronological modelling using OxCal v.4.2 and v.4.318,33 and the currently internationally agreed radiocarbon calibration curve for the Northern Hemisphere, IntCal1326. The compatibility of the radiocarbon dates on absorbed fatty residues with existing sites and regional chronologies was assessed by including the lipid radiocarbon dates into existing statistical frameworks in a position defined by archaeological information (for example, stratigraphy or seriation). Their compatibility with the existing chronologies were achieved by: (1) comparison of posterior density estimates for key modelled parameters with equivalent date estimates or known age by dendrochronology; (2) using the individual and model agreement indices18,33 in models containing fatty acid dates; and (3) comparing posterior density estimates for key parameters from models that include the fatty acid dates to a model that does not include the fatty acid dates (Supplementary Information 1). The sensitivity of existing chronological models to the addition of the new radiocarbon measurements was evaluated as above, after deliberately biasing the radiocarbon dates on pottery vessels to varying degrees while assessing the effect on posterior density estimates for the key parameters and indices of agreements (Supplementary Information 1–7).

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this paper.