Fish and maize: Bayesian mixing models of fourteenth- through seventeenth-century AD ancestral Wendat diets, Ontario, Canada

Freshwater and marine fish have been important components of human diets for millennia. The Great Lakes of North America, their tributaries and smaller regional freshwater bodies are important Native American fisheries. The ethnohistorical record, zooarchaeological remains, and isotopic values on human bone and tooth collagen indicate the importance of fish in fourteenth- through seventeenth-century ancestral Wendat diets in southern Ontario, which is bordered by three of the Great Lakes. Maize (Zea mays ssp. mays) was the primary grain of Native American agricultural systems in the centuries prior to and following sustained European presence. Here we report new Bayesian dietary mixing models using previously published δ13C and δ15N values on ancestral Wendat bone and tooth collagen and tooth enamel. The results confirm previous estimates from δ13C values that ancestral Wendat diets included high proportions of maize but indicate much higher proportions of fish than has previously been recognized. The results also suggest that terrestrial animals contributed less to ancestral Wendat diets than is typically interpreted based on zooarchaeological records.

Maize is a plant that uses the C 4 photosynthetic pathway with archaeological δ 13 C values in northeastern North America ranging from −15.1 to −7.4‰, while most native terrestrial plants and agricultural crops in the region utilize the C 3 photosynthetic pathway with archaeological δ 13 C values in northeastern North America ranging from −28.6 to −23.3‰ [26][27][28][29][30] . As a result, consumers of maize and their consumers have δ 13 C values that are higher than consumers of predominantly C 3 photosynthetic pathway plants and their consumers. δ 13 C analyses of human bone collagen and apatite suggest that maize consumption began to increase gradually in southern Ontario by ca. cal. AD 500 27 . Using δ 13 C values on human bone and teeth, Pfeiffer et al. 28,29 suggest that by the cal. fourteenth to fifteenth centuries AD, maize was a dominant source of calories for ancestral Wendat individuals, remained so in the sixteenth century AD, and increased in the seventeenth century AD. Using a linear mixing model with δ 13 C values they estimate that maize contributed >50% of ancestral Wendat diets during those centuries, and perhaps as much as 65% 28 . However, they suggest that elevated δ 13 C values may also reflect the consumption of freshwater fish because fish bone from ancestral Wendat archaeological sites have δ 13 C values that range mostly between −21‰ and −16‰ 28 . Pfeiffer et al. 28 report a mean dog collagen δ 13 C value of −11.1 ± 0.9‰ and suggest its consumption may have affected human values. Less attention has been paid to stable nitrogen isotopes (δ 15 N) in terms of ancestral Wendat diets, given that the primary concern of isotopic studies in southern Ontario has been to trace maize consumption histories and has thus focused on δ 13 C values.
Consumers of animal flesh have higher δ 15 N values than do plants and primary consumers of plants. Omnivores with higher proportions of terrestrial plants in their diets will have lower δ 15 N than omnivores with higher proportions of animal flesh in their diets. Of note is that despite the dominance of maize in the diet, δ 15 27,30 , suggest that ancestral Wendat diets included high-trophic-level freshwater fish.
There is ample evidence from the archaeological and ethnohistoric records that freshwater fish was an important contributor to ancestral and historical Wendat diets 15,17,31 . Based on the ethnohistorical record, Heidenrich 15 estimated that fish accounted for 9% of seventeenth century AD daily Wendat diets. Fish bone generally comprises large proportions of faunal assemblages on seventeenth-century and earlier ancestral Wendat sites on which fine-scale recovery was used, and this evidence suggests varied fishing strategies based on time of year and  Figure 1 does not contain copyrighted material. The map was produced in ArcGIS v 10.6 at the New York State Museum, Albany, NY by compiling GIS data (shapefiles) obtained from publicly available data from Statistics Canada, the U.S. Census, and the United States Geological Survey. settlement location 32,33 . While ancestral Wendat δ 15 N values suggest fish were important components of diets, no quantitative assessments have been made based on those values 28,31,34 . On the basis of δ 15 N values Pfeiffer et al. 28 suggest that proportions of fish in diets were highest during the fourteenth and seventeenth centuries AD, when ancestral Wendat communities were closest to the north shore of Lake Ontario and the south shore of Georgian Bay, respectively. They suggest dietary fish proportions decreased during the fifteenth and sixteenth centuries AD when communities were located at greater distances from these two fisheries. Hawkins et al. 17 suggest their analysis of a large dataset of fish remains supports Pfeiffer et al. 's conclusion, and the contribution of fish to diets was particularly low during the sixteenth century AD before recovering to substantial contributions during the seventeenth century AD.
