Survival of Absorbed and Visible Lipids

Eighty-two percent (n = 9) of the foodcrust samples and 88% (n = 28) of the absorbed residues exhibited satisfactory levels of lipid preservation (potsherds >5 μg.g−1; foodcrusts >100 μg.g−1; Craig et al. Reference Craig, Saul, Lucquin, Nishida, Taché, Clarke and Thompson2013; Evershed Reference Evershed2008). These interpretable samples represent altogether 28 vessel units. The average lipid concentrations are 104 and 356 μg.g−1 for the absorbed and foodcrust samples, respectively. Such levels are comparable to lipid yields reported in other studies of indigenous ceramics in northeastern North America (Taché and Craig Reference Taché and Craig2015) and confirm that organic matter tends to preserve well in these contexts.

Molecular Characterization

Dawson lipid profiles contain an array of saturated fatty acids (C_8:0 to C_30:0) dominated by palmitic (C_16:0) and stearic (C_18:0) acids, monounsaturated fatty acids with even numbers of carbon atoms (C_14:1 to C_24:1), and branched fatty acids ranging from C₁₃ to C₁₉. Several samples also contain a range of dicarboxylic acids (C₆ to C₁₈), potentially representing oxidation products of unsaturated fatty acids, despite their occasional association with other degradation processes occurring during burial (Baeten et al. Reference Baeten, Jervis, Vos and Waelkens2013; Copley et al. Reference Copley, Bland, Rose, Horton and Evershed2005; Regert et al. Reference Regert, Bland, Dudd, van Bergen and Evershed1998). Cholesterol and products of its degradation (primarily Cholest-5-ene, 3-methoxy-, [3.beta.]-) are present in 10 samples, indicating that animal resources were processed in these containers (Whelton et al. Reference Whelton, Hammann, Cramp, Dunne, Roffet-Salque and Evershed2021). In addition, plant biomarkers were recorded in 12 samples. These include methyl dehydroabietate and 7-oxo-dehydroabietate, biomarkers for pine resin (Jerković et al. Reference Jerković, Marijanović, Gugić and Roje2011; Mitkidou et al. Reference Mitkidou, Dimitrakoudi, Urem-Kotsou, Papadopoulou, Kotsakis, Stratis and Stephanidou-Stephanatou2008; Modugno and Ribechini Reference Modugno, Ribechini, Colombini and Modugno2009; Regert Reference Regert2004) in nine samples. These compounds may originate from the use of coniferous resin as a sealant (Reber and Hart Reference Reber, Hart and Hart2008) or from the smoke of campfires if pine was being used as fuel (Reber et al. Reference Reber, Kerr, Whelton and Evershed2019). Stigmastanol is present in four samples, and α-amyrin (pentacyclic triterpenoid, common among angiosperm resins; Bondetti et al. Reference Bondetti, Chirkova, Lucquin, Meadows, Lozovskaya, Dolbunova, Jordan and Craig2019) was identified in one sample (Supplemental Table 3). Significantly, n-dotriacontanol, previously established as a potential maize biomarker (Reber et al. Reference Reber, Dudd, van der Merwe and Evershed2004), was not detected in the Dawson samples following TMS derivatization of the acid methanol extracts.

