Prehistoric agriculture and social structure in the southwestern Tarim Basin: multiproxy analyses at Wupaer

The oasis villages of the Tarim Basin served as hubs along the ancient Silk Road, and they played an important role in facilitating communication between the imperial centers of Asia. These villages were supported by an irrigated form of cereal farming that was specifically adapted to these early oasis settlements. In this manuscript, we present the results from new archaeobotanical analyses, radiocarbon dating, and organic carbon isotopic studies directly from carbonized seeds at the Wupaer site (1500–400 BC) in the Kashgar Oasis of the western Tarim Basin. Our results showed that early farming in the oasis relied on a mixed wheat and barley system, but after 1200 BC was intensified through more elaborate irrigation, the introduction of more water-demanding legumes, and possibly a greater reliance on free-threshing wheat. These crops and the knowledge of irrigated farming likely dispersed into the Tarim Basin through the mountains from southern Central Asia. Improved agricultural productivity in the Tarim Basin may also have led to demographic and socio-political shifts and fed into the increased exchange that is colloquially referred to as the Silk Road.


Barley (Hordeum vulgare).
Three types of barley were identified at Wupaer. Based on grain apex morphology, we believe that both two-rowed (Hordeum vulgare var. distichon) (Fig. 2a) and six-rowed varieties of hulled barley (Hordeum vulgare var. vulgare) (Fig. 2b) were present. Additionally, smoother grained and more compact specimens of naked barley (Hordeum vulgare var. nudum) (Fig. 2c) were present. All three types contained a fusiform morphology. The grains of two-rowed hulled barley were the largest of the three types, with grain sizes reaching 6.0, 3.3, and 2.7 mm in length, width, and thickness, respectively. The lateral grains of six-rowed hulled barley were easy to recognize by their twisted form; this variety had grains smaller than the two-rowed type, and the average length, width, and thickness measurements were 4.22, 2.32, and 1.82 mm, respectively. Central grains on a spikelet of six-rowed barley were indistinguishable from grains of two-rowed barley, hence, we cannot discuss proportions of the two forms. Naked barley was slightly larger than six-rowed barley, and had an average grain size of 4.49 mm in length, 3.21 mm in width, and 2.26 mm in thickness.

Wheat (Triticum aestivum or T. aestivum/turgidum).
Wheat was the most abundant crop recovered from three of the pits, with a total of 548 carbonized grains in the assemblage, which accounted for 85% of the carbonized remains. The width/length ratio of most of the wheat grains was more than 2/3, suggesting that they fell into the identification index of compact morphotypes 40,41 (Fig. 3). While landrace varieties of any crop express greater variability within a variety than modern crop varieties, the compact wheat grains from Wupaer ( Fig. 2d) were nearly round and the sizes ranged from 2.4-4.5 mm in length, 1.6-3.7 mm in width, and 1.5-3 mm in thickness, the average width/length radio is 0.81. The handful of grains that we ascribed to the category of common wheat (Fig. 2e) displayed an oval shape and their size was 2.9-4.8 mm in length, 1.8-2.9 mm in width, and 1.5-2.0 mm in thickness, the average width/length radio was 0.58. Given the considerable variability between these grains, no clear break between the two populations, and the limited sample size, we cannot say with confidence if there were different morphotypes present or a high degree of variability within one formpossibly resulting from differences in watering regimes.
Millets. Millet grains were not recovered from the earliest contexts from the site, which seems to imply that they were later introductions, possibly in the late second millennium BC. Millets would have been known in this region and are recovered from earlier contexts in the mountains of Central Asia, hence, additional studies are necessary to verify their absence in the early layers at the site. There were four foxtail millet grains (Fig. 2f) recovered from Wupaer. One of the grains was measured, with a size of 1.8 mm in length, 1.3 mm in width, and 1.2 mm in thickness. A total of 11 grains of broomcorn millet (Fig. 2g) were found at the Wupaer site. One of these grains was measured as having a size of 1.8 mm in length, 1.5 mm in width, and 1.0-1.1 mm in thickness.  (Fig. 2h) found at Wupaer site, the average size of these legumes was 3 mm in length, 2.9 mm in width, and 2.0 mm in thickness. All of them had distinct radicals and embryo notches; they also had clear cotyledon splits. While we are cautious to make any interpretations based on two seeds alone, especially when we do not have representative comparative material for all wild Fabaceae in the region, there were two seeds that express some morphological similarity to cultivated Vigna sp. (Fig. 2i) as identified at sites in northern India and Pakistan 42,43 . Given the close proximity of Wupaer to the Kashmir Valley, where early Vigna have been identified, we think the possibility that there seeds represent a cultivated legume should be considered. Images of both seeds are presented in Fig. 2i, and they were both found at Wupaer 2-1, the size of seeds ranged from 3.1-3.  www.nature.com/scientificreports/ in archaeobotanical assemblages across arid Central Asia and they express a wide range in morphological variability; hence, we can't rule out the possibility that the two seeds mentioned above are exceptionally large Alhagi seeds.
