The TGR was constructed on the lower end of the upper reaches of the Yangtze River. It covers about 5.8 × 10⁴ km², straddles the Daba Mountains and southern Sichuan Plateau, and covers the low river gorges as well as the parallel valleys of eastern Sichuan. The TGR has a conventional maximum storage level of 175 m and a minimum storage level of 145 m (Chen et al., 2022; Ding et al., 2022). The SNR and DNR are typical tributaries of the TGR and are, respectively, ca. 75 km and 123 km from the Three Gorges Dam (Figure 1).

Ten sampling sites were set along the SNR and DNR to investigate the variability in WEOM chemistry (Figure 1). Five representative sampling sites were located along the SNR (SN01–SN05) and DNR (DN01–DN05) each, from upstream to downstream (<30 km). The sampling was conducted on 31 May 2018 in a period of water-level decline. We collected ca. 0–2 cm depth sediments as surface sediments. All samples were transported within 48 h to the lab in an icebox for WEOM extraction.

WEOM was extracted according to Hur et al. (2014). Specifically, a 2-mm filter was applied to air-dried sediment samples. Then, a 1 g sediment sample was mixed with 100 mL water (Mili-Q). The ratio of sediment–water followed Wang et al. (2021c), which was determined by a series of sediment/water ratio gradient extraction experiments. The sediment–water mixture was shaken (200 rpm) by a rotary shaker in the dark. The supernatant was collected and passed through 0.22 μm membrane filters (Millipore Express) to obtain WEOM for further analysis.

The minerals in the sediments were characterized based on X-ray diffraction (XRD) analysis with Cu-Kα radiation (40 Kv, 100 mA). The scanning ranges varied from 3° to 70°. The analysis of minerals was carried out on Jade 9 with ICDD. The particle size of sediments was determined by a laser particle size analyzer (Horiba LA-920) after ultrasonic dispersion (Yuan et al., 2020).

The OC content (OC%) in the sediments were measured by a Thermo Scientific FLASH2000 Series CNS Elemental Analyzer (He et al., 2020). Before OC analysis, 2 M HCl was used to treat air-dried sediments (∼0.2 g) to remove carbonate. Acid-treated samples were then rinsed with Mili-Q water to pH = 7. Finally, an isotope ratio mass spectrometer (Thermal Scientific) was applied to determine the table carbon isotope (δ¹³C) of sedimentary organic matter.

Nutrient concentration including PO₄ ^3-, SiO₃ ^2-, and DIN (e.g., NO₃-N, NO₂-N, NH₄-N) was measured by a segmented flow colorimetric auto-analyzer (Bran + Luebbe). Dissolved organic carbon (DOC) concentration of 0.22 μm filtered WEOM was determined by a TOC analyzer (Shimadzu TOC-L) with University of Miami deep-sea standards. 3D-EEMs and UV-Vis were measured on an Aqualog absorption-fluorescence spectrometer (Horiba) according to Wang et al. (2021a). In 3D-EEMs measurement, a 1 cm path length quartz cuvette was hired. We used Mili-Q water as blank and conducted the scan over 240–650 nm with increments of 3 nm and a scan integration time of 1s. In sample measurement, the blank was subtracted, and the inner filter effects was removed. Several optical parameters applied in various studies were also introduced in this work. We calculated specific ultraviolet absorbance at 254 nm (SUVA₂₅₄) for the aromaticity degree of CDOM (Weishaar et al., 2003), biological index (BIX) for autotrophic productivity (Huguet et al., 2009), freshness index (FrI) for the proportion of recently produced or fresh DOM (Parlanti et al., 2000), and humification index (HIX) for the humification degree of FDOM (Ohno, 2002). For detailed analysis of EEMs, a DOMFluor toolbox-based PARAFAC model was introduced (Stedmon and Bro, 2008).

