Economic Diversification Supported the Growth of Mongolia’s Nomadic Empires

Populations in Mongolia from the late second millennium B.C.E. through the Mongol Empire are traditionally assumed, by archaeologists and historians, to have maintained a highly specialized horse-facilitated form of mobile pastoralism. Until recently, a dearth of direct evidence for prehistoric human diet and subsistence economies in Mongolia has rendered systematic testing of this view impossible. Here, we present stable carbon and nitrogen isotope measurements of human bone collagen, and stable carbon isotope analysis of human enamel bioapatite, from 137 well-dated ancient Mongolian individuals spanning the period c. 4400 B.C.E. to 1300 C.E. Our results demonstrate an increase in consumption of C4 plants beginning at c. 800 B.C.E., almost certainly indicative of millet consumption, an interpretation supported by archaeological evidence. The escalating scale of millet consumption on the eastern Eurasian steppe over time, and an expansion of isotopic niche widths, indicate that historic Mongolian empires were supported by a diversification of economic strategies rather than uniform, specialized pastoralism.


Results
Preservation of samples. We analyzed 80 bone collagen and 108 dental enamel samples from 137 individuals from 60 archaeological sites (Tables 1-4; Fig. 1). Samples were separated into four chronological periods based on relative and absolute dating (Early [Neolithic -Bronze Age], Early Iron, Xiongnu, and Mongol; see Supplementary Table S4  All of the human bone collagen samples included in this study had atomic C:N ratios between 3.1 and 3.5 and were thus within the accepted range for good collagen preservation 35 (Supplementary Table S1). The collagen yields of these samples ranged between 6 and 30%, with none falling below 1%, a further check of data quality 35 . Furthermore, the majority of collagen samples have greater than 11% N and greater than 30% C, within acceptable ranges 36 . Each bone collagen sample was run in duplicate, and averages are presented in Supplementary Table S1 along with their standard deviation.
Dental enamel carbon stable isotope results. The data from δ 13 C values of human enamel bioapatite are divided into the same chronological periods as the bone collagen data. The Xiongnu population had the largest range of stable carbon isotope values, followed by the Mongol period and Early Iron Age ( Environmental differences. As stable carbon and nitrogen isotope values may vary in different environments (i.e. temperature and aridity), to adequately assess human δ 13 C and δ 15 N values from normal steppe (>200 mL of annual precipitation) and dry (<200 mL of annual precipitation) regions, we also determined the average values for each environmental type (Table 3). In these tables we have separated the previously published faunal stable carbon and nitrogen isotope values into the "steppe" or "dry" regions according to modern annual rainfall 37,38 . Statistical tests. Boxplots of our results can be found in Fig. 2A . For the bone collagen data, both the Xiongnu and the Mongol average δ 13 C values were significantly higher than those of Early individuals (p < 0.05). The same trend was seen for tooth enamel δ 13 C, with Early Iron, Xiongnu, and Mongol samples having δ 13 C significantly higher than that of the Early group (for the overall p < 0.05, and the specific pairwise comparisons are available in Supplementary Table 6). There was no significant difference between average dental enamel values for the Early Iron, Xiongnu, and Mongol periods δ 13 C (p > 0.05). Bronze Age δ 15 N values were also significantly higher (p < 0.05) than those of the Early Iron, Xiongnu, and Mongol periods (Supplementary Table 7).

Isotopic temporal trends in Mongolia and environmental impacts.