Organic carbon isotope rations in ancient grains.

Discussion
Agricultural strategies between 1500-400 BC in the southwestern Tarim Basin. The only crops recovered from sediments dating to the early phase of occupation at the Wupaer site (1500-1300 BC) were naked barley and compact wheat (Fig. 4). This is not surprising, given that increasing evidence has illustrated that these two grains were the earliest domesticated crops to spread through the mountains of Inner Asia 42 . The long-season cereals were likely well-adapted to environments of the more northerly latitudes and spread more rapidly eastward, than did legumes or other cereal varieties, such as tetraploid wheats. Both cereals spread into the Hexi Corridor, around the peripheries of the Tibetan Plateau, and eventually into the middle and lower reaches of the Yellow River by the end of the third millennium BC 44,45 . While we can only speculate about what the agricultural strategies in the Taklimakan Desert during this early period looked like, based on analogies with neighboring regions 44 , we believe that a low-investment form of agriculture, near river edges or spring promotes complimented simple herding strategies. Supporting our results from early Wupaer, archaeobotanical data from Xintala (ca. 1900-1500 BC) 22 , also showed that the agriculture was dominated by compact free-threshing wheat and naked barley (Fig. 6a). At  www.nature.com/scientificreports/ the Gumugou Cemetery (1886-1746 BC), wheat was the only crop recovered 20 , and the crop remains from the Xiaohe Cemetery 32 and the microfossil analysis of a desiccated cake from the North Keriya Cemetery 25 indicated that both wheat and millet were present. While further archaeobotanical investigation may clarify nuances in the farming system, it appears that cereal crops dominated cultivation systems across western Xinjiang during the period from ca. 2000-1200 BC. The dominance of cereals likely reflects long-standing traditions in farming systems across the mountains and deserts of Central Asia, where these cultivation strategies likely originated 46 . Many scholars have noted similarities in cultural aspects and cultivation technology among peoples in the mountains of Central Asia and those in western Xinjiang during this period 7 . In this regard, the dispersal of early farming traditions may have followed, what Frachetti 11 referred to as the Inner Asian Mountain Corridor (Fig. 6b).
Based on the archaeobotanical, isotopic, and archaeological evidence, we think there is reason to believe that occupation and cultivation of crops at the Wupaer site changed between the early (1500-1300 BC) and late periods (1200-400 BC). Counting naked and hulled forms of barley separately, six grain crops and at least one legume were present in the late Wupaer period (1200-400 BC) (Fig. 4). It is worth noting this is the first identification of a legume crop in an archaeobotanical record from the Taklimakan Desert. Additionally, it is the first attempt to subdivide hulled barley into two-and six-rowed forms. When taken in combination with the greater archaeological visibility of the later period, it seems likely that there was an intensification of farming strategies after 1200 BC. The greater archaeological visibility likely correlates with a greater population, which may have fed into the increased exchange at this period.
Similar shifts in cultivation practices have been observed in other areas of Inner Asia, for example, in northern Xinjiang, naked barley was replaced by varieties of hulled barley and legumes 47 . Spengler 45 presented the possibility that the shifts in dominant barley varieties across Central Asia during the first millennium BC may be tied into changes in investment in irrigation. Two-and six-rowed hulled barley was widely cultivated across southern Central Asia before the crops were introduced into Xinjiang. For example, two-rowed specimens have been recovered, dating as early as 5000 BC at Chagylli and Togolok in Turkmenistan 48 , and six-row barley grains were found at Aunu (4500-1700 BC) 47 , Gonur (Phase I, 2400-1950 BC) 49 , and Sarazm (ca. 3500-2000 BC) 50 . Preserved ancient peas have been found across a wider region, including Anau South (3000-1700 BC) 49 , Adji Kui 1 (2400-1300 BC) 51 , 1211/1219 (FS20, 1400 BC) 41 , and possibly Gonur (Phase I, 2400-1950 BC) 49 in southern Turkmenistan, as well as Tasbas (2a, 1441-1262 BC) 52 in the far east of Kazakhstan. As the core region of the Oxus cultural milieu, peoples in southern Central Asia had developed a prosperous agricultural tradition. Wheat, barley, and legumes had been widely utilized by agriculturists across the mountains and deserts of Central Asia before the second millennium BC 46,49,52 . Compared with developed irrigation agriculture in southern Central Asia, agriculture in northern Central Asia was usually associated with low-investment crops, such as naked barley and millets, which were characterized by wider tolerance in water and temperature but lower yields 44,52,53 .
