## Introduction

As one of the most recognizable world heritage sites, the Great Wall of China is a manifestation of the engineering capabilities and architectural achievements of multiple Chinese dynasties1. What is perhaps less well known, however, is that the iconic brick and stone walls built during the Ming Dynasty in the 15th century AD2, are only part of a series of multi-material fortifications that stretch across northern China from Hebei Province to Xinjiang Uyghur Autonomous Region (Fig. 1, Data S1)1,3,4,5,6,7. Walls and fortresses dating back to as early as the Warring States period (475–221 BC), were constructed using locally available materials, with reed fascines and wood bundles interbedded with gravel-mixed rammed earth (Fig. 2). Following the unification of China in 221 BC, an array of fascine and rammed-earth ramparts was established to protect against the powerful northern Xiongnu and Xianbei states8,9, and then in the early 2nd Century BC, these border defenses became essential in expanding the territories of the Han Dynasty from the central Chinese plains into the western frontier (including today’s Xinjiang and Gansu Province)4,10.

While remnants of the Han and later dynasty walls along the Shule River in Gansu and Inner Mongolia Autonomous Region have been surveyed11,12,13,14, isolated beacon towers (Fig. 2c) along the Kongque River (Fig. 1b) in Xinjiang are still relatively unknown despite being described in historical texts like the 5th Century AD “Book of Later Han”9. Nevertheless, recent surveys in Gansu and Xinjiang have identified new and well-preserved remnants of the ancient Great Wall in what was once the western reach of Imperial China15. These fortifications served as a military communication and warning system, symbolic political borders, and rest stops for merchants traveling along the ancient Silk Road4,6,7,16.

In the vast arid regions of central Asia, human livelihood is ultimately dependent on oases that supply freshwater to sustain both natural vegetation and agricultural crops17. Ancient walls, small and large forts, beacon towers, lookout platforms, and other structures were constructed and fortified using locally available plants harvested from such oases5, with Phragmites Adans. (Poaceae family), the cosmopolitan common reed, being the most conventional genus used in fascines as natural building material (Fig. 3). Phragmites is a highly successful C3 plant genus that has considerable variation with high phenotypic plasticity, a wide geographic distribution, and the ability to occupy aquatic and marginal habitats under various climate conditions18,19. When available, Phragmites was mixed with hardwood species in Great Wall fascines for added strength and durability1. Despite biochemical analyses on tissues from the species Phragmites australis (Cav.) Trin. ex Steud.20,21,22, reports on pollen23 and phytoliths24 of Phragmites in archaeological contexts, and the identification of a rope made from culms (stems) of Phragmites at the Gumugou Cemetery, a site dated to 3800 years cal BP and ~ 70 km east of Lop Nur25, no molecular characterization or isotope measurements have been taken on the ancient remains of Phragmites collected directly from the Great Wall itself.

Today, a large portion of northwestern China, including the Tarim Basin, the Hexi Corridor in Gansu, and the area west of the Helan Mountains of Inner Mongolia, has a semi-arid to arid continental climate with hot summers and cool, dry winters characterized by low rainfall and prolonged droughts26,27 (Fig. S1). This region, and specifically the eastern Tarim Basin, is a key geographical crossroad between Central and East Asia, holding political, military, cultural, and economic significance historically6. It has also been subject to both imperial expansion and agricultural intensification over the last two millennia15,28,29,30,31. Key to this expansion was the distribution of oases, unique landscapes that integrated natural and human-engineered ecosystems, which sustained human populations, economic growth, and political agendas17. Desertification brought on by natural climate changes and amplified by human activities32,33,34 intensified evapotranspiration35 and the proliferation of desert and xeric shrub-land habitats in the region36, reducing available oases today. Such changes have potential historical corollaries, however, as exhaustive irrigation farming and an overdrawing of highland tributaries during the Han Dynasty were believed to have changed local hydrology and caused salinification of lakes bordering the Tarim Basin30,37.

Extensive lacustrine31,38,39,40,41,42,43,44,45, speleothem46,47, and ice core48,49 records from northwestern China demonstrate long-term hydroclimate and environmental changes since the Han period. Yet, these records are distal to archaeological sites and do not necessarily reveal on-site (i.e., proximal) ecological subtleties that would have existed at various locations along oases in otherwise arid northwestern China when ancient Great Wall segments were built. In fact, there are no on-site paleoenvironmental reconstructions from northwestern China that speak directly to the creation and maintenance of altered ecosystems by the Han or later dynasties.

Here, we characterize the molecular preservation and distribution of plant wax n-alkanes and measure bulk carbon (δ13C) and nitrogen (δ15N) isotope ratios of accelerator mass spectrometry (AMS) dated Phragmites sampled directly from 13 Great Wall sections or beacon towers constructed during the Han, Jin, Tang, or Song Dynasties in today’s Gansu and Xinjiang. Our on-site data provide novel evidence for the source and diversity of reeds used in Great Wall building activities. We compare the ancient data with biochemical information obtained from modern P. australis available from the study area to reveal climatic and environmental dynamics during early Chinese historical periods with a focus on the rate, timing, and cause of hydrological changes. Biomolecular analyses of these organic materials preserved within ramparts may provide direct evidence of climatic and environmental conditions at specific historical points along the Great Wall. This study also highlights the future potential of these in situ material as valuable biogeochemical archives for studying human-altered ecosystems and hydrology.