WEOM was extracted for molecular analysis by solid-phase extraction (SPE) with PPL cartridges (Agilent Bond Elut). FT-ICR MS was introduced to WEOM analysis to reflect the molecular properties. The analysis was conducted with a Bruker SolariX FT-ICR MS (15.0 T) with an electrospray ionization (ESI) source (Research Center for Eco-Environmental Sciences, Chinese Academy of Sciences) (Tang et al., 2021). Molecular formulae, ranging from 120 to 1,000 Da, were assigned by the specific element limitation (C_≤80, H_≤200, O_≤40, N_≤3, and S_≤2) with a C/N threshold of >4. The element content, ratios, modified aromaticity index (AI_mod), and double bond equivalent (DBE) were obtained by weighted average calculation (Wang et al., 2019). Molecule types (e.g., peptides, H/C: 1.5–2.0, N > 0; unsaturated aliphatic compounds: UA, H/C: 1.5–2.0, N = 0; polycyclic aromatics: PCAs, AI_mod > 0.66; polyphenols: polyp, AI_mod: 0.50–0.66; highly unsaturated compounds: HU, AI_mod < 0.50, H/C < 1.5) were categorized following previous studies (Martínez-Pérez et al., 2017). The molecular lability index (MLB_L) was also calculated in WEOM molecular analysis (D'Andrilli et al., 2015).

To evaluate the regime of optical–molecular linkage in WEOM, the Spearman’s correlation between the molecular peaks of FT-ICR MS and spectral information was assessed (Stubbins et al., 2014). The optical and FT-ICR MS data of DOM in water of TGR was obtained from He et al. (2020), and a comparison of optical-molecular linkage between DOM in sediment and water of TGR was conducted. A principal component analysis (PCA) was introduced for WEOM dynamic clarification.

Median particle size (MPS) ranged from 4.1–18.7 μm and 5.0–17.6 μm in the SNR and DNR sediment, respectively (Supplementary Figure S1). MPS values declined from tributary upstream to downstream in both SNR and DNR. In terms of mineral composition, various minerals, including quartz, illite, feldspar, calcite, dolomite, and chlorite, were identified. The proportion of quartz, feldspar, and dolomite declined from tributary upstream to downstream in SNR and DNR, while the proportion of illite, calcite, and chlorite varied inversely (Figure 2).

OC% values ranged 0.6%–2.2% and 0.6%–0.8% in SNR and DNR, respectively (Supplementary Figures S2A, C). OC% values declined from tributary upstream to downstream in SNR and DNR (Supplementary Figure S2A, C). δ¹³C values varied from −28.3‰ to −26.1‰ in SNR and −27.3‰ to −26.1‰ in DNR and demonstrated an increasing trend from tributary upstream to downstream (Supplementary Figures S2B, D).

There was no significant spatial variation of nutrient concentration (PO₄ ^3-, SiO₃ ^2-, and DIN) in both SNR (PO₄ ^3-:1.13–4.54 µM; SiO₃ ^2-: 9.07–33.30 µM; DIN: 28.09–93.58 µM) and DNR (PO₄ ^3-:1.08–2.48 µM; SiO₃ ^2-: 12.78–20.14 µM; DIN: 46.72–89.27 µM) (Supplementary Table S1). DOC concentration ranged 197.50–235.83 µM and 174.17–224.17 µM in SNR and DNR, respectively (Supplementary Table S1). To assess the proportion of DOC in sediment, it was isolated in the unit of mg C/g sediment: DOC in SNR and DNR varied 0.24–0.28 mg C/g sediment and 0.21–0.27 mg C/g sediment (Supplementary Table S1), respectively, which were higher than that in surface sediments of the urban river (averaged 0.10 ± 0.02 mg C/g sediment) (Zhang et al., 2021).