The proportion of naked barley grains recovered from sites outside Xinjiang, notably in southwest Asia and Europe declined and gradually disappeared during the first millennium BC 47 . While our data are not robust enough to make an exact analogy, it is interesting to note that, at the Wupaer site, naked barley seems to disappear after 1200 BC. This may reflect a choice in cultivation by famers, as naked barley is easy to process and cultivate, but its yield is relatively low, and it is more suitable for small-scale population cultivation and processing. After the emergence of efficient processing tools such as millstones and irrigation systems in the first millennium BC, hulled barley with higher yields became more polular 47 .
Peas have been reported at the site of Qasim Bagh (2000-1500 BC) and Kanispur (2700-2000 BC) in the Kashmir Valley of the Pamir Mountains 43,54 . Vigna sp. became a common crop in South Asia at least five millennia ago 55,56 , and entered eastern China around 2000 years ago 57 . Interestingly, mountain passes, such as the Swat and Kashmir Valleys historically served as routes of connection between the northern Indus and Central Asia or Xinjiang. Therefore, it is likely that new crops were traded through these valleys during the late Wupaer period (1200-400 BC), and would have been introduced into Kashgar Oasis by agriculturists from southern Central Asia (Fig. 6b).
Agricultural populations did not only introduce new crops, but they reshaped the structure of agriculture across the Tarim Basin. Wheat was clearly a prominent part of the economy by the tail end of the second millennium BC, and may have become more important than barley at Wupaer (Fig. 6a). At other sites in the Taklimakan Desert, with wheat, such as Qunbake Cemetery (955-680 BC) of Luntai County 58 , iron sickles were also found 59 . Wheat grains were also found at the ancient urban center of Yuansha (Djoumboulak Koum) (ca. 400-0 BC) 23,24 and at the Sampula Cemetery (mid-first millennium BC) 60 .

Water management in the Tarim Basin.
Wheat is one of the most important cultivated crops in the world 61 , but it has a long cultivation period and it is water demanding, which made it difficult to intensify before mechanized labor. Stable carbon isotope (δ 13 C) analyses of ancient grains can help interpret the level of water supply available to those crops, and therefore, provides a rough estimate for assessing grain yields. Researchers have suggested that water input during the growing period is positively correlated with a carbon isotope discrimination (Δ 13 C) value of wheat grains 62,63 .
The boxplot of Δ 13 C values and reconstructed water input of ancient grains from Wupaer shows that wheat during late occupation period generally had higher Δ 13 C values than the early period ( Fig. 5a and Table S2). Paleoclimatic studies in this part of the world do not suggest that a shift to more humid conditions occurred in the regions during the transitional period 64 . Therefore, we suggest that the best explanation for the differences in water input was provided to wheat by cultivators during its growth period. Obviously, there are many factors that can affect water input and carbon isotope levels, but given that a greater management of water through Scientific RepoRtS | (2020) 10:14235 | https://doi.org/10.1038/s41598-020-70515-y www.nature.com/scientificreports/ gravity irrigation of mountain melt streams is expected at this period, the isotope data seems to support the archaeology [65][66][67] .
Paleoecologists have proposed that as early as the second millennium BC, salinization of top soils, possibly caused by irrigation, was already occurring at the Xintala site 22 . The irrigation systems found in Mohuchahangoukou 67 and at the Yuansha ancient city (Djoumboulak Koum) (ca. 400-0 BC) 23,24 suggest that people were significantly improving water management technology. It is reasonable to assume, given the isotope values, that farmers in the Wupaer region might have irrigated wheat, despite the fact that archaeologists were unable to identify ancient canals on the heavily wind-deflated landscape.