## Results

### Accelerator mass spectrometry (AMS) 14C dating

To constrain the chronology of each sampling location, we obtained eight AMS radiocarbon ages from four sites (Table 1 and see “Methods”). We determined that the Majuanwan Great Wall (Site 7 on Fig. 1), and thus the eastern cluster sites in Gansu (Sites 1–7), were built during the Han Dynasty between 132 and 116 BC (2082–2066 cal. B.P., 95% prob.), adhering to Han era historical records and artifacts associated with these structures (Supplementary Materials). The Yakelun and Sunji beacon towers (Sites 11 and 12) in Xinjiang were constructed during the Tang Dynasty and date between 677 and 726 AD, confirming recent archaeological findings50,51. Finally, and somewhat surprisingly, the Sishilidadun Beacon Tower (Site 14) dates to the Song Dynasty, 1030–1160 AD, when the eastern Tarim Basin was governed by the Western Liao Dynasty.

Other sampling locations were previously dated. The Milan Castle Heritage Site (Site 8) was constructed during the Tang Dynasty and radiometrically dated to ~ 770 AD28. The Yingpan City Buddha Tower and Burial Site (Site 9) was radiocarbon dated to a median age probability of 305 AD in the Jin Dynasty52, consistent with the funerary artifacts found in the Yingpan cemetery53. Due to similar archaeological contents found in burials54,55, the nearby Yingpan City Heritage Site (Site 10) is assigned to the Jin Dynasty as well, although it is likely that Yingpan City was continuously occupied since the Han Dynasty. Finally, we attribute the Tahaqi Beacon Tower (Site 13), to the Song Dynasty period due to its proximity and similar style of construction to Sishilidadun.

### Pyrolysis gas chromatography mass spectrometry (Py-GC-MS)

With minor variations in pyrolysis moieties among different sites, ancient Phragmites culms (n = 6) consistently exhibit excellent molecular preservation with abundant labile biomolecules (Fig. 4 and Fig. S2; Data S2). Lignin and polysaccharide pyrolysates dominate the molecular composition of culms from Han and Tang Great Wall segments, beacon towers, and fortifications and have a similar distribution pattern as modern P. australis (n = 4, Fig. 4). Ancient samples do, however, contain compounds that are not identified in modern analogs, such as apocynin and desaspidinol, though except for one sample, they lack amino acids (Data S2). Apocynin and desaspidinol likely derive from lignin decomposition products56 or as possible indicators of hardwood cross-contamination57, as hardwood species such as Tamarix sp. were sometimes mixed with Phragmites in Great Wall fascines1.

Major pyrolysate compounds identified in all samples are benzene and furan derivatives, phenol derivatives, and indole derivatives of amino acids (Data S2). Lignin moieties contain phenol, methyl and methoxy phenol, vinyl phenol, and vanillin, while polysaccharide moieties include furans and furfural, benzofuran, and levoglucosan. Lipids are detected primarily as n-alkanes and palmitic (C16) and stearic (C18) acids, but dodecanoic (C12) and tetradecanoic (C13) acids were identified in one modern P. australis leaf collected near Yumenguan (Site 5). Indoles (e.g., Indole, 3-methyl-) indicate the presence of amino acids in modern P. australis samples. There are some variations in compound distribution among ancient samples collected from different sites (Figs. 4 and S2). For example, ancient culms from Han Dynasty Yumenguan (Site 5), have fewer lignin derivatives but contain identifiable fatty acids compared to contemporary Majuanwan (Site 7) samples which have more abundant lignin derivates but fewer overall polysaccharide compounds (Data S2). Nevertheless, the Tang era Sunji Tower sample has a similar suite of pyrolysis products as the older Yumenguan and Majuanwan culms, further highlighting the excellent preservation of Han era samples.

Even though site-specific pyrolysates may vary, the high polysaccharide and lignin fiber content of ancient Phragmites, compounds that provide strength and durability to culms, coupled with the arid regional climate15, resulted in the long-term preservation of these plant parts in the wall fascines as revealed with SEM (Fig. 3). This excellent preservation also suggests that the absence of leaves and inflorescences/infructescences in the walls was an intentional sorting process for high fiber material, as these plant parts would have also been preserved had they been used in wall construction.

### Lipid concentration and distribution

Culms from ancient wall segments or beacon towers have wider n-alkane distributions than their modern P. australis analogs even though they contain relatively lower quantities of lipids. The concentration of n-alkanes (C21–C33) in ancient culms is approximately 12-times lower than that of modern samples (Data S1), having between 12 and 8610 μg (1160 ± 1702 μg/g; n = 38) per gram of dry material (μg/g) compared to the 4163 to 32,296 μg/g (14,065 ± 8042 μg/g; n = 12) measured in modern plants. Overall, there is a significant difference in C21–C33 n-alkane abundance between modern and ancient samples as shown by a Student’s t-test (two-tailed, p = 0.0002) and Mann–Whitney U test (p = < 0.0001).