SUVA₂₅₄, relating to the aromaticity of CDOM in WEOM, ranged 4.89–5.36 L mg-C⁻¹ m⁻¹ in SNR and 4.17–5.21 L mg-C⁻¹ m⁻¹ in DNR (Supplementary Table S2). HIX, representing the humification of FDOM in WEOM, varied 0.55–0.81 in SNR and 0.47–0.56 in DNR (Supplementary Table S2). Both SUVA₂₅₄ and HIX decreased upstream to downstream in SNR and DNR (Figure 3, Supplementary Figure S3). On the other hand, there was a downstream toward increase in BIX (ranging 0.47–0.52 in SNR and 0.47–0.57 in DNR) and FrI (ranged 0.47–0.52 in SNR and 0.47–0.56 in DNR), which hints the autotrophic productivity in both tributaries (Figure 3). The semi-quantitative EEMs-PARAFAC further supported the variation of light-absorbing components in WEOM. Specifically, four fluorescent components were identified: humic-like components (C1, C2, and C3) and protein-like component (C4) (Supplementary Figure S4). Compared to downstream, there was a higher relative proportion of C1, C2, and C3 upstream, while the relative proportion of C4 varied inversely (Figure 3).

A total of 9,018 and 10,196 unique compounds were characterized by FT-ICR MS in SNR and DNR, respectively; the number of identified formulae in each sample varied between 5,749 and 6,884: an average of 2,378 ± 54 CHO compounds, 2,494 ± 61 CHON compounds, 1,396 ± 91 CHOS compounds, and 89 ± 10 CHONS compounds in SNR and 2,560 ± 15 CHO compounds, 2,585 ± 95 CHON compounds, 1,443 ± 78 CHOS compounds, and 103 ± 8 CHONS compounds in DNR were identified (Supplementary Table S3). CHO compounds were present in relatively higher abundance than N- and S-containing compounds. CHO, CHON, CHOS, and CHONS compounds exhibited homogeneity in relative abundance in upstream and downstream samples. AI_mod, DBE, and m/z values declined from the tributaries upstream to downstream, whereas MLB_L values increased (Figure 4). In addition, PCAs, polyp, and HU abundance varied 77.5%–83.1% in SNR and 79.5%–83.7% in DNR and declined in the tributaries upstream to downstream (Figure 4). Peptides and aliphatic compounds abundances varied 13.9%–15.0% in SNR and 13.4%–17.4% in DNR and increased from the tributaries upstream to downstream (Figure 4).

Optical parameters including FrI, BIX, and SUVA₂₅₄ exhibited significant correlations with AI_mod, DBE, MLB_L, and m/z (r > 0.5 or < −0.5), whereas no significant correlation was observed between HIX with molecular information (Figure 5A), preliminarily indicating that there was a heterogeneity linkage between the optical and molecular properties of DOM in sediment. Specifically, the significant positive correlation between SUVA₂₅₄ and AI_mod, DBE, and m/z demonstrated that CDOM compounds with high aromaticity tracked closely with high aromatic and unsaturated molecules with high molecular weight (Weishaar et al., 2003; Huguet et al., 2009), which might be derived from terrestrial sources (Seidel et al., 2015; Painter et al., 2018). The negative linkage among FrI, BIX and AI_mod, DBE, m/z indicated that the autotrophic activity derived CDOM pool is probably linked to low aromaticity, unsaturated molecules (Wagner et al., 2015) (Figure 5A), which could be involved in microbial activities (Zhou et al., 2019; Messetta et al., 2020). Humic material has been proven to consist of various macromolecules derived from the assemblage of individual, small molecules by the linkage of dispersive forces (Simpson et al., 2002; Romera-Castillo et al., 2014), which might result in the ambiguous linkage between HIX, the parameter relating to the humification degree of CDOM, and molecular information (Figure 5A). In terms of specific compounds, humic-like and protein-like components correlate significantly to terrestrially sourced PCAs, polyp, HU (Seidel et al., 2015), and autochthonously sourced peptides and aliphatic compounds (Kellerman et al., 2018), respectively, exhibiting the equal effectiveness of CDOM components and molecular groups in tracking DOM sources. Therefore, the covariance of the optical and molecular properties of DOM in sediment could be influenced by biogeochemical processes (e.g., microbial activity) and sources.