Based on differences in δ 13 C and δ 15 N values in bone collagen and dentine collagen samples, Pfeiffer et al. 28 suggest that there were differences between adult and child diets, respectively. Specifically, they suggest maize comprised larger proportions of child than adult diets, while animal protein comprised smaller proportions of child than adult diets.
Potentially complicating interpretations of δ 15 N values is the role of white-tailed deer (Odocoileus virginianus) in subsistence 34 . While deer is sometimes considered to have been an important component of ancestral Wendat diets 28 , deer bone frequently does not comprise high percentages of faunal remain assemblages 35 . Seventeenth-century AD ethnohistoric records indicate that deer was not a major component of Wendat dietsdeer were scarce in the area occupied by the Wendat at that time 15 . To the contrary, good deer habitat occurs to the south where ancestral Wendat communities lived prior to consolidation south of Georgian Bay in the seventeenth century AD 35 .
Here we apply a Bayesian dietary mixing model, MixSIAR 36 , to calculate the proportions of food sources in ancestral Wendat diets. MixSIAR allows multiple isotopic tracers (e.g., δ 13 C and δ 15 N) and uses a Markov Chain Monte Carlo simulation to model the probability of proportions of food sources in a consumer's diet based on the isotopic values of the food sources (e.g., maize, fish) and the consumer (human). As opposed to identifying point estimates of the proportions of certain foods in diets (e.g., maize) from the analysis of single isotopic tracers, Bayesian mixing models, including MixSIAR used here, incorporate the uncertainty in the isotopic values of the food sources, the consumer, and the trophic enrichment factor (TEF) between the food sources and the consumer in the model to ultimately produce a probability of proportions of food sources in a consumer's diet. An assumption of these models is that the isotopic tracer values of the food sources are representative of the diets of the sampled humans.
We use δ 13 C and δ 15 N values in human dentine and bone collagen and tooth enamel from Pfeiffer et al. 's 28,29,34 ancestral Wendat research, recently obtained isotope values for maize from ancestral Wendat sites, and a large database of terrestrial animal (n = 404) and freshwater fish (n = 170) bone collagen isotope data from southern Ontario archaeological sites, which we compiled from several sources 28,[37][38][39][40] , to model fourteenth through seventeenth century AD ancestral Wendat diets. We specifically test four hypotheses generated by the results of Pfeiffer H 2 : Diets incorporated proportionately more fish during the fourteenth and seventeenth centuries AD than during the intervening centuries, with fish being particularly minor dietary components in the sixteenth century AD diets. H 3 : Fish exploitation was largely concentrated on high trophic-level species. H 4 : Children, as reflected in dentine collagen and tooth enamel isotope values, had higher proportions of maize in their diets than did adults as reflected in bone collagen isotope values.
This article provides results of the first application of multi-tracer Bayesian dietary modeling on a large series of Native American isotopic values from eastern North America. The combined modeling of δ 13 C and δ 15 N values results in the first quantitative estimates of fish proportions in ancestral Wendat diets based on archaeological isotopic data.

Results
All model output and code are presented in the Supplementary Information File. Results of the three-source (i.e., maize, fish, and terrestrial species) MixSIAR models by century are summarized in Fig. 2 and Table 1. Isotopes are differentially incorporated into tissues such that the isotopic values in collagen (bone or dentine) derive mainly from ingested protein, while values in tooth enamel derive from the whole diet [41][42][43][44] . This routing of isotopes may influence interpretations of diet in populations if they are based solely on one tissue type. Additionally, these tissues may reflect different periods in life depending on when the tissue forms. Tooth enamel and dentine collagen form during tooth development and do not readily remodel, so the δ 13 C and δ 15 N values are reflective of intake at younger ages. Bone collagen, however, does remodel and, in general, the isotopic values reflect intake of an older individual. By comparing values from different tissues, such as tooth enamel and bone collagen, a more comprehensive understanding of dietary source proportions is derivable.
Mean values for the models evince similar trends, that is, the proportion of maize consumed increases while the proportion of fish consumed decreases over time. The dentine collagen-based model exhibits an increase in maize proportions through time, a steady decrease in fish proportions, and a slight increase in terrestrial animal proportions in the seventeenth century AD. The bone collagen model exhibits an increase in maize proportions from the fourteenth to fifteenth century AD followed by a plateau, an overall decrease in fish proportions, but with a plateau in the fifteenth and sixteenth centuries AD, and an increase in terrestrial animal proportions but with a plateau in the fifteenth and sixteenth centuries AD. Overall the dentine collagen model mean values exhibit higher maize and lower terrestrial animal proportions than the means of the bone collagen model. As would be (2019) 9:16658 | https://doi.org/10.1038/s41598-019-53076-7 www.nature.com/scientificreports www.nature.com/scientificreports/ expected these results are substantially the same when the results of individual site models are placed in relative chronological order (Supplementary File S1.4.3).