Instead, the most distinctive molecular feature of the Dawson pottery sherds is the presence of biomarkers for aquatic products—namely, long-chain C₁₈-C₂₂ (o-alkylphenyl)alkanoic acids (APAAs) with C₂₀/C₁₈ ratios above 0.06, along with isoprenoid fatty acids (4,8,12-Trimethyltridecanoic, phytanic, and pristanic acids; Bondetti et al. Reference Bondetti, Scott, Courel, Lucquin, Shoda, Lundy, Catalina, Léa and Oliver2021; Craig et al. Reference Craig, Forster, Andersen, Koch, Crombé, Milner, Stern, Bailey and Heron2007; Evershed et al. Reference Evershed, Copley, Dickson and Hansel2008; Hansel et al. Reference Hansel, Copley, Madureira and Evershed2004). These account for 78% (n = 29/37) of the interpretable samples and 71% (n = 20/28) of the interpretable vessel units analyzed (Figure 3). Dihydroxy fatty acids, also used to identify aquatic resources (Cramp and Evershed Reference Cramp, Evershed, Holland and Turekian2014), were not detected in the Dawson residues. The APAAs form during the protracted heating of mono- and polyunsaturated fatty acids, implying that vessels were subjected to at least one hour of heating at 270°C, or at 200°C for five hours, conditions easily achieved through both boiling or roasting (Bondetti et al. Reference Bondetti, Scott, Courel, Lucquin, Shoda, Lundy, Catalina, Léa and Oliver2021; Lucquin et al. Reference Lucquin, Robson, Eley, Shoda, Veltcheva, Gibbs and Heron2018). The conditions required to form the highly diagnostic APAAs are not always met during food preparation, and both APAAs and isoprenoid acids, which occur at low concentrations relative to other lipid molecules, may be lost through exposure to the burial environment. In phytanic acid, a ratio of the diastereomers SRR/RRR (3S,7R,11R,15-phytanic acid/3R,7R,11R,15-phytanic acid) exceeding 75.5% has been shown to be characteristic of aquatic resources (Lucquin et al. Reference Lucquin, Colonese, Farrell and Craig2016, Reference Lucquin, Robson, Eley, Shoda, Veltcheva, Gibbs and Heron2018). In this study, 46% (n = 11/24) of the samples for which this value was obtained meet this criterion. The remaining samples, characterized by SSR/RRR ratios ranging from 41 to 72, may reflect the processing of ruminant resources, an interpretation further supported in cases where no aquatic biomarker is present (samples #1, 2, 7, 11, 12). Samples with SRR/RRR ratios below 75.5% and aquatic biomarkers may, on the other hand, reflect a mixture of ruminant and fish resources. The possibility that some of these containers (samples #21, 27, 33, 37, 41) were used to process freshwater shellfish, also shown to have lower diastereomer ratios (Admiraal et al. Reference Admiraal, Colonese, Milheira, da Rocha Bandeira, Demathe, Pereira dos Santos and Fossile2023), is worth considering despite the absence of bivalve shells in the faunal assemblage from Dawson. Nonetheless, aquatic resources clearly were a favorite ingredient among the Dawson cooks, but were these fish prepared alone or mixed with other foods? Isotopic measurements of fatty acids and bulk charred deposits provide a more quantitative measure of food input.

Figure 3. Typical partial gas chromatogram of a lipid extract from Dawson-site ceramics showing evidence of degraded aquatic oil from potsherd 9S3B (sample #19). The partial m/z 105 ion chromatogram (inset) shows ω-(o-alkylphenyl)alkanoic acids with 16 (?), 18 (+), 20 (*), and 22 (#) carbon atoms. Cn:x are fatty acids with carbon length n and number of unsaturations x; DCx are α,ω-dicarboxylic acids with carbon length x; br are branched-chain acids; TMTD is 4,8,12- trimethyltridecanoic acid; IS is internal standard (n-hexatriacontane).

