Introduction

Rivers transport substantial amounts of terrestrial OM from land to the ocean, where it may either be oxidised and released as CO2 into the atmosphere or transported and ultimately buried in marine sediments1,2, regulating the global carbon cycle3,4. Clay-sized mineral aggregates play a pivotal role in this lateral transport process, functioning both as effective vectors transporting terrestrial OM from land to ocean5,6, and critical protector matrices that enable long-term organic carbon (OC) burial in marine sediments7,8. Yet, the mechanisms by which clay-sized mineral aggregates stabilise OM during river transport remain inadequately understood.

Sorption to mineral surfaces has long been recognised as the primary mechanism for protecting OM from degradation9,10,11,12. This view is supported by empirical evidence of the coupling between the contents of organic carbon (OC) in sediments and the specific surface area of the minerals13,14, making mineral surface area the primary control. However, the varying fates of terrestrial OM diverge from predictions based solely on mineral surface area: terrestrial OM is progressively lost, selectively replaced by marine-derived OM, and exhibits differential stabilisation patterns between rock-derived and soil-derived OM, despite mineral surface area remains invariant15,16,17. These discrepancies suggest that mineral surface area alone cannot adequately explain the role of minerals in regulating the fates of OM during lateral transport, underscoring the need to investigate how OM and minerals are spatially organised and stabilised within clay-sized aggregates.

Emerging high-resolution microscopy reveals that OM stabilisation within the clay-sized mineral aggregates involves both mineral surface sorption and pore occlusion11,18,19. At mineral surfaces, OM does not form a uniform monolayer but instead exhibits a discrete and patchy distribution. This distribution is shaped by multiple binding mechanisms, including ligand exchange, cation bridging, and hydrophobic interactions, highlighting the important role of chemical properties of mineral surfaces and OM in their spatial organisation20. The onion-layer and zonal structure models provide conceptual frameworks for this heterogeneous organisation, proposing that nitrogen-rich, microbially processed OM preferentially adsorbs as a basal layer21,22,23,24,25. This initial layer could facilitate the subsequent adsorption of organic molecules compared to bare clay surfaces26,27. While these models are supported by direct microscopic evidence in terrestrial soils26,27, their applicability to fluvial and marine sediments, where saturation and redox conditions differ markedly, remains unclear. Furthermore, these surface-focused frameworks overlook the pore-occluded OM in the clay-sized fraction, hence providing only a partial understanding of the spatial organisation of OM and minerals at this scale. This pore-occluded OM may be stabilised through fundamentally different mechanisms compared to the surface-associated OM, where small pore diameters and tortuous pore networks restrict microbial access and oxygen diffusion23,24. As such, distinguishing between surface-bound and pore-occluded OM within the clay-sized fraction, and determining whether these two pathways selectively protect OM of different origins is essential for understanding the varying fates of mineral-associated terrestrial OM during land-to-ocean transit.

This study explores how OM and minerals are spatially organised within the clay-sized fraction, and whether this spatial organisation correlates with their chemical composition. If specific OM types consistently localise within distinct structural domains (e.g., surface-bound vs. pore-occluded), it would imply that minerals confer divergent stabilisation mechanisms based on OM composition. However, a longstanding technical challenge lies in differentiating these two OM pools within clay-sized mineral aggregates. Unlike intact soil aggregates in bulk samples or aggregates larger than 20–50 µm28, where physical separation techniques can effectively isolate pore-occluded OM24,28,29, the clay-sized fraction is more recalcitrant to full dispersion using traditional separation methods11,30. To overcome this limitation, we employed NanoSIMS imaging technique coupled with a machine-learning image processing method. This approach enables automated, high-resolution, and direct discrimination of surface-bound and pore-occluded OM domains within intact clay aggregates (Methods), allowing us to explicitly quantify whether a systematic spatial-chemical coupling occurs in clay-sized mineral aggregates.

A natural laboratory for examining OM-mineral association

We investigated the Luzon and Taiwan rivers, which border the northeastern South China Sea, the largest marginal sea in the western Pacific5. These rivers transport vast amounts of organic matter (OM) and fine-grained minerals to the northeastern South China Sea. Understanding the nature and dynamics of OM-mineral associations in these fluvial sediments is essential for elucidating the fate of terrestrial OM and evaluating its preservation in the South China Sea shelf and slope. Luzon, the largest island in the Philippines, is characterized by active volcanism and tectonic activity31. Due to volcanic weathering, the clay assemblages of rivers draining Luzon are predominantly composed of smectite (83–87%) with minor kaolinite (10–17%)32. In Taiwan, the combined effects of a humid climate, high precipitation, and frequent seismic activity have made it one of the fastest-eroding mountain belts globally. Taiwanese rivers export substantial fluxes of suspended sediment to the ocean, accounting for approximately 2% of the global sediment flux6. Sediments collected from the Choshui and Kaoping river sites consist primarily of chlorite and illite, with proportions of 44%:51% and 39%:61%, respectively5.