The lower concentration of n-alkanes in Great Wall samples is expected given their antiquity. However, the carbon preference index of the C21–C33 n-alkanes (CPIC21–C33), a metric used to examine the odd-over-even carbon number predominance and as an indicator for hydrocarbon maturity58, indicates that no significant degradation occurred among the longer chain n-alkanes from ancient samples. Both modern and ancient CPIC21–C33 values are ≥ 2.0 (Fig. 5, Data S1), which is typical of plant-derived CPI values59,60. There is also no significant difference between modern and ancient reed CPI in a Student’s t-test (two-tailed, p = 0.634) and Mann–Whitney U test (p = 0.827). Like lignin and polysaccharide, lipid preservation is contingent on the dry regional climate which likely facilitated the preservation of these organic archaeological remains over time15,61,62.

Figure 6 shows the ternary diagrams of the C27, C29, and C31 relative abundances for n-alkanes from ancient Phragmites and modern P. australis. The wider chain-length distribution of n-alkanes in ancient reeds contrast with those of modern samples, suggesting that ancient Phragmites used in the construction of the Great Wall were harvested from habitats that were likely more diverse and growing under cooler and wetter climate conditions relative to today, as chain-length distribution in P. australis has been shown to be a function of temperature and water availability22,63. Of the 33 ancient samples containing enough lipid material for GC-FID analysis, 13 (39.4%) have C27 as the most dominant n-alkane, while C29 and C31 are most abundant in 9 (27.2%) and 8 (24.2%) samples, respectively. The sample from Sishilidadun has C21 as the most dominant compound. This wider n-alkane distribution contrasts with modern reeds where 9 of the 12 samples (75%) have the C29 homologue as the most dominant compound, while only 2 (16%) samples from Sishilidadun have C27 as the most abundant alkane, and 1 sample (8%) from Yumenguan has C31 (Fig. S3).

The average chain lengths (ACL21–33) for Han and Tang aged reeds overlap with the distribution of modern ACL21–33 values (Fig. 5) and are consistent with previous reports of modern P. australis ACL from China20,21,22,64. ACL21–33 values from the Jin and Song Dynasty samples, however, are progressively lower than their modern counterparts (Fig. 5). Although there was no significant difference in ACL21–33 values between all modern and ancient reeds in a Student’s t-test (two-tailed, p = 0.617) and Mann–Whitney U test (p = 0.616), ACL21–33 tracks higher in all modern samples from individual sites where both were sampled, except Majuanwan (Fig. S4). ACL values have been shown to correlate with higher growing season temperature and aridity65,66,67, and therefore, the higher ACL values in modern plants are consistent with the nearly 1 °C increase in regional temperature over the past 50 years68 and the enhanced aridity that has resulted from intensive irrigation farming that began in the middle of the 20th Century in northwestern China37. Selective pressures may favor the production of longer n-alkane chain lengths under hot or arid conditions69,70, and the extant P. australis likely suffer from water stress brought on by significant evapotranspiration and elevated 21st Century temperatures, both of which drive ACL values higher. The uncertainty associated with using plant wax distribution, CPI, and ACL as environmental indicators, however, necessitates the application of stable isotope measurements to make stronger inferences on climatic and environmental change.

### Bulk carbon isotope analysis

Ancient reeds from Han (Sites 1–7), Jin (Sites 9 and 10), Tang (Sites 8, 11, 12), and Song (Site 14) samples yield bulk δ13C values between − 25.3‰ and − 22.6‰ (− 23.9 ± 0.7‰; n = 42), while modern bulk δ13C values, corrected for the Suess effect, range between − 24.6 and − 20.8‰ (− 22.9 ± 1.3‰; n = 12) (Fig. 7; Data S1).

There is a clear geographic patterning in δ13C signatures of modern P. australis samples. Modern samples exhibit significant differences in δ13C values between samples from eastern (n = 7) and western (n = 5) sites (Student’s t-test two-tailed, p = 0.001; Mann–Whitney U test, p = 0.003; see Table 2), as western samples are 2.0‰ higher on average than eastern P. australis, having mean corrected δ13C of − 21.7‰ and − 23.7‰, respectively (Figs. 7, 8a). An opposing pattern is observed in ancient Phragmites samples, however, as the mean Han Dynasty δ13C value from the east (n = 26) is 0.5‰ heavier than western (i.e., Jin, Tang, and Song; n = 16) samples (Figs. 7, 8b; Table 2). Interestingly, there is no significant difference between corrected modern and Han δ13C values in the eastern cluster (Table 2), as both have means of − 23.7‰. On the other hand, corrected modern samples from Xinjiang are 2.5‰ higher than their ancient analogs with modern P. australis having a mean δ13C value of − 21.7‰ compared to the Jin, Tang, and Song Dynasty sample mean δ13C values of − 24.2‰ (Table 2).