Isotopic Characterization

Bulk Stable Isotope Analysis. Despite limitations arising from uncertain isotope endpoints of different foodstuffs and diagenetic alteration (Heron and Craig Reference Heron and Craig2015), EA-IRMS of charred deposits adhering to the interior walls of ceramic vessels can provide a number of useful lines of information, especially when combined with other analytical techniques (Hastorf and DeNiro Reference Hastorf and DeNiro1985). Bulk δ¹³C isotope values from 10 carbonized deposits range from −17.3 to −24.4‰, indicating variable C₃ and C₄ inputs that must include maize and that are consistent with a mixture of terrestrial C₃ and C₄ plants, terrestrial mammals, and freshwater fish. A comparison of the Dawson site carbon and nitrogen isotope values with previously published data obtained on the region's earliest pottery (ca. 3000–2400 BP) from a range of inland and coastal sites is informative (Taché and Craig Reference Taché and Craig2015). Figure 4 shows almost no overlap between early pottery and Dawson δ¹³C values. Although both the early and Dawson pots include samples with elevated carbon values, the early pottery samples systematically have higher δ¹⁵N values. In this case, the elevated carbon measurements are likely due to a contribution from marine resources, an interpretation further supported by the coastal location of these samples, the presence of aquatic biomarkers in the residues, and the absence of maize agriculture at this time. The comparatively low δ¹⁵N values associated with samples from Dawson, on the other hand, leave little doubt that maize contributed to these residues—information that was not readily accessible through molecular characterization alone. Comparison between early pottery and Dawson C:N ratios also shows a tendency for higher C:N ratios at Dawson (unpaired t-test, t = 2.79; p = 0.027). Given that Atomic C:N ratios are indicative of the amount of protein versus other macromolecules such as carbohydrates and lipids (Bondetti et al. Reference Bondetti, Chirkova, Lucquin, Meadows, Lozovskaya, Dolbunova, Jordan and Craig2019), these higher values are consistent with a greater contribution from plant tissues at Dawson. The difficulty surrounding the identification of plant resources, maize specifically, through GCMS analysis of Dawson pottery likely results from a mixing of resources in these containers. Indeed, given that each lipid compound identified in a residue may derive from any single resource ever processed in a pot, without biomarkers the contribution of plants to a chromatogram tends to go unnoticed if mixed with other lipid-rich resources (Evershed Reference Evershed2008).

Figure 4. Bulk stable carbon and nitrogen isotope data obtained from internal carbonized residues adhering to Iroquoian pottery from the Dawson site (black circles) and the region's earliest (ca. 3100–2300 cal. BP) pottery from inland (filled gray circles) and coastal sites (open gray circles; data previously reported in Taché and Craig Reference Taché and Craig2015). The median and ranges (2σ) of experimentally charred aquatic and terrestrial animals (Craig et al. Reference Craig, Saul, Lucquin, Nishida, Taché, Clarke and Thompson2013) are also shown.

Single-Compound Isotope Analysis. To further investigate the source of lipids recovered from the Dawson assemblage, the δ¹³C values of the two main fatty acids—palmitic (C_16:0) and stearic (C_18:0)—was determined using GC-combustion-isotope ratio MS (GC-C-IRMS) analysis. Twenty-five absorbed lipid samples representing 24 distinct vessels were analyzed by GC-C-IRMS and compared to reference value ranges of the most likely food candidates to have been processed in these pots: freshwater aquatic oils (indistinguishable from nonruminant adipose and C₃ plants based on isotope values alone), wild ruminant adipose, and maize (Figure 5). An ellipse representing the reference value range for marine resources was also included in Figure 5 to illustrate potential equifinality issues met when interpreting residues representing mixing of resources; however, given the site's location and the extremely sparse findings of marine fauna (see above), fats derived from marine animals would seem highly improbable. The results show a wide range of δ¹³C values for palmitic (C_16:0) and stearic (C_18:0) fatty acids from −21.8 to −30.9‰ (palmitic) and from −21.0 to −31.9‰ (stearic; Supplemental Table 3). Although several samples fall within the reference value ranges of freshwater aquatic oils and wild ruminant adipose, it is notable that almost half of the Dawson data (n = 12) are found outside the reference value ranges of the most likely food candidates to have been processed at Dawson. Although a contribution from anadromous/catadromous fish species (characterized by isotopic values falling between freshwater and marine taxa) cannot be excluded, a more likely explanation for these intermediate values considering other sources of data (faunal, botanical, isotopic, contextual) would be that, rather than a single source, residues from Dawson contain a complex distribution of lipids from C₄ plants (i.e., maize) and freshwater aquatic oils, at the very least. This independent line of evidence provided by GC-C-IRMS values provides further support to the combination of isotopic and molecular characterization of Dawson residues discussed above and requires further deconvolution.