In both tooth enamel models (Fig. 2c,d), maize proportions increase with a plateau during the fifteenth and sixteenth centuries, while fish proportions decrease with a plateau in the fifteenth and sixteenth centuries. The tooth enamel carbonate models calculate higher mean proportions of maize and lower proportions of fish in Wendat diets, over the four centuries when comparing the collagen (bone and dentine) to tooth enamel model results (Fig. 2). This implies that the collagen models slightly under represent the proportion of maize in the diet. These results indicate we can reject H 1 -the dentine model exhibits a consistent increase in maize, the bone collagen model does not show an increase in the seventeenth century AD, and the tooth enamel models show an overall increase in maize with a plateau between the fifteenth and sixteenth centuries AD.
The  Fig. 2 and Table 2. We cannot reject H 3 given that high δ 15 N fish have the highest dietary proportion estimates for fish throughout the sequence in both models. www.nature.com/scientificreports www.nature.com/scientificreports/ Based on the preceding, it is evident that child and adult diets were different in terms of maize and terrestrial animal proportions, but not for fish. As a result, H 4 cannot be rejected-child and adult diets did differ in the respective samples.

Discussion
Previous δ 13 C analyses of human bone and teeth and the ethnohistorical record suggest maize was a major component of ancestral Wendat diets, perhaps comprising upwards of 65% of the diet. Bayesian analysis shows that maize dietary proportions increased in importance from the fourteenth to seventeenth centuries AD, at least in adults. As evidenced by the dentine collagen model, maize appears to also have been an increasingly important component of child diets through the sequence. The bone-and dentine-collagen models suggest that about 50% of the protein derived from maize, and the tooth enamel models imply mean maize proportions over 60% throughout most of the analyzed sequence supporting the assessment that maize at times was the major component of ancestral Wendat diets. One estimate of fish consumption of 9% is based on seventeenth-century ethnohistorical records 15 . The Bayesian three-source models indicate fish proportions decreased through time, but the mean calculations for proportion both the dentine collagen and enamel models of fish in the diet was always above 9% across the fourteenth through seventeenth centuries. In adult diets decreased dietary fish proportion was replaced by terrestrial animals. The three-source model using enamel δ 13 C follow the same trends for child diets with increasing maize and decreasing fish proportions. The decreased proportions of fish in child diets was replaced primarily by increased maize, combined with terrestrial animals in the seventeenth century AD in both the dentine collagen and enamel models. Previous δ 15 N analyses suggest high-trophic-level fish, that typically have high δ 15 N values, were an important component of ancestral Wendat diets. The five-source collagen models show that the high δ 15 N fish group, which includes salmonids, is the dominant fish source in all bone and dentine collagen model iterations.
These results suggest that maize and fish were regular components of ancestral Wendat diets throughout the sequence. Both constituted large proportions of the diet contributing to the formation of dentine and bone collagen as well as tooth enamel. Other foods, such as terrestrial animals contributed substantially less. Deer flesh apparently was not the primary source of animal protein, rather, fish was the primary source, probably due to an easier access to fish compared to deer. Large fish like salmons can be passively trapped from spring to fall with wooden hoop nets while deer were captured using fence and pens and needed a significant number of experienced people to build the structures and drive the deer into them 15 . The increased proportion of terrestrial animals in the seventeenth-century AD bone collagen model may reflect deer consumption, which would be at odds with poor deer habitat in the region south of Georgian Bay. In earlier centuries, with villages located in favorable deer habitat, the mean dietary proportion of terrestrial animals contributing to collagen formation was <10%, while during the fourteenth century, i.e. when villages were more concentrated close to Lake Ontario and closer larger fish sources, the mean value is ~2%. This latter value may also be explained if deer harvesting was seasonal as recorded in the ethnohistoric record and deer flesh was limited to immediate consumption or associated feasts. Similarly, the ethnohistoric record indicates dog flesh, which has high δ 13 C, was generally eaten ceremonially and at feasts rather than frequently 15 . This occasional consumption would not influence human isotopic values.
As with previous isotopic analyses of human tissue in southern Ontario 27-30 , our analysis is limited by small sample sizes. However, the present analysis contributes to our understandings of ancestral Wendat diets by  www.nature.com/scientificreports www.nature.com/scientificreports/ allowing us to assign quantitative estimates of fish and terrestrial animal dietary proportions for the individuals represented in the available samples. As in other areas of the world, while changes occurred over the centuries considered here, freshwater fish was a critical component of these individuals' diets, while terrestrial protein was of substantially less importance. These results should inform future inferences about ancestral Wendat diets based on zooarchaeological analyses.