In previous studies, the combination of optical characteristics with FT-ICR MS has greatly broadened our understanding of the complex behaviors of DOM in water (Herzsprung et al., 2012; Stubbins et al., 2014; Timko et al., 2014; Kellerman et al., 2015; Wagner et al., 2015). To comprehensively assess the characteristics of DOM in sediment, we compared the optical–molecular linkage between DOM in sediment and in water (He et al., 2020) (Figures 5, 6), further demonstrating the unique association of the optical and molecular properties of DOM in sediments (Figures 5, 6). Compared to DOM in water, which exhibited homogeneously significant correlation among all optical and molecular parameters, the optical and molecular property of DOM in sediment presented limited association (Figure 5). Especially, HIX and SUVA₂₅₄, the parameters represent humidification degree and aromaticity of CDOM, respectively, and always tracked the similar molecule pools in water DOM (Williams et al., 2010; Dixon et al., 2014; He et al., 2020), showed significant different associated behaviors in sediment DOM. This was also consistent with the observation of Singh et al. (2014), indicating that HIX might be less effective in tracing aromatic compound in sediment DOM than in water DOM for structured matrices of sediment that influence the function of DOM (Zsolnay, 2003), or that there was limited overlap between humified and aromatic molecules of DOM in sediment. Moreover, to provide finer insights into the difference between optical–molecular linkage in sediment and water DOM, an optical component (e.g., humic-like and protein-like components)-related van Krevelen Diagram with marginal density plots is presented (Figure 6). A highly divergent distribution of CDOM associated molecules from DOM in sediment and water was observed, indicating the intrinsic difference between DOM in sediment and water. Specifically, less molecules correlated with CDOM components of DOM in sediment than that in water were observed, in consistent with the correlation variation of optical and molecular parameters (Figure 5). Molecules positively correlated with humic-like CDOM exhibited relatively higher O/C and lower H/C than those positively correlated with protein-like CDOM in both DOM in sediment and in water, demonstrating that, although sources or cycling behaviors varied, there was still universal molecular information in natural DOM (Zark and Dittmar, 2018).

With the bulk, optical, and molecular properties of WEOM characterized, the variation of WEOM in tributaries was revealed. In the SNR and DNR, the proportion of terrestrially derived WEOM (e.g., PCAs, HU, and polyp compounds) decreased significantly along the tributaries upstream to downstream transects, while autochthonous derived WEOM (e.g., UA and peptide) increased (p < .01) (Figures 3, 4) (Seidel et al., 2015). Meanwhile, δ¹³C, SUVA₂₅₄, HIX, AI_mod, m/z, and DBE exhibited higher values upstream than downstream, indicating that WEOM upstream tends to be more aromaticity with a higher molecular weight and unsaturation degree than that downstream (Figures 3, 4) (Hansen et al., 2016). PCA provided the preliminary insights into the variation dynamic of WEOM, which exhibited that the higher terrestrial signaled upstream samples had positive score values of PC1 and appeared around MPS, quartz, feldspar, and dolomite, while the higher autochthonous signaled downstream or tributary estuary samples had negative score values on PC1 and appeared around calcite, illite, and chlorite (Supplementary Figure S5). The analysis indicated that sediment property (e.g., mineral composition and particle size) was involved severely in the dynamic of WEOM and hinted that sediment property influenced the dynamics of WEOM.