Higher δ 13 C values in individuals from the Early Iron Age, Xiongnu, and Mongol periods could be the product of the increased direct consumption of C 4 crops or wild plants or animals consuming C 4 plants. It should also be understood that both Mongolians and foreign travellers would have been moving within and outside of the imperial borders, and dietary intake likely varied greatly in different regions. In areas with environmentally-linked variation in wild C 4 and C 3 plant distributions, such as Mongolia, it is important to rule out a climatically driven change (see Supporting Information Text 1). Modern plant samples from Mongolia have yielded δ 13 C values ranging from −28.3 to −23.4‰ for C 3 photosynthetic pathways and an average δ 13 C of −14.7‰ for plants following the C 4 photosynthetic pathway 39 . Notably, wild C 4 plants make up a much smaller proportion of Mongolian and other Central Asian environments than C 3 plants 40,41 . Overall, contemporary studies suggest that leaf δ 13 C values decrease with increasing mean annual precipitation 42 , both as a product of reduced C 4 plants in wetter landscapes and aridity-driven changes in δ 13 C among C 3 plants (see Supplemental Text S1). While C 4 plants make up a relatively limited portion of the biotic community today, we established local isotopic baselines for Mongolia in the past using archaeological fauna in order to determine if shifts in δ 13 C values through time are the product of environmental variations or social and economic choices. Isotopic studies of modern and archaeological herd animals have shown differences in δ 13 C values between more and less arid regions [43][44][45] , and that there is variation in the availability of C 3 and C 4 plants across the country 46,47 . While there were no fauna associated with the human remains collected for this study, we were able to use previously published faunal stable isotope data from the Minusinsk Basin of Siberia (just north of Mongolia)(MNSK, AD, AM; n = 21) 20,29 , the Gobi (BGC; n = 14) 30,48 , Gobi-Altai (SBR; n = 5) 30 , and north central Mongolia (EG; n = 13) 30 areas to show that regional herbivores generally consumed C 3 plants, with some having higher stable carbon isotope values, indicative of C 4 plant consumption, in the hyper-arid desert regions 30,48 .
Statistical tests further support this assessment, with humans having higher δ 13 C values than the available fauna in all periods, with the greatest difference occurring in the Xiongnu and Mongol periods (p ≤ 0.005) ( Figs. 2A and 3A). For terrestrial faunal remains (Fig. 3A), there is a significant correlation between δ 15 N and δ 13 C bone collagen values (R 2 = 0.64, p-value <0.001) which is a product of higher levels of aridity leading to a higher availability of C 4 plants in the natural vegetation cover. However, no such correlation is observed in humans (Fig. 3A,B), either between δ 15 N and δ 13 C bone collagen values (R 2 = 0.01, p-value = 0.15) or between δ 15 N bone collagen and δ 13 C enamel values (R 2 = 0.05, p-value = 0.13). Given this, alongside the consistent elevation of human δ 13 C values over the available fauna δ 13 C values, this indicates that higher δ 13 C values in human bone collagen and enamel is a product of direct consumption of non-wild C 4 plants.  www.nature.com/scientificreports www.nature.com/scientificreports/ Mean bone collagen δ 13 C values for faunal remains from steppe regions are typically C 3 (−19.3 ± 1.3‰), and the stable carbon isotopic offset between bone collagen of herbivores and carnivores is c.  www.nature.com/scientificreports www.nature.com/scientificreports/ food consumption. Furthermore, given that bone collagen reflects primarily consumed protein, and that millet has a poor protein content, the dietary caloric contribution from millet was likely much higher than its protein contribution 50 . This is corroborated by human enamel δ 13 C values given that this isotopic proxy reflects the carbon dietary mix 50 . Steppe human enamel samples for the later periods show mean δ 13 C values higher by c. 3.5‰ when compared to the Early period. For dry areas, mean human enamel samples δ 13 C values are higher by c. 1‰ when compared to the Early period, which indicates a temporal increase in millet-based food consumption although considerably smaller than that observed in the steppe regions as shown also in the model estimates for millet caloric contributions (Fig. 4).