Methods
For the Bayesian models, we used the MixSIAR GUI v. 3.1 in R v. 3.5.2 45 . Stable carbon (δ 13 C) and nitrogen (δ 15 N) isotope values from southern Ontario archaeological sources were used as the isotopic tracers. Data for animal bone recovered primarily from southern Ontario archaeological sites dating from AD 900-1700 (source) (Supplementary Information File S.2) were compiled from published sources 28,[37][38][39][40] . Nine of the assayed fish bones are from the St. Lawrence Valley in Quebec and six are nineteenth-century sites in southern Ontario 38 . Data for humans (consumers) were obtained from Pfeiffer et al. 28 ( Supplementary Information File S3). To utilize the human tooth enamel data in the Bayesian models that employ collagen stable isotope data for sources and trophic enrichment factors, we converted the tooth enamel δ 13 C values using a Δ 13 C enamel-collagen value of −5.7‰ (Supplementary Information File S1.2). (Zhu and Sealy, 2018). For the δ 15 N values, we used the dentine and bone collagen δ 15 N of the same teeth. Maize δ 13 C and δ 15 N values were obtained from ancestral Wendat site samples, reported here for the first time (Table S. 4.1) and are the first published δ 15 N values on maize in the region. Three-source and five-source models were run and presented on data from each site separately, then run and presented combining data from sites into their respective centuries as in Pfeiffer et al. 28 . Since Hawkins et al. (2019) found no zooarchaeological evidence of Atlantic salmon (Salmo salar) in 17 th c. Wendat localities, we ran additional models of the 17 th century data in MixSIAR removing samples of this species from the fish source (3 source model) and the high δ 15 N fish source (5 source model). Results of these additional models showed extremely limited changes in the modelled proportions of sources consumed (Supplementary Information File S1.4.1.1, S1.4.6.1) The three-source model groups included maize (n = 14), all fish (n = 170), and all terrestrial prey (n = 404). The five-source-model groups had the all fish group split into three different groups based on statistically significant differences in δ 15 N values among fish taxa. The statistically significant differences in δ 15 N values also capture known ecological differences in the analyzed fish species with the high δ 15 N fish encompassing offshore fish species, while the medium and low δ 15 N fish groups contain nearshore fish species. Therefore, the five-source model groups included, maize (n = 14), all terrestrial prey (n = 404), high δ 15 N fish (n = 104), medium δ 15 N fish (n = 43), and low δ 15 N fish (n = 23) (see Supplementary Information File S1.1). Terrestrial prey included the following species: bear (Ursus americanus, n = 50), beaver (Castor canadensis, n = 3), deer (Odocoileus virginianus, n = 191), fox (Vulpes vulpes, n = 6), groundhog (Marmota monax, n = 18), muskrat (Ondatra zibithecus, n = 3), porcupine (Erethizon dorsatum, n = 3), rabbit/hare (Leporidae, n = 8), raccoon (Procyon lotor, n = 47), squirrel (Sciurus carolinensis, n = 14), and turkey (Meleagris gallopavo, n = 61). While obviously not capturing all potential variation in isotope values for the various taxa and isotopic groupings, the source value dataset is large enough to avoid modeling issues arising from small source sample sizes 46,47 in the isotopic groups used here. Additionally, we found no statistically significant differences between salmon data from Ontario Ancestral Wendat sites and the remaining salmon samples implying baseline values from different watersheds were similar for the fish sources in the models (Supplementary Information File S.1.1). Because the maize δ 15 N values were higher than expected we carried out a series of experiments to determine if charring affects maize isotopic values and adjusted the maize δ 15 N values used in the Bayesian modeling accordingly (Supplementary Information File S.4). Sources were input as means and standard deviations. Collagen to collagen source (food) to consumer (human) trophic enrichment factors (TEF) used in the collagen models were +1.1‰ ± 0.2‰ for δ 13 C and +3.8‰ ± 1.1‰ δ 15 N 48,49 . The maize (source) to consumer (human) TEF for collagen was +5.0‰ ± 0.1‰ for δ 13 C, and +3.0‰ ± 0.1‰ for δ 15 N 43 . The Bayesian mixing models met the criteria of the Gelman-Rubin and Geweke diagnostics. Results are reported as means and standard deviations as well as percent credible interval range (posterior probabilities) from 0.025 to 0.975.

Data availability
All data generated or analyzed during this study are included in this published article and its Supplementary Information File.