Figure 5. δ¹³C values of C_16:0 and C_18:0 n-alkanoic acids extracted in (a) Dawson site samples with (filled gray circles) and without (open gray circles) aquatic biomarkers. The data are shown against modern reference values expressed as 68% confidence ellipses (Supplemental Table 4); (b) authentic mixes of maize and lake trout in 10% increments (asterisk symbols). The data are shown against modern reference values expressed as 68% confidence ellipses (Supplemental Table 4) and average isotopic endpoints and mixing lines in 10% increments for hypothetical mixes generated in R of maize with (1) freshwater aquatic resources and (2) wild ruminant adipose fats.

Mixing Models

Identifying mixtures of foods, determining whether ingredients were cooked together or through sequential use of the pot over time, and making inferences about the relative proportions of various food types are major challenges met when interpreting lipid profiles, which, to this day, preclude the reconstruction of ancient recipes. To illustrate potential effects of mixing, we considered simple linear mixing models and more complex Bayesian mixing models. We also conducted actualistic experiments, where modern maize from an organic Amish farm in northern Washington County (New York state) and lake trout from Cayuga Lake (New York state) were mixed in 10% increments as described in Hart et alia (Reference Hart, Taché and Lovis2018; Supplemental Text 1). The results of these validate the theoretical maize–freshwater fish mixing line and suggest the mixing of maize and freshwater aquatic oil as one likely explanation for the δ¹³C measurements of palmitic and stearic fatty acids (C_16:0 and C_18:0) observed on, at least some, of the Dawson samples (Figure 5a). As recently proposed by Admiraal et alia (Reference Admiraal, Colonese, Milheira, da Rocha Bandeira, Demathe, Pereira dos Santos and Fossile2023) using theoretical mixing curves, caution must be taken when interpreting Δ¹³C values below −1.1 as evidence for the presence of ruminant adipose or dairy fats when C₄ plant oils are a plausible component. Indeed, Figure 5b shows—for the first time empirically—that such negative Δ¹³C values could also be observed when mixing maize with aquatic oils. Of the Dawson vessels, six samples have Δ¹³C values below −1.1 and therefore could be interpreted as a mixture of freshwater fish and maize, potentially with a ruminant fat—for example, from the artiodactyl (that is, deer and moose), which is plentiful at the site. Interpretations of such negative values are further complicated by the identification of the full set of aquatic biomarkers in four of these samples, whereas the other two contain all three isoprenoid acids and SRRs of 71% and 80%, also consistent with the processing of aquatic resources despite the absence of ω-(o-alkylphenyl)alkanoic acids. This raises the possibility that more than two food sources are represented in the residues.

The use of Bayesian mixing models was advocated to better understand the proportional contribution of more than two food sources to lipid residues (Fernandes et al. Reference Fernandes, Millard, Brabec, Nadeau and Grootes2014). These models must rely on a range of reference data, ideally as representative as possible of the environmental and cultural contexts under investigation. Such a comparative baseline was sought for the Dawson site, where the limited reference values from northeastern North America were complemented with reference values obtained in comparable environmental contexts around the world (Supplemental Table 4). To verify the efficiency of our model parameters to estimate the relative contributions of different food sources to a residue, it was first tested against the authentic maize-lake trout mixtures. In the first test, where only two food sources (maize and freshwater fish) were included, the Bayesian model generated by FRUITS reliably estimated the proportions of both ingredients, despite relatively high uncertainty ranges (Figure 6). When a third food source (wild ruminant adipose) is considered as a potential explanation for the values associated with the lake trout–maize experimental mixtures, we observe a trend toward underestimating the proportion of freshwater fish in samples, except in cases where fish represent only a small proportion (less than 20%) of the mixes. It should also be noted that the contribution of wild ruminant adipose, which should be nil, is overestimated when standard deviations are considered, but significantly, the model predicts that a zero or negligible (less than 3%) percent contribution is credible in all cases at the 95% confidence interval. Despite the addition of wild ruminants as a potential food source in Figure 7, it is also reassuring that the predicted trends remain consistent with the gradual increase in maize / decrease in fish in the authentic mixes.