The distinct mineral compositions of these sediments are representative of global sedimentary environments, making this region an ideal natural laboratory for examining the role of minerals in the preservation of terrestrially derived OM. Furthermore, recent studies have shown that OM preserved within these mineral assemblages exhibits variable fates during land-to-ocean transit16,17, highlighting the need to investigate the processes governing the stability of MAOM in these fluvial sediments.

Results

Spatial distribution of OM within fine-grained mineral assemblages

To directly reveal the patterns of OM-mineral associations, we employed high-resolution microscopic techniques, combining nano-scale secondary ion mass spectrometry (NanoSIMS) combined with multichannel machine-learning image analysis (Methods)33. This approach allowed us to visualize and quantify the distribution of OM patches within mineral assemblages at the submicron scale, and to correlate these distributions with the chemical characteristics of both OM and associated minerals.

Our analysis of 1091 organic patches and 4255 inorganic particles revealed two dominant modes of OM-mineral association within the clay-sized fraction of fluvial sediments. In the first mode (Mode I), a considerable proportion of OM patches (red in Fig. 1(a-c)) have their surfaces overlapping those of mineral particles, forming discrete coatings across the mineral surfaces (grey particles in Fig. 1(a-c)). Here, OM preservation via this mode is referred to as surface-coating OM. By contrast, the second mode (Mode II) involves OM patches that appear isolated within the pores of mineral aggregates, with more than 99% (area/area) of their surface area not in direct contact with mineral surfaces (green in Fig. 1(d–f)). These patches, which we refer to as pore-occluded OM, are likely preserved due to the physical protection provided by the small diameter and high tortuosity of aggregate pore networks. This structural confinement limits microbial activity and oxygen exchange, thereby creating an environment conducive to OM preservation.

Fig. 1: Two submicron modes of OM-mineral associations in fine-grained mineral assemblages.
figure 1

a-c, surface coating OM (mode I) is assigned as red (a–c). d-f, pore occluded OM (mode II) is assigned as green. Mineral grains/assemblages are illustrated as grey. The yellow frames of each panel highlight the mineral assemblages in the absence of organic patches.

Full size image

Our direct microscopic analyses corroborate previous findings by confirming the presence of both surface-coating and pore-occluded OM within clay-sized fractions of fluvial sediments. These results raise questions concerning whether OM stablised via these distinct modes differs in chemical composition.

Compositional variations in OM-mineral associations: C/N and δ¹³C composition

To investigate whether the spatial distribution of OM within clay-sized mineral assemblages correlates with its chemical composition, we quantified the C/N and δ¹³C values of each OM patch at a resolution of 78 nm. These measurements were calibrated against bulk elemental analysis and isotope ratio mass spectrometry (EA/IRMS) data from the same samples (Methods).

Our analysis revealed a high degree of heterogeneity in C/N and δ¹³C values at the microscale. Nitrogen-rich (C/N ratios <5–10) and nitrogen-poor regions (C/N > 10) coexist on a single mineral aggregate (Fig. 2(c–d), aggregate diameter ranging from 0.70 to 10.58 µm, with an average diameter of 1.50 µm). When the two preservation modes were examined separately, this microscale heterogeneity in C/N composition resolved into distinct patterns for each mode (Fig. 2(a–d)). Nitrogen-rich OM was primarily found in Mode I, where OM coats the surfaces of mineral particles, while carbon-rich OM was predominantly preserved in Mode II, within the pores of mineral assemblages. The C/N ratios of surface-coating OM patches were 2–3 times lower than those of pore-occluded OM (Fig. 2(c–d), Table S1). Notably, the N-rich OM may also form a basal layer at the inner surfaces of aggregate pores, as illustrated in Fig. 2d. This arrangement could facilitate the subsequent accumulation of C-rich OM, consistent with the layered organisation proposed in the “onion-layer” or zonal structural models21,22. These findings suggest that OM patches with differing C/N compositions are preferentially preserved in specific sites within the clay-sized mineral assemblages. This spatially selective accumulation pattern occurs in both Taiwan and Luzon Island rivers, despite the sediments receiving different OM sources between these sites12,34. In contrast, δ13C composition did not show distinct differences between the two preservation modes (Fig. 2e–f, h), Table S1), likely due to overlapping δ13C signatures of the major organic carbon sources for each site.