Further evidence for C 4 plant consumption is offered by the distribution of isotopic values. For faunal remains there is a positive significant correlation (R 2 = 0.67, p-value <0.05, correlation coefficient =1.126) between δ 15 N and δ 13 C bone collagen values which is expected given that an increase in aridity leads to a higher availability of C 4 plants in the vegetation cover. A similar correlation, albeit with isotopic offsets, would be expected if humans relied exclusively on animal products. However, no clear environmentally-driven correlation is observed for the human groups. There is no significant correlation for the Early (R 2 = 0.65, p-value <0.08, correlation coefficient = 0.65) and Xiongnu (R 2 = 0.0, p-value <0.46, correlation coefficient = 0.11) periods, and although the correlation is significant for the Mongol period (R 2 = 0.13, p-value <0.02, correlation coefficient = 0.39), it only explains 39% of the variability. For similar δ 15 N bone collagen values across the human individuals, there are wide ranges in δ 13 C collagen values. Whereas during the Xiongnu period one can observe a significant negative correlation (R 2 = 0.0, p-value <0.46, correlation coefficient = 0.11), which implies the contribution from a food source with higher δ 13 C values but lower δ 15 N values when compared to animal food sources. These isotopic relationships are indicative of varying individual intake of a food with elevated δ 13 C values, such as millet, and having relatively uniform δ 15 N and δ 13 C values across regions with varying levels of aridity.
Bayesian spatial modelling of C 4 plant caloric consumption. To further confirm that the increased δ 13 C values in human bone collagen and tooth enamel through time is a product of the consumption of crops rather than changing availabilities of baseline C 4 /C 3 plant ratios or the availability of samples in different local environments, we developed a Bayesian model to produce a C 4 dietscape, representing estimates of spatial distribution of C 4 plants based on per capita caloric consumption (See SI for detailed discussion). Stable carbon isotope data of dental enamel was used, and individuals were separated into two periods, Early (Neolithic -Bronze Age) or Late (Xiongnu, and Mongol). The results for the two models show that during the Early period C 4 caloric contributions were very low across Mongolia, likely including consumption of local plants and livestock consuming natural vegetation, with mean estimates varying between c. 2.5 and 5.0% of calories (interpolation 1-sigma uncertainty up to 0.5% calories) (Fig. 5A,B). During the later periods, the variability in millet-based food consumption increases considerably as shown by the range in the mean estimate (between 3 and 26% of per capita millet calories) and in the 1-sigma interpolation uncertainty for each location (between c. 3 and 6% of per capita millet calories) (Fig. 5C,D). The C 4 plant dietscape for the late period also shows that millet consumption is concentrated in central northern Mongolia (reaching the highest mean value [26% per capita millet calories]), an area where environmental increase of carbon values would not be expected naturally (Figs. 4 and 5).

Isotopic indicators of diet through time in mongolia.
Our results clearly demonstrate an increase of human consumption of C 4 plants during the imperial periods in ancient and historic Mongolia (Figs. 2, 3, and 5). While high δ 15 N values in human bone collagen relative to the faunal data ( Fig. 2A and 3A) supports evidence for human reliance on dairy and meat products throughout the periods under study, the change in C 4 plant   Table 3. Average human and faunal bone collagen δ 13 C and δ 15 N values between the steppe (>250 mL annual precipitation) and dry (<250 mL annual precipitation) regions.   In the rest of Mongolia during the tenure of the Xiongnu and Mongol empires human stable carbon values became more varied with increasing numbers of individuals displaying bone collagen and tooth enamel δ 13 C values suggestive of moderate to high C 4 plant consumption, with the number of individuals with such values reaching their peak during these imperial periods. We also observe the largest range and diversity of δ 13 C and δ 15 N values during the imperial periods. This is likely due to diverse subsistence strategies being pursued across each empire, reflecting different environmental zones and levels of imperial support. This is result of the extensive range of each empire, and includes the knowledge that not everyone that died in Mongolia would have been "Mongolian", but these individuals likely lived and died within the empires. Since the majority of the individuals analyzed were excavated from elite imperial tombs, and human remains representing other sectors of society are lacking at present, attributing all outliers to non-local outsiders would be to dismiss the agency of Mongolian populations and provide something of a 'colonial' narrative.