Improved ability for water management is a sign of increased productivity and time investment in sedentary economic practices, but the size and shape of wheat grains do not appear to be affected by the increased water input. The measurements of wheat morphology show that compact wheat grain size during the late Wupaer period was not larger than that during the early Wupaer period (Fig. 5b). This would support assumptions that the highly compact morphotypes, as discussed by Spengler 45 , reflect non-plastic genetic mutations or aspects of complex spherococcoid syndromes. In some parts of Asia, dwarfing of semi-dwarfing traits in cereals are linked to ecological adaptations, such as greater drought tolerance, higher yields, and tolerance to heavy snow cover 45,46 . The highly compact grains in the Taklimakan Desert after 1200 BC, may suggest that semi-dwarfing varieties of wheat were preferred given their adaptability to the extreme continental climate of the Tarim Basin.  Central Asian archaeology is how best to model changes in social orders over time, especially during a period often referred to at the Bronze/Iron Age Transition (ca. 1200-700 BC). Traditional models of social development in Central Asia claim that this period marked the first appearance of highly specialized mobile pastoralists 30,[68][69][70] . The heavy focus on socio-political dynamics is tied directly into the assumption that all people in this broad part of the world lacked agriculture, making them a unique case study for imperial formation in the absence of grain surplus [71][72][73][74] . Scholarship in Central Eurasia often focuses on linear complexity models 11,72 , claiming that social orders in Central Asia were different in prehistory than in parts of the world where agricultural surplus was tied into sedentism and demographic expansion. While it is beyond the scope of this paper to discuss the specific nature of political systems in the oases of Xinjiang, they are often likened to city-states or small-scale polities. Archaeological evidence from across Xinjiang illustrate a greater density of sites and more expansive settlements during the first millennium BC 7 . Understanding what role irrigation played in these cultural changes in this hyper arid region will provide a reference for discussing the Bronze/Iron Age Transition in the arable mountain valleys across Eurasia. At the period of this supposed switching to more mobile and specialized pastoral economies, archaeobotanical data have illustrated that an intensification of agriculture was underway and more dense farming villages were forming in the mountain foothills to the west 75 . Chang 76 has argued for a much more intertwined economic system across the Tien Shan Mountains during the first millennium BC, and Spengler et al. 77 have suggested that the intensification of farming through irrigation was a key driver of socio-political changes in Central Asia. Li 7 recently discussed the intensification of irrigated farming in northern regions of Xinjiang, and other scholars have discussed the increasingly more prominent role of irrigation in farming in the foothills of Central Asia 78 . Miller et al. 79 suggest that even more intensive crop-rotation systems were being implemented in parts of Central Asia by the mid-first millennium B.C. and Spengler et al. 80 have discussed the importance of these complex multi-cropping systems in the Pamir Mountains by the tail end of the second millennium BC. Archaeobotanical data have also been used to demonstrate an increased prevalence of water-demanding crops in southern Central Asia 79 , and more intensive irrigation systems may have existed from the Murghab Oasis 81 , to Khorezm 82 , and Semirechye 77 in the late first millennium BC. Wilkin et al. 83 recently suggested that the introduction of millet and possibly cereal farming in Mongolia may have played a role in the development of more complex social systems there as well. The apparent switch to more heavily irrigated crops during the key transitional period in the oases on the peripheries of the Taklimakan Desert, further support the growing evidence that illustrate a link between more elaborate and hierarchical political systems and farming in Central Asia 43 .
Scholars have suggested that increasing the labor input into irrigation might have accelerated the emergence of city states 9 . According to historical texts 2,3,84 , more than 25 city-states were present in this region when the envoy of Zhang Qian arrived. He was appointed as ambassador by the emperor of the Han Dynasty to contact counties to the west of the imperial boundaries and reached this region in 129 BC. Considerable ruins of ancient towns of proto-urban centers dating before the Han Dynasty (before 202 BC) exist across the Taklimakan, such as Ahetu, Yuansha (Djoumboulak Koum), Andier, Qiemo, Loulan, Qiuci, and Wushikate 1 (Fig. 6b). Some of these ancient city ruins have been directly dated; for example, radiocarbon dating results show that Yuansha (Djoumboulak Koum) was constructed between ca. 400-0 BC (recalibrated by inCal 13, 2σ) 24,29,85 . Likewise, human activities at Loulan ancient city began as early as 200 BC 86 . Historical and archaeological data indicate that the oasis cities of the Taklimakan Desert were constructed before the colonization of the region by the West Han Dynasty 9 .