Fig. 2: Compositional differences between OM patches preserved via the two modes.
figure 2

a, b, Surface coating OM (mode I) and pore occluded OM (mode II) are circled by red and green lines respectively. c-f, Spatial distribution of C/N and δ13C composition of the surface coating OM (mode I, c, e) and pore occluded OM (mode II, d, f). g, h, Box plots of C/N (g) and δ13C (h) differences between the OM preserved via two modes: boxes represent the range 25–75% with medians as the middle lines, and whiskers represent standard deviation.

Full size image

Mineralogical compositional differences: OM-attached vs. exposed surfaces

Mode I preservation is characterized by OM occurring as discrete patches heterogeneously distributed across mineral surfaces, leaving large portions of the minerals either partially coated or completely exposed (Fig. 1(a–c)). To investigate whether mineral composition influences OM distribution, we analyzed the Si/Al and Si/Fe ratios of mineral particles at a 78 nm resolution. These measurements were calibrated against X-ray diffraction (XRD) data from previous studies on the same samples (Methods).

Our analysis revealed a high heterogeneity in the Si/Al and Si/Fe ratios across the sampled mineral surfaces (Fig. 3). However, when comparing OM-attached versus exposed mineral surfaces, a clear compositional trend emerged. OM-attached surfaces consistently exhibited lower Si/Al and Si/Fe ratios compared to exposed surfaces (Fig. 3(a–f), the ranges of Si/Al and Si/Fe ratios for the OM-attached surfaces and exposed surfaces are listed in Table S1). In the Luzon rivers, 40–50% of OM-attached surfaces had Si/Al ratios below 2, in contrast to just 15% for exposed surfaces. Similarly, 25–40% of the OM-attached mineral surfaces had Si/Fe ratios below 4, compared to only 10% of exposed surfaces. Similar preferential OM attachment to surfaces with lower Si/Al and Si/Fe ratios were observed in Taiwan river systems (Fig. 3(e, f), Table S1). This compositional preference may play a key role in determining where and how OM is stabilized in the clay-sized fraction.

Fig. 3: Compositional differences between mineral surfaces with OM attached and those that are exposed.
figure 3

a, b NanoSIMS imaging showing the distribution of Si/Al ratios across mineral particle surfaces. c, d NanoSIMS imaging illustrating Si/Fe distribution across mineral particle surfaces. Yellow circles highlight mineral surfaces where OM accumulation occurs (Mode I), while unmarked areas represent exposed mineral surfaces without OM attachment. e, f Box plots comparing Si/Al and Si/Fe ratios between OM-attached and exposed mineral surfaces. The whiskers depict 10–90% range, while the boxes represent the range 25–75% interquartile range, with the median indicated by the central line.

Full size image

Primary controls on OM distribution modes: preferential association of nitrogenous compounds with iron

Our analysis shows that the OM preserved via the two modes present considerable compositional differences, raising fundamental questions concerning whether the compositional differences determine the spatial distribution of OM. Here, we found that C/N ratios of surface-coating OM presents a significant positive correlation with Si/Fe ratios of the attached mineral surfaces (Fig. 4), suggesting a preferential association between nitrogenous OM and Fe-rich mineral surfaces. This pattern occurs across the Luzon and Taiwan rivers, irrespective of their distinct mineralogical composition, emphasizing the key role of iron in binding nitrogenous compounds. As such, nitrogen-rich OM predominantly coats mineral surfaces (Mode I), while carbon-rich OM, which exhibits less affinity for direct mineral associations, tends to be occluded within the pores of mineral aggregates (Mode II). This indicates that iron distribution across mineral surfaces plays a critical role in the localization of nitrogenous OM within mineral assemblages, providing an explanation for the distinct spatial localization of nitrogen- and carbon-rich OM in the fine-grained mineral assemblages.

