Actually, our sediment samples were collected from the same section as Li (1995) and Fang et al. (2003) proposed. In order to get the fresh and uncontaminated sediments, an exploratory trench (0.5 m deep and wide) was dug. And then a total of 229 unweathered sediment samples were collected from the trench from the bottom to the top of the Maogou section, from the Tala Formation to the Hewangjia Formation, generally with the sampling interval of 2 m. Most of the samples were composed of silt sandstones, mudstones. All the sediment samples were marked with thickness and stored at -20 C for further analyses.

The samples (150 g each) were powdered (80 mesh) and extracted with dichloromethane/methanol (93:7) in a Soxhlet extractor for 70 h, and the solvent was removed by distillation. The extracts were condensed and weighed. Because the extracted material was only 1–5 mg, fractionation using chromatography on silica gel or alumina was avoided to prevent the potential loss of components such as the trace alkanes and the oxygen compound. After natural air-drying and dilution using chloroform, the total extracts were analyzed directly using GC-MS. The blank samples were run under similar conditions. The GC-MS spectra of blank samples showed that the lipid molecules to be discussed in this paper were not found.

GC-MS analysis was performed using a HP 5973 MSD interfaced to a HP 6890 gas chromatograph fitted with a 30 m × 0.25 mm i. d. fused silica capillary column coated with a film (0.25 μm) of 5% phenyl-methyl-DB-5. For routine GC analysis, the oven was programmed from 80 to 300°C at 3°C/min with an initial and final hold time of 1 and 30 min, respectively. Helium was used as carrier gas at a linear velocity of 32 cm/s, with the injector operating at a constant flow of 0.9 ml/min. The MS was operated with an ionisation energy of 70 eV, a source temperature of 230°C and an electron multiplier voltage of 1900 V over a range of 35–550 Da.

An aliquot of each sample was acidified with HCl to remove carbonates before analysis. Then, samples were washed with deionized water until a neutral pH value was reached. Next, samples were dried in an oven at 90°C. Pretreated samples were analyzed using a Flash 2000-MAT 253 system. The Flash 2000 was fitted with an oxidation-reduction tube filled with Cr₂O₃, Cu and silver-bearing CoO. Treated samples were oxided at 960°C with flowing oxygen. Helium was used as carrier gas. IAEA-600 (caffeine) was used as standard sample (Coplen et al., 2006). Each sample was analyzed twice, and final averaged results were expressed as ‰ relative to the VPDB (Vienna Peedee Belemnite) standard. During the analysis, 15 standard samples were determined to monitor the accuracy of the instrument. Reproducibility, as determined from the replicate standard samples, were better than 0.05‰ for δ¹³C_org of organic carbon.

Basically, the n-alkanes ranges from C₁₆ to C₃₂ with an obvious maximum carbon number at C₁₈ throughout the entire section, characterized by unimodal distribution patterns. Moreover, some samples show a significant bimodal distribution, with C₁₈ and C₂₉ as the main peaks. A distinct odd-over-even predominance of the long-chain n-alkanes is observed throughout the profile. Representative relative abundance of n-alkanes from the Maogou section sediments are shown in Figure 2.

In terms of the n-alkane derived proxies, the ratio of C₂₇/C₃₁ and (C₁₇-C₂₁)/(C₂₇-C₃₁) values were calculated in this study. Specifically, the ratio of (C₁₇-C₂₁)/(C₂₇-C₃₁) can be calculated using the following equations: ( C 17 - C 21 ) / ( C 27 - C 31 ) = ( C 17 + C 18 + C 19 + C 20 + C 21 ) / ( C 27 + C 28 + C 29 + C 30 + C 31 )

Besides, the δ¹³C_org values of the Maogou section sediments varies between -31.0‰ and -22.0‰, with an average value of -26.0‰ (Figure 3C).

Typically, n-alkanes are widely distributed in sedimentary organic matter derived from organisms living in the study area and their catchments, and different compositions of n-alkanes can directly reflect different biota (Pearson et al., 2007). Normally, the ratio of (C₁₇-C₂₁)/(C₂₇-C₃₁) of n-alkanes can indicate the relative contribution from lower organisms including algae, cyanbacteria, fungi and microbes relative to that from terrestrial higher plants, and thus the corresponding climate conditions (Eglinton and Hamilton, 1967; Cranwell et al., 1987; Rielley et al., 1991; Meyers and Ishiwatari, 1993; Ficken et al., 2000). Besides, the ratio of C₂₇/C₃₁ is usually applied to indicate the relative abundance of trees versus grasses, as trees produce high abundance of C₂₇ or C₂₉ n-alkane, while grasses typically generate high C₃₁ n-alkane. Consequently, a high C₂₇/C₃₁ ratio indicates a thrive of trees at the expense of grasses, and thus more contribution from trees to the organic matter (Cranwell et al., 1987).