Individual bone collagen and tooth enamel δ 13 C values for the Xiongnu and Mongol empires range between those indicative of a pure C 3 diet to those that suggest heavy C 4 plant consumption. Interestingly, during this period, a few individuals had δ 13 C values lower than those of the Early period which, alongside lower δ 15 N values, indicates a staple intake of C 3 plants, likely crops such as wheat and barley. Historical and archaeobotanical sources suggest that cereal crops were commonly cultivated or obtained through trade during the Mongol period 13,51-56 . In addition to grains, carbonized fruit and nut remains have been recovered from sediments at the Mongol capital of Kharakhorum (also used during the Mongol rule in the Yuan Dynasty) showing the diversity of imported plants through the presence of rice (Oryza sativa L.), over a dozen cultivated fruits, including grapes (Vitis vinifera L.), figs (Ficus carica L.), and jujube (Ziziphus jujube Mill.), as well as vegetable and oil-seed crops. There are also remains of spices -notably a few, such as black pepper (Piper nigrum L.) and caraway (Carum carvi I.), that were imported along the trade routes with South Asia, and would have involved transport across distances of up to 2000 kilometers 12 .
The resulting bone collagen δ 13 C and δ 15 N values have been plotted to show this increase of dietary diversity over time (Fig. 2A). From our data, alongside the growing corpus of biomolecular, archaeological, and historical data, it is evident that the Xiongnu and Mongol Empires had complex imperial structures that facilitated increasingly diverse subsistence economies. The combination of crop cultivation in tandem with dairy pastoralism would have allowed these empires to sustain a diverse economic surplus that defended against livestock depletion from harsh winters, crop loss, or volatile political episodes. Diverse dietary values likely also reflect an increasingly cosmopolitan society in which dietary heterogeneity within populations increased with growing migration, trade and interaction, and the emergence of increasingly elaborated elite statuses. The diversity could also reflect temporal political shifts within the time-span covered by our sampling groups, with trade routes to Karakorum decreasing in volume during the Mongol Period after the switch of the capital in 1260 and ending with the end of the Yuan Dynasty in 1368 12,57 , for example.  58 . Nevertheless, our data clearly highlight that pastoral lifestyles did not preclude the inclusion, and later intensification, of crop use. Millet's suitability to arid environments combined with its short growing period is compatible with the often peripatetic, mobile lifestyles of pastoralists 22,26 . Indeed, during the Xiongnu and Mongol empires, we see clear evidence for human dietary reliance on millet in a significant proportion of individuals. Although some scholars contend that all grains were either extorted or imported from China and other exterior polities 6 , we argue that our data, alongside existing archaeobotanical and archaeological findings 25,59,60 , provide clear evidence for imperial reliance on locally grown crops in the Xiongnu and Mongol heartlands, as well as the coordination of diverse economic connections and exchanges 12 . These discoveries bolster the notion of an economically diverse population across much of Mongolian history 14,61,62 .
Agricultural tools for plowing, hoeing, and grinding have been uncovered from permanent Xiongnu settlements in Mongolia, implying local plant cultivation and processing 63 , and charred remains of millet, barley, and wheat grains have been recovered through flotation at pit-house villages at Boroo 64 and ephemeral campsites 61 . Studies in the Egiin Gol valley and at the large site complex of Ivolga have illustrated the presence of long-season cereal crops (wheat and barley) in the Iron Age, which represent more labor investment in farming practices than millets 61,63 . At Ivolga, this occurs alongside evidence of ploughshares at permanent settlements 63 , as well as written accounts of crops suitable for the northern steppe being managed by imperial Xiongnu administrators, such as the 'Lord of Millet Distribution' , referred to in the 1st century C.E. Chinese accounts 54 . Millet grains, still articulated in their chaff, have been found within the graves of Xiongnu rulers at Gol Mod and Noyon Uul 27 as well as of local elites throughout the steppe 60 . There are also uncharred grains found within Xiongnu pit-house villages, all of which were unprocessed (i.e. with palea and lemma) and thus most likely not transported long distances 60 instead representing local production and consumption (Fig. 4).