Fig. 4: Micro-spatial coupling between C-rich OM patches and Fe-rich mineral surfaces. revealed by NanoSIMS imaging.
figure 4

a, d, g, Spatial localization of OM patches on mineral surfaces. Grey particles are mineral grains/assemblages and surface-coated OM is shown in red. b, c, e, f, h, i, Spatial distribution of C/N ratios of surface-coated OM and Si/Fe ratios in the underlying mineral surfaces. Blue indicates low C/N and Si/Fe ratios, while green to red represents progressively higher ratios. j, correlations between Si/Fe ratio of OM-attached mineral surfaces and C/N composition of the OM coatings. OM-attached mineral particle surfaces (n = 736) are divided into six groups: Si/Fe=0-5, 5-10, 10-15, 15-20, 20-25, 25-100. The symbols represent the mean values of Si/Fe and C/N ratios of the mineral particle surfaces and their OM coatings in each of the six groups, with the error bars indicating ± 0.5 standard deviation of each dataset.

Full size image

Discussion

A conceptual framework for the structural organisation of OM in clay-sized mineral aggregates

Our high-resolution analysis reveals and quantifies the spatial organisation of distinct OM types and mineral phases within the clay fraction of fluvial sediments. Existing conceptual frameworks have either emphasised OM heterogeneity on mineral surfaces (e.g. “onion layer” and zonal structure models21,22) or addressed pore-occlusion in bulk or coarser aggregates35,36. However, the spatial organisation of different types of OM and minerals within the clay-sized fraction, and whether this spatial organisation correlates with their chemical composition, remain poorly understood11,30, despite the critical role of this fraction in long-term OC burial and in transporting terrestrial OM from land to ocean5,7,37. We address this critical gap by identifying a consistent chemical–spatial coupling within the clay fraction of fluvial sediments: N-rich OM preferentially associates with iron-rich mineral surfaces, whereas C-rich OM is mainly protected within aggregate pores. This pattern holds across sediments with varying OM sources and contrasting clay mineralogy, revealing a previously unresolved mechanism of divergent mineral protection.

The surface enrichment of N-rich OM offers a mechanistic explanation for the widely observed decline in C/N ratios with decreasing particle size and increasing density in riverine and marine sediments. This pattern is consistent with the preferential accumulation of proteinaceous and amide-rich microbial residues in the fine fraction38,39. Conceptual models such as the “onion layer” and zonal structures propose that OM is distributed in multiple layers on mineral surfaces, with N-rich compounds forming stable inner layers that facilitate further OM accumulation 21,22,40. High-resolution analyses by Kopittke et al. (2018) support these models by providing direct imaging evidence for the preferential association of microbial N-rich compounds with mineral surfaces in terrestrial soils. Our findings extend this framework beyond terrestrial soils and suggest that nitrogen-selective attachment mechanisms also operate in fluvial clay fractions.

More importantly, we observed a preferential association between N-rich OM and Fe, indicating that the discrete OM distribution is spatially structured by Fe distribution across mineral surfaces. While previous models have primarily focused on the chemical composition of associated OM, our findings highlight Fe as a key spatial determinant of OM heterogeneity at the microscale. Furthermore, the consistent Fe–N association is observed in both the Taiwan and Luzon river systems despite differences in OM sources and their contrasting clay mineralogy5,32. The Fe signals detected in this study likely reflect a mixture of reactive Fe oxides and structural Fe embedded in clay minerals. Although Si/Fe ratios are slightly higher in smectite-dominated sediments (Luzon) than in illite-/chlorite-dominated sediments (Taiwan) (Fig. 3h, Table S1), it remains unclear whether this difference reflects selective Fe oxide coatings or variations in structural Fe content. We hence do not attempt to distinguish the relative contributions of Fe oxides and clay mineralogy to OM stabilisation. Nonetheless, associations between OM and Fe oxides have been more frequently reported in previous studies, likely due to their higher binding efficiency with OM41,42,43. If the Fe in our samples primarily reflects Fe oxides, our finding may align with Sollins et al. (2009), who reported low C/N ratios in Fe-rich soils regardless of mineral composition. At the molecular level, the N-rich organic coatings on Fe (hydr)oxides might contain abundant proteinaceous amide nitrogen, likely originating from microbial synthesis44. Negatively charged bacterial cell walls can rapidly bind with metals44, and microbially-derived compounds such as extracellular polymeric substances (EPS) have strong affinities to Fe-rich particles45. Additionally, compounds like oxygen- and nitrogen-rich amino acids form inner-sphere complexes with iron oxides, a process widely observed in sediments worldwide42. These interactions facilitate the preservation of reactive organic compounds while more refractory molecules are selectively degraded, helping to explain the long-term persistence of proteinaceous OM in oxic pelagic sediments over million-year timescales8.