In our study, abundant n-alkanes were detected in all the 229 samples from the Maogou section. Generally, the n-alkanes displayed bimodal or unimodal distribution patterns, suggesting that the organic matter were contributed by mixed sources, including lower organisms and terrestrial higher plants. Generally, the ratio of (C₁₇-C₂₁)/(C₂₇-C₃₁) and C₂₇/C₃₁ displayed a similar variation trend throughout the sequence (Figures 3A,B), indicating the variations of organic matter sources. From 29 to 21.8 Ma, the ratio of (C₁₇-C₂₁)/(C₂₇-C₃₁) and C₂₇/C₃₁ remained relatively low-value stage except some fluctuations, suggesting a high proportion of grasses contributed to the organic matter. During the period of 21.8–18 Ma, both the (C₁₇-C₂₁)/(C₂₇-C₃₁) and C₂₇/C₃₁ values increased, implying the source of organic matter changed and lower organisms and trees flourished during this interval. From 18 to 14 Ma, the lipid-derived proxies showed a decreasing trend, indicating the sources of trees and lower organisms were limited at this time. Between 14 and 4.3 Ma, our records indicated that the contribution from lower organisms constantly remained constrained, and the abundance of trees slightly increased. According to these observations, we may infer that different organic matter sources at different timescales corresponded to the varying climate conditions.

Terrestrial high plants can normally be divided into three categories, C₃ plants, C₄ plants and CAM plants based on their photosynthesis characteristics and carbon atomicity. C₃ and C₄ plants can be identified according to their carbon isotope values, as C₃ plants are characterized by an δ¹³C value ranging from -40‰ to -20‰ with an average of -27‰, and C₄ plants have a distinct δ¹³C value between -19‰ and -9‰ with an average value of -12‰, respectively (Smith and Epstein, 1971; Oleary, 1981). It is also generally acknowledged that C₄ plants adapt to an environment which is typical of high temperature, high aridity and low atmospheric CO₂ concentration (Wang and Greenberg, 2007). Particularly, C₄ plants have greater water-use efficiency than C₃ plants (Collatz et al., 1998). In the present study, there is no exact relationship between the δ¹³C_org values from the Maogou sediments and the reconstructed atmospheric CO₂ concentrations based on the carbon isotopic analyses of diunsaturated alkenones and planktonic foraminifer (Pagani et al., 1999), suggesting the concentration of CO₂ may not be the main factor influencing the vegetation in the Linxia Basin (Figure 3). Additionally, previous study indicated that terrestrial higher plants contributed significantly to the organic matter in the sediments in the Linxia Basin (Fan et al., 2007). Consequently, the δ¹³C_org values may reflect the information of the vegetation communities and its associated climate conditions, such as temperature and/or precipitation (Wang and Deng, 2005; Kohn, 2010; Wu et al., 2018). Overall, the δ¹³C_org values of the sediments throughout the ∼29 to 4.3 Ma varied from -31.0‰ to -22.0‰ (Figure 3C), indicating that C₄ grasses were possibly not significant in the Linxia region prior to 4.3 Ma, which is consistent with the conclusion proposed by Fan et al. (2007).