This 2.5‰ heavier δ13C value of modern P. australis in the western cluster and the uniformity in eastern modern and Han aged samples, suggest a differential rate of environmental change on opposite sides of Lop Nur. We attribute the higher δ13C values of western P. australis to 13C-enrichment that occurrs in plants growing in environments with higher temperatures and rates of evapotranspiration71,72. Plants in arid or hot environments are proportionally enriched in 13C compared to those growing in cooler or well-watered locations because the rate of water loss intensifies as plants must augment stomatal conductance to preserve leaf water71,72. Annual mean temperature and precipitation at Yuli City, which represents our modern western samples’ climate parameters, averages 12.1 °C and 37.2 mm, respectively. This is ~ 5 °C warmer and half the annual precipitation of that in Yumen City from east of Lop Nur (Fig. S1). Thus, the extensive aridity and higher evapotranspiration rate in Xinjiang has a significant fractionation effect on bulk carbon isotope values, resulting in the mean 2.5‰ difference between modern and ancient samples, as well as the mean 2.0‰ difference between modern samples from Xinjiang and Gansu. It is reasonable to infer that the rate of change over time has increased as bulk δ13C from modern P. australis collected at Yingpan City (Sites 9 and 10) is 2.7‰ heavier on average than Jin Dynasty Great Wall culms from the site, while modern samples from the Sishilidadun Beacon Tower of the Song Dynasty (Site 14) are on average 2.8‰ higher than ancient reeds from this location.

### Bulk nitrogen isotope analysis

Ancient reed samples yield large variations in bulk δ15N, with values ranging from 0.8 to 33.5‰ (9.3 ± 6.7‰; n = 42). Extremely heavy δ15N values were obtained in samples from the Tang Dynasty Milan Castle Heritage Site (Site 8, 27.5 ± 6.0‰; n = 3) and in the Song Sishilidadun Tower (Site 14, 15.9 ± 4.0‰; n = 3). In general, δ15N values in Han (6.9 ± 3.3‰; n = 26) and Jin (6.1 ± 3.2‰; n = 5) era samples are significantly different from Tang (16.3 ± 10.4‰; n = 8) and Song (15.9 ± 4.0‰; n = 3) aged samples (Table 3), due to the heavy δ15N values obtained from Milan Castle and Sishilidadun (Data S1). This may suggest that the δ15N composition of plant or soil nitrogen in Tang aged samples was influenced by temperature and precipitation at the large, regional scale73,74,75, but also by human activities at the small, site-specific scale. The extremely high δ15N values at Milan Castle and Sishilidadun Tower coincide with other evidence of high (i.e., ≥ 15‰) δ15N values from Xinjiang52,76, indicating that human activities such as crop fertilizing, in addition to climatic parameters, may have elevated δ15N values at these locations.

The culms used for the construction of the Milan Castle yield an exceptionally high mean δ15N value of 27.5 ± 6.0‰, which is about 12‰ heavier than that of Sishilidadun. The 15.9 ± 4.0‰ δ15N from ancient Sishilidadun stands in exceptional contrast to the modern 0.24 ± 0.04‰ value obtained from extant P. australis growing at the site. The only other modern δ15N value obtained, 13.4‰, comes from the Yingpan City Heritage Site (Site 10). As for other ancient samples, the Cang Ting Sui Beacon Tower (Site 5) and wall segments at the Great Wall Heritage Site (Site 6), locations which are only 5 km apart, have mean δ15N values of 8.3‰ and 2.0‰, respectively, showing the large variability and local variations between sites (Fig. 8c). Based on mean δ15N of all ancient Phragmites except those from Milan Castle and Sishilidadun (7.2 ± 3.6‰; n = 36), the local differences in δ15N more likely reflect the use of fertilizers around large population centers in the past, which would have resulted in potential agricultural runoff that drove δ15N values higher in wild reeds that were then harvested for wall building. Fertilizers derived from manures or guano are enriched in 15N and have a significant impact on plant nitrogen isotopic values73,77,78,79.

## Discussion

Although the rammed-earth and reed fascine segments of the ancient Great Wall do not elicit the amount of visual attraction as the brick and stone masonry of the Ming Dynasty, they offer a wealth of unique scientific information on the sourcing of natural organic building materials in addition to paleoclimatic and environmental signals. Our AMS results (Table 1) confirm unsynchronized ages for construction of Great Wall segments along the Shule River in Gansu Province and the beacon towers and associated fortifications along the Kongque River in Xinjiang. As our radiometric ages corroborate with archaeological findings11,12,13,14 (Supplementary Materials), we are confident that the ancient reeds analyzed here were in fact original Great Wall building material.