Figure 6. Actual against predicted percentage contribution of (a) C₄ plant oil and (b) freshwater aquatic oil obtained by applying FRUITS Bayesian modeling to experimental mixtures of modern maize and freshwater lake trout in 10% increments by dry weight. The boxes represent a 68% credible interval, whereas the whiskers represent a 95% credible interval. The horizontal continuous line indicates the mean, whereas the horizontal discontinuous line indicates the median.

Figure 7. Actual against predicted percentage contribution of (a) C₄ plant oil, (b) freshwater aquatic oil, and (c) wild ruminant adipose fats obtained by applying FRUITS Bayesian modeling to experimental mixtures of modern maize and freshwater lake trout in 10% increments by dry weight. The boxes represent a 68% credible interval, whereas the whiskers represent a 95% credible interval. The horizontal continuous line indicates the mean, whereas the horizontal discontinuous line indicates the median.

To illustrate the contribution of different food sources to 25 residues from the Dawson site, a Bayesian mixing model was implemented in FRUITS (Supplemental Text 1). In this model, δ¹³C_16:0 and δ¹³C_18:0 values were used as proxies, and three food groups were considered as potential sources: freshwater aquatic oils, C₄ plant oils, and ruminant adipose fats, with δ¹³C_16:0 and δ¹³C_18:0 reference ranges and concentrations as defined in Supplemental Table 5. Other potential sources—such as fruits, tubers, nuts, and leguminous plants—were not included in this model. Many of these have very low lipid content, whereas others, such as nuts, are likely to have values that overlap with freshwater aquatic oils. Given the predominance of aquatic biomarkers in our samples, the freshwater aquatic oil food source was favored. The mixing model presented here should thereby be viewed as illustrative rather than definitive. Its main utility undoubtedly lies in its ability to identify and estimate C₄ plant oil contribution to the residues, especially given that this foodstuff was invisible from the chromatographic profiles. Significantly, the model generated predicts the presence of variable amounts of maize in almost all the potsherds, with a 0% contribution credible in only two cases, represented by samples 15 and 27 (Figure 8a). According to the model, 40% (n = 10) of the samples contain over 50% of maize by dry weight. The latter have elevated isotope carbon values and significantly lower lipid yields (Mann-Whitney U = 39; p = <0.05) than the remaining 15 samples, indicating that these elevated carbon values reflect the presence of maize rather than marine aquatic resources in the residues from Dawson samples.

Figure 8. Estimated percentage contributions of lipids from different food sources to Iroquoian ceramics from the Dawson site using nonconservative model parameters. The boxes represent a 68% credible interval, whereas the whiskers represent a 95% credible interval. The horizontal continuous line indicates the mean, whereas the horizontal discontinuous line indicates the median.

Despite higher standard deviations associated and general uncertainties with the predicted proportions of freshwater oil and ruminant fats, the Bayesian mixing model shows a general trend where the amount of aquatic resource is inversely proportional to the amount of maize (Figure 8b). The predicted proportion of ruminant varies in a less predictable manner. In this case, a percent contribution below 3% is credible in 52% (n = 13) of the samples, which, as mentioned earlier, could be observed in cases where a food source absent from the sample is added to the mixing model. On the other hand, a minimum percent contribution exceeding 10% in six samples (#17, 19, 16, 10, 11, 21) likely reflects a real presence of ruminant mammal fats in these containers, although exact proportions cannot be estimated due to high uncertainty ranges (Figure 8c).