In terms of this background, we attributed the obvious fluctuations of δ¹³C_org values to the climate variations somewhat arbitrarily, with positive δ¹³C_org values corresponding to a relatively high abundance of C₄ plants. In the present study, the records of δ¹³C_org values seemed more sensitive to climate variations as it revealed more stages of climate shifts. From 29 to 21.8 Ma, the δ¹³C_org values exhibited an overall increasing trend, indicating a relatively high proportion of C₄ plants. Coincidently, the C₂₇/C₃₁ ratio displayed a decreasing trend, suggesting an expansion of grasses. As previous study has confirmed that the grasses usually applied the C₄ photosynthesis (Wang and Greenberg, 2007), our results indicated that the vegetation evolution from trees to grasses reflect the vegetation type variations from C₃ to C₄ plants under the relatively warm and dry climate condition (Ao et al., 2021). The pollen assemblages also indicated an open woodland-steppe vegetation mainly composed of Chenopodipollis, Polygonaceae, Compositae for the herbs and Quercoidites, Betulaepollenites, Fraxinoipollenites and Cupressaceae for the woody plants under a dry condition during this time (Ma et al., 1998). While between 21.8 and 18 Ma, the C₄ grasses shrank while the C₃ trees flourished in this interval, implying a gradual increase in humidity, as indicated by the decreasing δ¹³C_org values and high C₂₇/C₃₁ ratios, corresponding to a general coniferous forest (Ma et al., 1998). During this interval, the high (C₁₇-C₂₁)/(C₂₇-C₃₁) ratio possibly indicated an increasing contribution from lower organisms to the organic matter. Normally, the δ¹³C_org of the typical lake algae in fresh water is about −28‰ (Meyers, 1994; Meyers, 2003), thus enhanced inputs of lower organisms may result in negative δ¹³C_org values in our records. In addition, the sedimentary face was identified as floodplain-shallow lake (Fang et al., 2016), furthering indicating an increasing humidity. Thereafter, a sharp increase in δ¹³C_org values demonstrated that C₄ grasses increased at the expense of C₃ trees from 18 to 14 Ma, indicating the climate was turning to warm and dry conditions gradually, possibly corresponding to the MMCO (Figure 3D) (Zachos et al., 2001; Zan et al., 2015). During this interval, the C₂₇/C₃₁ and (C₁₇-C₂₁)/(C₂₇-C₃₁) ratio decreased, suggesting an enhanced contribution from terrestrial higher plants such as grasses and limited inputs of lower organisms, which was in good accordance with the increasing δ¹³C_org values. The warm and dry climate conditions was beneficial to the growth of grasses. Although this period was recognized as coniferous forest, the percentage of herbs was increasing (Ma et al., 1998). Specifically, the initial stage of the MMCO was characterized by a relatively humid condition as reflected by the proxies, which was also recorded by the pollen assemblages from the northern Tian Shan (Tang et al., 2011) and Linxia Basin (Ma et al., 1998). After the MMCO, the climate was generally dry from 14 to 9.1 Ma, although the humidity showed a slight increase, generally consistent with the vegetation type as coniferous and broad-leaved mixed forest. A rapid increase in δ¹³C_org values and a sharp decrease in C₂₇/C₃₁ and (C₁₇-C₂₁)/(C₂₇-C₃₁) ratio within only less than 1 Ma from 9.1 to 8.5 Ma were observed, suggesting a significantly strengthened aridity at this time under the background of long-term global cooling in the late Cenozoic (Zachos et al., 2001; Westerhold et al., 2020), in accordance with the steppe forest expansion at ∼8.5 Ma (Ma et al., 1998). This aridity sustained to ∼6.4 Ma and then the humidity displayed a slight increase afterwards. Coincidently, the vegetation evolved towards coniferous and broad-leaved mixed forest, implying the climate became relatively wet (Ma et al., 1998). Considering the refined age-depth model and sample resolution, our records generally agreed with the pollen analysis.

It also should be pointed out that throughout the sequence, the variations of δ¹³C_org values corresponded well with the changes of lithology basically. This interesting phenomenon was primarily observed in the Dongxiang Formation sediments. The high δ¹³C_org values occurred when the gray-green clay deposited, and the red bed was characterized by low δ¹³C_org values. Further studies are necessary to figure out the potential driving mechanisms regarding the frequent fluctuation of the δ¹³C_org values.

As mentioned above, our records based on the n-alkane-derived proxies and δ¹³C_org values are overall in good accordance with the pollen records in the Linxia Basin (Ma et al., 1998). Furthermore, the sporopollen records from the Tianshui Basin in the NE Tibetan Plateau demonstrated a temperate, warm-temperate broad-leaved forest between 17.1 and 14.7 Ma, forest or forest-steppe between 14.7 and 11.7 Ma, broad-leaved forest during 11.7–8.5 Ma and steppe vegetation from 8.5 to 6.1 Ma (Hui et al., 2011). Similarly, the sporopollen reports in the Jiuxi Basin revealed a semi-moist climate from 13 to 11.2 Ma, warm and moist climate between 11.2 and 8.6 Ma, warm and semi-moist climate during 8.6–5.6 Ma, semi-arid climate during 5.4–4.9 Ma and arid climate condition in the interval of 4.9–2.2 Ma (Ma et al., 2005). Also, Tang et al. (2011) reported that the climate was wet during the late Oligocene and shifted to dry conditions between 23.8–17.3 Ma, and subsequently ameliorated to a relatively wet stage to 16.2 Ma, but then turned to dry condition until 4.2 Ma and the aridity reached a peak at 13.5 Ma. Despite the lack of long-term comparison of paleoclimate records, our results are in good concert with the periodical reports from previous studies.