Scholars working in Mongolia have extensively discussed the formation of hierarchical political systems and greater concentrations of population densities in the absence of farming, often describing imperial systems in Central Eurasia as unique due to their economic basis [65][66][67][68][69] . In other parts of Asia, farming is linked to demographic expansion and the congregation of greater population densities 70 . Notably, millet farming is linked to urbanization 71,72 and imperial formation 73 in East Asia. Boserupian economics suggest that increased investment in farming, along with a diversified economy and higher levels of cultural exchange, often lead to a demographic transition [74][75][76] . The data presented in this paper suggests that, while Mongolian empires have often been seen as outliers in global comparisons of imperial structures, they were in fact, like many others around the world, highly reliant on economic diversification, local adaptations to a diversity of environments, and the creation of reliable and stable subsistence resources and economic surpluses [8][9][10] . www.nature.com/scientificreports www.nature.com/scientificreports/ Mongolian empires have traditionally conjured up exotic ideas of mobile pastoral specialists who roamed the Asian steppes attacking more sedentary communities [1][2][3][4] . While prominent in the public sphere, such preconceptions have also directed the type of questions academics have asked. For example, comparative analysis of Mongolian empires with others around the world has been limited, with 'Steppe Empires' often portrayed as deficient or somehow doomed to failure in the absence of reliable crop-based surplus 6,77 . As in other parts of Central Asia, where occupation sites have been hard to come by 78,79 , simplistic projections of ethnographic and ethnohistoric datasets into the past have been common in Mongolian archaeology. We hope to have demonstrated how multidisciplinary approaches, built on datasets from different parts of Mongolian imperial networks, can begin to provide novel insights into their economic systems and, perhaps most importantly, their geographic and temporal variability. While there is no doubt that the Xiongnu and Mongol empires were unique, they were also built upon many of the same tenants of economic diversity, stability, and reliability that have characterized imperial structures throughout prehistory and history, demonstrating the importance of a core set of underlying variables in both enabling and driving the formation of empires.

Methods and Materials
Sites and materials analyzed. All bone and tooth samples included in this study were collected from the National University of Mongolia's Department of Archaeology during the winter of 2016. Bone collagen was analyzed for carbon (δ 13 C) and nitrogen (δ 15 N) stable isotopes, and the carbonate of dental enamel bioapatite was measured for carbon stable isotopes (δ 13 C), with some individuals analyzed for both bone collagen and bioapatite (see Supplementary Table 3). Time periods for samples ranged from the mid-fifth millennium B.C.E. to the Mongol Empire, as dated by AMS radiocarbon methods where possible (see below).
Samples of bones and teeth were collected from archaeological sites across the country of Mongolia, through varying environmental and topographical zones. Where possible, we collected a tooth and long bone fragment from individuals from each time period. While we aimed to assemble an equal number of samples from all time periods, the collection was dominated by individuals from the imperial periods, resulting in fewer individuals prior to the Iron Age. Bone collagen was preferably extracted from rib bones, but occasionally other bone fragments were employed (clavicle, femur, crania). δ 13 C and δ 15 N stable isotope measurements of human bone collagen inform primarily on protein source 80 , and the bones sampled (i.e. ribs) represent a period of diet of approximately the last 20 years of life 80 .
By contrast, tooth enamel δ 13 C values are indicative of the whole dietary carbon (carbon mix of protein, lipids, and carbohydrates) consumed during enamel formation 33,49 . First molars mineralize before an individual is 3 years old, second molars are fully formed around age 8, and third molars, if present, are completely mineralized between the age of 7 and 16 81 . To avoid tooth samples that might show a breastfeeding isotopic signal contribution in older children and adolescents we preferentially selected second and third molars. First molars were chosen only when both the M2 and M3 were unavailable.