The different ages obtained from the Xinjiang beacon towers are somewhat surprising. On the one hand, the Tang aged material from Sites 11 and 12 confirm recent archaeological findings from a nearby tower50,51, whereas the younger age obtained from Sishilidadun (Site 14), one of the northern-most beacon towers along the Kongque River, suggests that wall-building activities lasted into the post-Tang era and may have been a defensive practice adopted by nearby states like the Qocho Uyghur Kingdom or imported from northeastern China when the Western Liao Dynasty was established in Central Asia. Nevertheless, our new radiometric dates shed light onto the previously conflicting archaeological age uncertainties for these ancient structures.

Both the pyrolysis and n-alkane data demonstrate the exceptional preservation of the ancient reeds used in Great Wall construction, despite the lower quantity of certain biomolecules in archaeological samples. Desiccation enhances preservation for plant material at both structural and molecular levels61,62, so it is expected that the dry climate of northwestern China helped facilitate the longevity of ancient reeds protected within wall ramparts. The excellent molecular preservation of these ancient reeds is confirmed with SEM observations (Fig. 3). Both external and internal views of ancient culm samples show identical cellular features, including epicuticular wax crystals, epidermis, internal membranes, vessel elements, parenchyma cells, fibers, etc., with their modern counterparts having only slight degradation. The preservation of these in situ organic material, therefore illustrates the potential of using these site-specific, common reeds as a proxy source for studying climatic and environmental change at a smaller, local scale. As fragments of common reeds are included in the construction mix of building materials in ancient, arid Central Asia, they will be essential to archaeological studies seeking to illuminate the influence of human activities on the ecology and distribution of oases in the eastern Tarim Basin since the Han period. Additionally, this technique can be adopted elsewhere and applied to any ancient ruin that preserves organic materials.

While northwestern China has become warmer and dryer since the Han Dynasty30,48,80, the large δ13C offset between modern P. australis and Jin, Tang, and Song Dynasty aged Phragmites indicates variability in the localized ecological response to changing climate and surface-water hydrology. We suggest that the uniformity in ancient δ13C values recorded in Great Wall Phragmites culms from across the eastern Tarim Basin (Fig. 8a) may reflect the once wider availability of regional oases, likely due to previous homogeneous wetter climatic conditions brought on by a stronger Asian monsoon that penetrated further into western China32,33,44,48. On the other hand, the uniformity in bulk δ13C between Han aged and modern P. australis samples collected in Gansu suggests that surface-water hydrology within the Shule River catchment has been relatively consistent over the last two millennia. This contrasts with the high values of bulk δ13C in modern samples collected from the western cluster that now suffers a greater degree of environmental stress due to elevated temperatures and a higher degree of aridity and evapotranspiration, suggesting that climatic conditions in Xinjiang have drastically changed over the past century.

The regional environmental change in China’s northwestern frontier is of explicit concern in the discussion of various episodes of migrations81,82,83 and cross-cultural exchanges of technologies, military, farming, and pastoral activities, with the eastern Tarim Basin acting as a crossroad in those narratives84,85. Both Han and Tang Dynasties were periods of significant expansion of the Chinese empire into the northwest, a phenomenon that historians largely attribute to a unified central power, economic growth, cultural achievements, and more importantly, strong military capabilities86. However, recent studies have pointed out the possible link between climate change and state affairs and regional conflicts, especially along China’s northwestern border between nomadic and farmer groups87. Our data provide additional evidence linking the availability of locally sourced material with wall building activities during the Han, Jin, and Tang periods, with conflict possibly incentivising the construction of walls or beacon towers87. Nevertheless, further research is needed to determine whether this only occurred when climate conditions sustained sizeable oases in the region, or other ways in which climate played a role in shaping historical change and development88,89,90.

The Tarim Basin’s Taklamakan Desert is characterized by extreme aridity and extensive evapotranspiration, which results in large plant 15N enrichment due to intensive evaporation and low mean annual precipitation91. Located at the center of the Eurasian continent, the Taklamakan Desert is the world's second largest shifting sand desert, with evaporation reaching as high as 1500 mm yearly and annual precipitation being only between 50 and 80 mm on the basin’s edges and 17–25 mm at the center68,92. As δ15N values of plant roots, plant litter, and soil organic matter increase with decreasing precipitation73,93, the entire Tarim Basin has some of the most elevated terrestrial δ15N values in Eurasia52,76,91, as exemplified by the δ15N value of 13.4‰ from modern P. australis sampled at Site 10.

The extremely high δ15N values at the Tang-era Milan Castle, however, suggest that fertilizers derived from manure or guano had a significant effect on wild plant nitrogen isotopic values73,77,78,79. The mean 27.5 ± 6.0‰ obtained from Milan Castle could only be reached through anthropogenic input, whether deliberately or as agricultural runoff. The high, 15–20‰ δ15N values of Yingpan Man and associated wheat/barley, broomcorn millet, grape, goats, and sheep52, all suggest that high nitrogen values correspond to human subsistence strategies, specifically when compared to our relatively low results from Yingpan City (Sites 9 and 10, 6.1 ± 3.2; n = 5), and from the site’s modern P. australis δ15N value of 13.4‰. The δ15N data from Yingpan City would therefore suggest that local Phragmites were not purposely fertilized here, nor did they grow in an area influenced by agricultural runoff. Though we are currently unable to determine whether Phragmites used in the construction of Milan Castle were purposely fertilized and deliberately managed for wall construction, or had naturally collected nutrients from agricultural runoff, the extremely high value here compared to sites like Yingpan City suggests that the possible manuring hypothesis deserves further investigation.