Supporting the discussions of crop dispersal discussed above, mtDNA analyses of fifteen human remains excavated from Yuansha (Djoumboulak Koum) suggests a relatively close relationships with modern populations of southern Central Asia and the Indus Valley, as well the ancient population of Chawuhu 23,24,85 . This conclusion further indicates bilateral connections between oasis populations in southern Central Asia and arid western China (Fig. 6b). Scholars have already suggested that the irrigation technology present in Xinjiang in prehistory may have spread into the region through the mountains of Central Asia from southern Central Asia 67 .
The multiproxy data that we present in this study leads us to conclude that early cities in the Tarim Basin were closely related to the populations from southern Central Asia. Eastward-moving agriculturists from southern Central Asia introduced the new crops, including legumes, and new irrigation technology into the Tarim Basin after 1200 BC (Fig. 6b). All of these factors likely articulated into a complex cultural realm leading to increased populations, the formation of large oasis towns, and increased exchange along the proto-Silk Road.

Methods
Radiocarbon dating. Seven carbonized seeds from three pits were chosen for dating by Accelerator Mass Spectrometry (AMS). The pretreatment process included: (1) cleaning the surface of the samples; (2) treating the samples with an acid-alkali-acid treatment; (3) combusting the sample to produce carbon dioxide; and (4) reducing it to graphite. Radiocarbon ages were measured at the Australian Nuclear Science Technology Organization (ANSTO) in Sydney, Australia. Other dates were measured at Beta Analytic in Miami, USA, which served as comparison between the two labs. The radiocarbon ages were calibrated by using the IntCal13 calibration curve in Oxcal v4.3.2 87 .
Archaeobotanical analyses. The cluster of sites at Wupaer is highly wind deflated and most sites consisted of surface scatters, as is characteristic of archaeological sites in arid regions of Eurasia. The lack of stratigraphic integrity hinders archaeobotanical investigation, but after a close survey of the sites we identified a few cultural layers and ash pits with secure contexts. We collected 11 sediment samples of 20-L at three of the sites in the cluster, the sites were designated as Wupaer 1, 2, and 3 (5 samples, 3 samples, and 3 samples were collected Scientific RepoRtS | (2020) 10:14235 | https://doi.org/10.1038/s41598-020-70515-y www.nature.com/scientificreports/ respectively). Flotation of the sediments was performed at a nearby fresh-water spring, using a basic bucket method, and screens of 0.3 and 0.6 mm. The samples were placed in a bucket filled with water and manually agitated with a clean wooden stick. Secondly, the floating carbonized remains were decanted into a screen and were cleaned in spring water. Lastly, carbonized remains were collected in cloth bags and dried in the shade. All carbonized remains were identified and measured under a stereoscopic microscope (Leica M205c) in the laboratory of Institute of Vertebrate Paleontology and Paleoanthropology (IVPP), Chinese Academic of Science (CAS), Beijing. Seed identification in this study relied on published archaeological and modern data 88,89 as well as the modern plant seed bank of the environmental archaeology laboratory of IVPP.
Carbon isotope analysis. Fifteen carbonized wheat grains were selected for measuring the ratio of organic carbon isotopes. The pretreatment process proceeded as follows: (1) carbonized seeds were ground into powder by using an agate mortar and seed powder was placed into a small beaker; (2) carbonate was removed by hydrochloric acid then washed to neutrality by using distilled water; (3) roughly 15 mg of dried powder was placed into a silver box and compressed into a small ball with tweezers; and (4) the sample was placed into the chamber for measuring the ratios of organic carbon isotopes. All samples were measured by using a MAT 253 Mass Spectrometer in the isotope laboratory of the IVPP, Chinese Academic of Science, Beijing.
In order to assess the available water supply for growing wheat plants during seed formation, equations proposed by Farquhar et al. 90 and Araus et al. 91 were used, as follow: where Δ 13 C is carbon isotope discrimination, δ a and δ p are δ 13 C of the atmosphere and the δ 13 C of wheat grains. The δ a date has been obtained from air bubbles trapped in Antarctic ice cores 92 .
where WI is water input, Δ 13 C is carbon isotope discrimination.