In particular, the sediment grain size could be sorted by hydrological conditions, resulting in relatively larger grain sizes of sediment particles upstream and smaller downstream. Grain size could influence the structure of the pore system in particles and the relative number of reactive surface groups per unit mass of particle by spatial and diffusional confinement (Vereecken et al., 1989); this might provide relatively more effective binding points for H-enriched compounds (e.g., autochthonously derived OM) in small particles and enhance the carrying of relatively unsaturated compounds (e.g., terrestrially derived OM) with O atoms or OH groups in large particles. This was consistent with the positive relationship between HIX, SUVA₂₅₄, AI, and MPS and the negative relationship between BIX, FrI, and MPS (Supplementary Figure S6). In terms of mineral composition, a complex dynamic of OM molecules on the mineral–OM interface was revealed, which might be constrained by various factors, including the surface chemistry of minerals and OM property (Leinemann et al., 2018). This multiple-factor influenced and complex combination of minerals and OM would lead to the preferred binding of mineral and OM molecules (Kalbitz et al., 2005). Specifically, the higher abundance of quartz, feldspar, and dolomite upstream might prefer the adsorption of OM molecules with high aromaticity (Kaiser and Guggenberger, 2000; Kalbitz et al., 2000). Moreover, the environment context might influence the interfacial energy of quartz, feldspar, and dolomite, enabling recalcitrant molecules to coexist. This study indicated significant covariance among quartz, feldspar, or dolomite and individual molecules with high molecular weight and aromaticity (Supplementary Figure S6). PCAs, HU, and polyp compounds were enriched with quartz, feldspar, or dolomite content (Supplementary Figure S6). Calcite, illite, and chlorite have a relatively high specific surface area (SSA) (Cui et al., 2017; Wu et al., 2020), which might favor the accumulation of high saturated organic compounds. In addition to the surface topography, the crystal structures of illite and chlorite (e.g., the sheet of edge-sharing octahedra and corner-sharing tetrahedra) would limit crystal growth and maintain particle size (Chen et al., 2018) and, consequently, might also facilitate H-enriched OM molecule accumulation. This is also consistent with the close covariance between individual molecules with low molecular weight and high H/C (e.g., peptides and unsaturated aliphatics) and calcite, illite, and chlorite in this study (Supplementary Figure S6). Based on this, a sediment property-dependent conceptual model of WEOM dynamics was developed (Figure 7). In this model, tributary sediment regulates the dynamic of WEOM, which leads to higher terrestrial WEOM accumulation upstream and higher autochthonous WEOM accumulation downstream by regulating the distribution of a mineral matrix and sediment particle size (Figure 7).

Reservoirs are hot spots of carbon cycling with autochthonous and allochthonous carbon mixing, and they accommodate various carbon-related biogeochemical processes (Dean and Gorham, 1998; Battin et al., 2009; Tranvik et al., 2009; Li et al., 2019). With the quality assessment of WEOM, an essential form of organic carbon buried, the carbon dynamic information in reservoir would be hinted. Bio-incubation experiments of OM in the TGR (He et al., 2020) have demonstrated the positive relationship between the aromaticity of OM molecules and bio-degradation resistance (Wang et al., 2021c) (Supplementary Figure S7). This indicates that high terrestrially aromatic WEOM in the tributary upstream would be at the high end of bio-degradation resistance. Furthermore, the high autochthonous WEOM, which was less aromatic and more labile, accelerated the downstream tributary, which has a much greater water depth (>50 m) with lower oxygen content and water temperature than those upstream (Huang et al., 2019). The low oxygen content and water temperature would limit heterotrophic microbial activity and might constrain the biodegradation of autochthonous WEOM downstream (Bond-Lamberty and Thomson, 2010; Davidson et al., 2012). Thus, reservoir operation inducing WEOM dynamics is likely to be beneficial to OM preservation both upstream and downstream and devotes to carbon burial in tributaries. However, assessing the quantitative contribution of WEOM in tributaries to carbon burial in the whole reservoir requires further research.

This study has some limitations. For instance, considering the limited efficiency of water extraction, only some minerals (e.g., calcite, illite, and chlorite) which could fix OM molecules by adsorption were analyzed; the analysis of other minerals (e.g., iron-bearing minerals) which also fix OM molecules in a complex mode needed to be integrated into a more efficient extraction (e.g., base extraction) to delineate the dynamic of OM in more detail. Nevertheless, a novel insight into carbon sequestration and burial in the TGR was presented in this investigation, contributing to better clarification of carbon cycling interference by reservoir construction and operation, especially in the global context of the increasing construction of reservoirs.