It has been suggested that widespread irrigation projects and agricultural intensification quickly decreased the amount of surface water along the eastern Tarim Basin, reducing overall lake levels44,80. However, the uniformity in our ancient Phragmites δ13C values indicates that surface-water availability, and the extent of oases, was more homogenous from the Han through Song Dynasties across both sides of Lop Nur (Fig. 8a), implying little change of surface-water hydrology due to human activities during these periods. Moreover, it may have only been relatively recently37,68 that elevated temperatures and aridity had a significant influence on plant δ13C in our study region, especially in the western cluster samples (Fig. 8a). Our data, therefore, is consistent with recent temperature records in Xinjiang indicating that the Tarim Basin experienced significant, monotonic warming with an average increase of nearly 1 °C from 1955 to 2000, unevenly distributed across time and space68, leading to the intense evaporative stress on modern plants.

Our work demonstrates the potential for paleoenvironmental reconstructions applying molecular and biochemical methodologies to organic materials that are well preserved in ancient Great Wall segments and beacon towers in the western frontier from important periods in Chinese history. Our study represents the first attempt to reconstruct the source and local habitats of Great Wall building materials that contain information on the impact of climate change on local environmental settings. Along with other regional and global climate proxies, this data illuminate site-specific environmental records that speak to localized natural or human-induced environmental changes in northwestern China. Building upon this study, more research based on higher resolution sampling strategies of co-occurring organic materials with other molecular isotope climate proxies from newly-surveyed beacon towers in Xinjiang and Inner Mongolia will certainly yield valuable insights into regional archaeology and environmental changes across time and space. Such methods are not limited to northwestern China, however, and can be applied to any ancient structure that was built, at least partially, using locally-sourced organic materials.

## Methods

### Site locations and sampling

Ancient Phragmites culms (n = 45), identified only to the genus level, were sampled from exposed fascines of remnant wall segments, beacon towers, or fortification ruins (Fig. 2). Modern culms (n = 8) and leaves (n = 4) belonging to the species P. australis were sampled directly from wild plants growing in the study area (Data S1). All available modern P. australis populations in the study area were sampled. Both ancient and modern samples were collected from 14 sites with the permission from local archaeology authorities in Gansu and Xinjiang during joint field expeditions in 2011 and 2016 (Fig. 1, Data S1). Geographically, these sites are grouped as eastern (Sites 1–7) and western clusters (Sites 8–12, 14; Site 13 did not yield data), separated by the now dried Lop Nur Lake basin (Fig. 1b). Climatologically, this region represents one of the driest areas in China with mean annual precipitation of only 66.5 mm at Yumen City (40°16′ N, 97°2′ E) and 37.2 mm at Yuli City (41°21′ N, 86°16′ E), localities representing the climate of the eastern and western side of the Lop Nur basin, respectively. There is also a regional mean annual temperature (MAT) difference, with MAT at Yumen City being 7.5 °C compared to 12.1 °C of Yuli City (Fig. S1). However, different paleoenvironmental proxies suggest wetter climate conditions with higher lake levels and precipitation in northwestern China during the Han and Tang Dynasties38,39,40,44,47,48,94,95,96,97,98. In contrast, lake records demonstrate decreased moisture availability and significant landscape change toward the end of the Han Dynasty and shortly afterwards31,44,98,99.

Scanning Electron Microscopy (SEM) was performed on culm pieces of both modern P. australis and an ancient Phragmites from the Sunji beacon tower (Site 12) that were cut with single-edged blades. The external and internal surfaces of these pieces were then coated with a 15 nm thin film of gold and observed using a JEOL JSM-IT200 Scanning Electron Microscope under low vacuum at 15kv.

We only collected samples from wild P. australis from Chinese public land guided by Chinese colleagues with permission granted by their respective research institutes in accordance with relevant guidelines and regulations. This species is not considered at Risk of Extinction and is listed as Least Concern by the IUCN Council, thus, no special permission or license was required for collection100. All collected specimens were identified by Dr. Qin Leng and voucher specimens (see sample ID numbers in Data S1) were equally divided between the Institute of Earth Environment at the Chinese Academy of Sciences in Xi’an, Shaanxi and at the Laboratory for Terrestrial Environments in Bryant University, Smithfield, Rhode Island.