Moreover, our reconstructed paleoclimate conditions based on the n-alkane-derived proxies and the δ¹³C_org values exhibited significant variations especially during the typical climate events, which behaved synchronously with previous studies. During the period of ∼26 to 25 Ma, our results suggested the grasses expanded, and contributions from trees and lower organisms declined, roughly in association with the late Oligocene Warming (Zachos et al., 2001). Sporopollen records from the Nima Basin indicated that broad-leaved trees increased obviously after 25.6 Ma, in response to the late Oligocene warming, further suggesting the development of south Asian monsoon and accompany of global warming (Wu et al., 2019). At ∼23 Ma, the increase of the (C₁₇-C₂₁)/(C₂₇-C₃₁) and C₂₇/C₃₁ ratio implied that trees and lower organisms characterized by negative the δ¹³C_org values increased, leading to the decrease of δ¹³C_org values from the sediments. This most apparent explanation for the change of organic matter source at ∼23 Ma is the Mi-1 Glaciation as revealed by the compiled oxygen isotope records (Zachos et al., 2001). Palynological results from the Xining Basin showed the cold-tolerant conifers shrived while thermophilic plants declined during the Mi-1 Glaciation at ∼23 Ma (Miao et al., 2013a). While during the interval of ∼18 to 14 Ma, the (C₁₇-C₂₁)/(C₂₇-C₃₁) ratio and C₂₇/C₃₁ value were decreasing, while the δ¹³C_org values exhibited an opposite variation, which suggested that C₄ grasses gradually flourished when the warm and wet climate shifted to warm and dry conditions during the MMCO (Zachos et al., 2001; Westerhold et al., 2020).

The pollen records from both the Tianshui Basin and northern Tian Shan suggested a warm and humid early stage of MMCO turned to warm and dry conditions in the latter periods (Hui et al., 2011; Tang et al., 2011). Moreover, the timing and duration of the MMCO was reported from nearby records as 18–14 Ma in the Qaidam Basin (Miao et al., 2011; Miao et al., 2016) and 17–14 Ma in the Xining Basin (Zan et al., 2015).

Our records demonstrated another significant climate shift at ∼9 to 8 Ma. During this period, the (C₁₇-C₂₁)/(C₂₇-C₃₁) and C₂₇/C₃₁ value decreased dramatically, and the δ¹³C_org values of sediments showed a sudden increase within less than 1 Ma, suggesting a prominent plant variation from C₃ to C₄ plants, which was ascribed to the enhanced aridification. Our previous work on n-alkan-2-ones and microbial communities from the Maogou section sediments also identified this severe climatic variation (Wang et al., 2012; He et al., 2020). Besides, the pollen assemblages also indicated an obvious aridification event happened at ∼8.5 Ma (Ma et al., 1998). The sporopollen data from the Tianshui Basin recorded a rapid development of steppe at about 8.5 Ma, which suggested a permanent drying of the Asian interior at this time (Hui et al., 2011). Yang et al. (2016) reported that carbonate-derived Sr concentrations and Sr/Ca ratios on the northeastern Tibetan Plateau exhibited weakened chemical weathering intensity and pedogenesis at ∼8.6 Ma corresponding to a cold and dry climate linked with the enhanced aridification. Furthermore, stable isotope evidence including the δ¹³C and δ¹⁸O values of carbonates and δ¹³C_org values from the Linxia Basin indicated that the most severe aridity occurred from 9.6 to 8.5 Ma (Fan et al., 2007).

In summary, based on these observations, the reconstructed paleoclimate conditions in the Linxia Basin generally followed the global climate variation trend. Specifically, our records indicated an overall warm and dry period of late Oligocene warming and MMCO, which was in favor of the growth of C₄ grasses, corresponding well with the global climate as revealed by the oxygen isotope records (Zachos et al., 2001; Westerhold et al., 2020). And the dry climate conditions spanning 14–4.3 Ma generally coincided with the global cooling since the expansion of polar ice-sheets at 14 Ma (Zachos et al., 2001). Under this circumstance, the expansion of C₄ grasses at 9–8 Ma cannot be sufficiently explained by the global climate trends. Thus, other possible factors need to be considered. The uplift of the Tibetan Plateau may be responsible for the aridification, as extensive evidence corroborated that the tectonic uplift of the Tibetan Plateau occurred approximately at ∼8 Ma (An et al., 2001; Lease et al., 2007; Miao et al., 2012; Yang et al., 2016). The uplift of the Tibetan Plateau resulted in a geographic barrier which blocked moisture transportation from neighboring oceans, leading to an arid continental interior due to the rain shadow effect (Boos and Kuang, 2010), and therefore an enhanced aridification. Moreover, Miao et al. (2012) proposed that the uplift of the TP resulted in the changes of topography, which strongly influenced the moisture pattern in Central Asia during Miocene times, particularly the precipitation rates in the Linxia Basin were significantly affected by the tectonic uplift of the surrounding mountains.