Stable isotope analysis methods. Bone collagen. We selected ribs for bone collagen analysis as representative of the last c. 20 years of life 80 . Collagen was extracted from each rib sample following standard procedures 35 . Approximately 1 gram of pre-cleaned bone was demineralized in 10 ml aliquots of 0.5 M HCL at 4 °C, with changes of acid until CO 2 stopped evolving. The residue was then rinsed three times in deionized water before being gelatinized in pH 3 HCl at 75 °C for 48 hours. The resulting solution was filtered, with the supernatant then being lyophilized over a period of 24 hours.
After calculating the collagen yield, all purified collagen samples (~1 mg) were located in tin capsules to be analyzed in duplicate at the Department of Archaeology, Max Planck Institute for the Science of Human History by the elemental analyzer/continuous flow isotope ratio mass spectrometry (EA-IRMS) using a ThermoFisher Elemental Analyzer coupled to a ThermoFisher Delta V Advantage Mass Spectrometer via a ConFloIV system. Tooth enamel. Teeth or tooth fragments were cleaned using air-abrasion to remove any adhering external material. 8 mg of enamel powder was obtained using gentle abrasion with a diamond-tipped drill along the full length of the buccal surface or fragment in order to maximize the period of formation represented by the resulting isotopic analysis for bulk samples. Enamel powder was pre-treated using a protocol to remove any organic or secondary carbonate contaminates 21 . This consisted of a series of washes in 1.5% sodium hypochlorite for 60 minutes, followed by three rinses in purified H 2 O and centrifuging, before 0.1 M acetic acid was added for 10 minutes, followed by another three rinses in purified H 2 O (as per 35 ).
Following reaction with 100% phosphoric acid, gases evolved from the samples were analyzed to stable carbon and oxygen isotopic composition using a Thermo Gas Bench II connected to a Thermo Delta V Advantage Mass Spectrometer at the Max Planck Institute for the Science of Human History, Jena (MPI-SHH). Carbon and oxygen isotope values were compared against an International Atomic Energy Agency (NBS 19) and in-house standard (MERCK). Replicate analysis of internal bovid enamel standards suggests that machine measurement error is c. ± 0.2‰ for δ 13 C and ± 0.2‰. Using a Thermo Gas Bench 2 in tandem with a Thermo Delta V Advantage Mass Spectrometer at MPI-SHH, gases produced from a reaction with 100% phosphoric acid were analyzed for stable carbon and oxygen isotopic composition. We compared the resulting values against International Standards (IAEA-603 (δ 13 C = 2.5; δ 18 O = −2.4); IAEA-CO-8 (δ 13 C = −5.8; δ 18 O = −22.7); USGS44 (δ 13 C = −42.2)); as well as an in-house Bayesian dietary modelling. Caloric estimates of millet intakes were obtained using the Bayesian mixing model FRUITS having as input data individual tooth enamel δ 13 C values and local food isotopic values adjusted for spatial variations due to varying environmental conditions 80 . To achieve the latter, we grouped site locations into the categories of "steppe" and "dry" depending on modern day annual precipitation. Steppe sites have a range from 250-350 mm in precipitation per annum and arid sites have below 250 mm of yearly rainfall. It was assumed that the enamel δ 13 C signal is defined by the dietary carbon mix 50 . To extrapolate the spatial distribution of per capita millet caloric intakes (dietscape) a Bayesian additive mixed model with error-in variables [83][84][85] available as an online app via the Pandora & IsoMemo initiatives was employed 86 . Dietscapes were generated for two main periods corresponding to a temporal divide defined by the intensification of millet consumption as observed from the interpretation of raw isotopic data, into Early (Bronze Age) and Late (combining the Early Iron Age; Xiongnu; Mongol periods). Modelling at a higher chronological resolution was not possible given a lack of data for shorter time periods. Further details on dietscape modelling are available in Supplementary Text S3.
Radiocarbon and archaeologically classified dates. AMS radiocarbon dates were conducted at the Oxford Radiocarbon Accelerator Unit (ORAU), Oxford, England, UK (n = 14; bone collagen and dentine) 87 and at the University of Groningen, Faculty of Science and Engineering, Groningen, The Netherlands (n = 25; bone collagen and dentine) 88