### Accelerator mass spectrometry (AMS) 14C dating

The chronology of the different sampling locations was established through radiocarbon dating on selected plant samples using accelerator mass spectrometry at the Institutional Center for Shared Technologies and Facilities at the Institute of Earth Environment, Chinese Academy of Sciences in Xi’an, China. Dates were calibrated to cal yr before A.D. 1950 (i.e., cal 14C yr B.P.) by IntCal20 using the OxCal v. 4.2.3 software at 95% probability or 2 standard deviations (2σ) (Table 1). For sites where radiocarbon was not applied, the chronology of each location was determined using archaeological artifacts and historical texts11,12,13,14 (Supplementary Material). All available evidence indicates a Han Dynasty age for sampling sites in the eastern cluster (Sites 1–7). Bamboo slips and wooden tablets recovered from beacon towers at Yumenguan (Site 5, Fig. 1), Majuanwan (Site 7, Figs. 1, 2a,b), and Dunhuang11,13 date them from 98 BC to 137 AD101. Other archaeological artifacts, such as silk textiles, literature, and the construction style of the beacon towers, further attribute these military structures to the Han Dynasty and as key stops along the ancient Silk Road13,102. Two AMS dates obtained from Majuanwan yielded radiometric ages of 132 and 116 BC (± 20 years), confirming its Han Dynasty age (Tables 1, S1).

The ages for sampling sites in Xinjiang (Sites 8–12, 14) are grouped into three categories in terms of their chronology (see Supplementary Materials for detailed description):

1. 1.

Site 8 (Milan Castle Heritage Site) is radiometrically dated to ~ 770 AD and was constructed during the Tang Dynasty28. Sites 11 and 12, beacon towers north of the Kongque River, are also attributed to the Tang Dynasty based upon archaeological findings50,51. Our new radiometric dating of four Great Wall Phragmites from Sites 11 and 12 yielded calendar years between 677 and 726 AD, placing them within the Tang Dynasty.

2. 2.

The Buddha Tower at the Yingpan City Heritage Site (Site 9) was radiocarbon dated to 305 AD, placing it in the Jin Dynasty52. A wall section from the Yingpan City Heritage Site (Site 10) was previously assigned to both the Han and Jin Dynasty based upon burials and associated archaeology54,55. We attribute Sites 9 and 10 to the Jin Dynasty as Yingpan was likely built during the Han and continuously occupied through the Jin period.

3. 3.

The Sishilidadun Beacon Tower (Site 14) is the youngest of the sites studied, as we obtained two AMS ages with radiometric dates of 1030 and 1160 AD. These dates indicate that the structure was built during the Song Dynasty (960–1279 AD), although Song political borders are not known to have extended into the Tarim Basin.

Most of the ancient reed materials used in construction were culms, as leaves have rarely been recovered from these ancient ruins (Fig. 2b). Modern culms and leaves of native P. australis were also sampled at six of the sites that contained reed stands near the ancient ruins to serve as modern comparisons (Sites 1, 5, 7, 9, 10, 14). Morphologically, the culms of ancient reeds are indistinguishable from their modern counterparts (Fig. 3). All samples were kept frozen in the laboratory until analyzed.

### Pyrolysis gas chromatography mass spectrometry (Py-GC-MS)

Modern (n = 4) and ancient (n = 6) plant samples were analyzed using Py-GC-MS to test for the molecular distributions and preservation of organic compounds at Bryant University, Smithfield, Rhode Island. Samples were pyrolized using a CDS 5250 Pyroprobe by combusting at 610 °C for 20 s to convert macromolecular compounds to GC amenable products. Compound detection and identification were performed using an Agilent 7890A GC System equipped with a Thermo TR-1 capillary column (60 m length, 0.25 mm i.d. and 0.25 μm film) coupled to a 5975C Series Mass Selective Detector (MSD). The GC oven was programmed from 40 °C (5 min hold) to 100 °C at 10 °C/min, then to 300 °C at 6 °C/min (25 min hold). Helium was the carrier gas with a constant flow of 1.1 mL/min. The MS source was operated at 250 °C with 70 eV ionization energy in the electron ionization (EI) mode and the MS Quadrupole mass analyzer was set to 150 °C with a scan rate of m/z 50–500. Samples were held at the pyroprobe interface for at least 5 min at 300 °C for additional thermal extraction and to remove volatile impurities before gas chromatography. Compounds were identified by comparing their spectra with those reported in the literature103,104. Duplicate analyses of each sample were conducted for analytical consistency.

### Chemical analysis of plant wax lipids

Plant culms and leaves were lyophilized and ground, then extracted with Dichloromethane:Methanol (9:1, v/v) using ultrasonication at 40 °C in three, 30-min cycles at the Institute of Earth Environment, Chinese Academy of Sciences, Xi’an. The total lipid extracts were dried under nitrogen and separated into two fractions through silica gel column chromatography using hexane and methanol, respectively, with n-alkanes being eluted in the hexane fraction. Quantification of n-alkanes was performed using an Agilent 6890 Series instrument equipped with a split-injector, HP1-ms GC column (60 m length, 0.32 mm i.d. and 0.25 μm film), and a Flame Ionization Detector (FID). Samples were injected in split mode (split ratio 4:1) and the GC oven was programmed from 40 °C (1 min hold) to 150 °C at 10 °C/min, then to 315 °C at 6 °C/min (20 min hold). Helium was the carrier gas with a constant flow of 1.2 mL/min. Sample peak areas were compared with an external standard mixture (C21, C25–C33, odd numbered n-alkanes, 50 ng/μL) for compound identification and quantification. Specifically, sample n-alkane GC-FID peak areas (PA) were converted to concentrations assuming that the response of each compound is identical to that of the standard mixture (approximately 3–5 PA/ng). A calibration curve was not necessary as the range of n-alkanes measured were not outside the standard carbon homologue range.

Generally, long-chains (C27–C35 n-alkanes) are most abundant in terrestrial plants105,106, while submerged and floating aquatic macrophytes contain more mid-chain compounds (C23–C25 n-alkanes)107,108, and short-chains (C17–C21 n-alkanes) are dominant in algae109,110. The average chain length, or the weight-averaged number of carbon homologues of the odd-numbered C21–C33 n-alkanes, is used (cautiously) as both biosynthetic source and climate proxies and was calculated as follows:

$$ACL = \frac{{21\left( {C_{21} } \right) + 23\left( {C_{23} } \right) + \cdots + 33\left( {C_{33} } \right)}}{{C_{21} + C_{23} + \cdots + C_{33} }}$$

where Cx is the abundance of the chain length with x carbons. The carbon preference index (CPI) examines the odd-over-even carbon number predominance of hydrocarbons and distinguishes sedimentary organic matter deriving from terrestrial plants and that from bacterial or petroleum sources60,111,112,113, and as an indicator for hydrocarbon maturity58. Alkanes deriving from land plants display carbon chains typically (60.7%) with CPI > 5, but have been shown to range between 1 and 99, with most (81.2%) plants having values > 259. Samples containing petrogenic and marine inputs or mature/degraded samples are characterized by considerably lower CPI values of ≤ 1. CPI was calculated using the abundances of odd and even chain lengths from C21 to C33 with the following formula:

$$CPI = \frac{{\left( {C_{21} + C_{23} + \cdots + C_{31} } \right) + \left( {C_{23} + C_{25} + \cdots + C_{33} } \right)}}{{2 \times \left( {C_{22} + C_{24} + \cdots + C_{32} } \right)}}$$

### Bulk carbon isotope analysis

Culms from modern and ancient reeds were washed with distilled water and treated with 2 M HCl for 24 h at room temperature to remove any potential carbonates before combustion (4 h, 860 °C) in a vacuum-sealed quartz tube in the presence of Ag foil and CuO. The purified CO2 gas was then analyzed for carbon isotopes using a Finnigan MAT251 gas mass spectrometer at the Stable Isotope Biogeochemistry Laboratory at the Institute of Earth Environment, Chinese Academy of Sciences, Xi’an. The national standard GBW04407 (δ13CVPDB = − 22.43 ± 0.07‰) was analyzed between every twelve samples. The precision of repeated measurements of the laboratory standard was < 0.1‰. Sample carbon isotope ratios (δ13C) are expressed as parts per thousand (‰) relative to the international VPDB standard and defined by the following equation:

$$\delta {}^{13}C = \left( {\left( {\frac{{{}^{13}C}}{{{}^{12}C}}Sample \div \frac{{{}^{13}C}}{{{}^{12}C}}Standard} \right) - 1} \right) \times 1000$$

Since modern and ancient reed δ13C values were compared, + 1.9‰ was added to all modern values114,115,116 (Data S1). This is to correct for the 13C Suess effect, or the differences in atmospheric δ13CCO2 between the value averaged for 2011 and 2016 of − 8.4‰117, and the pre-industrial δ13CCO2 value of − 6.5‰118.

### Bulk nitrogen isotope analysis

The nitrogen isotope ratios of the dried plant samples were determined at the Stable Isotope Biogeochemistry Laboratory at the Institute of Earth Environment, Chinese Academy of Sciences using a FLASH EA1112 elemental analyzer interfaced with a Delta-Plus continuous-flow isotope ratio mass spectrometer (IRMS). All the δ15N values used a KNO3 reference material (δ15N 6.27‰) and an international isotope reference material (IAEA-N3; δ15N 4.70‰) to control the analytical accuracy of the EA-IRMS. Repeated analyses of the laboratory soil standards with confirmed δ15N values were performed daily to ensure instrumental accuracy. The standard deviation for repeated sample analyses was better than 0.3‰. The δ15N of each sample is expressed as ‰ relative to the international AIR standard and defined by the following equation:

$$\delta {}^{15}N = \left( {\left( {\frac{{{}^{15}N}}{{{}^{14}N}}Sample \div \frac{{{}^{15}N}}{{{}^{14}N}}Standard} \right) - 1} \right) \times 1000$$

Finally, two-tailed Student’s t-tests, assuming unequal variances, and Mann–Whitney U tests were used to test the significance in differences between sample sets using an Alpha (α) of 0.05 (see Tables 2, 3). Student’s t-tests and Mann–Whitney U tests were run using PAST 4.03 (additional nonparametric statistical tests, Kruskal–Wallis and Spearman’s rank, were also used to test significance in differences between sample sets and are presented in the Supplementary Materials). Mean values are reported along with their standard deviations.