In contrast, we find that carbon-rich, plant-derived OM predominantly resides within the pores of clay-sized mineral aggregates. This finding challenges the prevailing assumption that such OM is primarily occluded within large, intact soil aggregates, while fine-grained mineral-associated OM (MAOM) is dominated by microbial inputs35. A recent global meta-analysis has similarly reported substantial enrichment of carbon-rich, plant-derived OM within the fine-grained MAOM46. Our spatially resolved observations build upon this finding by demonstrating that plant-derived OM stabilised within these smaller aggregates exhibits distinct spatial localisation compared to microbial OM. This spatial preference implies that the persistence of C-rich OM is largely governed by physical inaccessibility within the aggregate pore network. Specifically, the size, geometry and orientation of pores and channels constrain the diffusion of larger molecules and oxygen exchange with the surrounding environment and limit accessibility of microbe consumers, hence slowing the degradation of enclosed OM24,29. Notably, our method did not fully resolve the molecular diversity of the organic compounds or their potential surface binding mechanisms. We hence do not exclude the possibility that specific C-rich components, such as carbohydrates or alkyl C, may also associate with mineral surfaces under specific conditions38,47. Further clarifying the molecular diversity of OM and mineralogical characteristics will advance understanding of the processes governing spatial–chemical coupling.

Implications for the fates of terrestrially derived OM during land-to-ocean transit

Our results suggest that minerals provide divergent protections to C- and N-rich OM with each pool stabilized by distinct mechanisms and likely exhibiting different responses when subject to environmental disturbances. Although the clay-sized aggregates are more stable compared to the bulk aggregates, their pore structures can still be altered by changes in salinity, pH, and hydrodynamic forcing frequently occurred during lateral transport from land to ocean48. Changing pore size, tortuosity, and connectivity with surrounding water affect fluid flow and diffusion within and around aggregates49, modifying microbial accessibility and oxygen exchange, and ultimately exposing pore-occluded OM to degradation and reducing its persistence50,51. Meanwhile, the transport pathways of pore-occluded OM (carried within aggregates) differ from those of OM-coated single particles due to differences in size, density, and structure between aggregates and individual particles52.

Our results, along with prior studies on sediments from the Luzon and Taiwan rivers and the northeastern South China Sea4,16,17,53, support this hypothesis. The C/N and δ13C compositions of typical OM sources in these regions suggests that the OM patches preserved in the clay-sized fraction of the two Luzon rivers are primarily composed of microbially-transformed nitrogen-rich and soil & plant-derived carbon-rich OM27,54,55,56. Besides, a considerable proportion of fossil OM is preserved in Taiwan river sediments (Figure S3a-d)12. In the clay-sized fraction of Luzon rivers, microbial-derived OM (92%) is mostly preserved via surface coating on Fe-rich mineral surfaces (Mode I), while soil & plant-derived OM is equally preserved by both modes (Figure S3e). In Taiwan rivers, microbial OM is also predominantly preserved via surface coating on Fe-rich mineral surfaces (87%), whereas soil and plant OM is mainly pore-protected (74%). Fossil OM, however, is primarily preserved via Mode I via coating Fe-rich mineral surfaces (85%). According to our framework, the OM sources, preserved via a preferential association to Fe-rich mineral surfaces, should follow distinct transport pathways compared to the pore-occluded OM, in response to environmental changes occurred during the land-to-ocean transit. Blattmann et al. (2018; 2019)’s investigation on samples collected by sediment traps deployed in the northeastern South China Sea provide supportive evidence: soil-derived, carbon-rich OM (preserved through Mode II) is more likely to be lost during land-to-ocean transport, while rock-derived OM (preserved through Mode I) reaches the deep ocean16,17.

If, as our findings suggest, minerals confer divergent protection to different OM types, then mineral-associated OM transported by rivers may follow distinct cycling pathways in response to environmental changes. This variability is not captured by current models that emphasise mineral surface area or sorption, while overlooking the spatial and chemical specificity of OM–mineral interactions. The spatial–chemical coupling of OM and minerals within clay-sized aggregates remains in its infancy, with fundamental gaps in analytical techniques, quantification strategies, and mechanistic understanding. Our study only represents an initial step, and substantial efforts are required toward addressing these gaps by resolving OM–mineral associations at the clay-fraction scale. For example, to expand the generality of this framework, we suggest targeted laboratory experiments to examine how surface-bound and pore-occluded OM pools respond to changes in salinity, pH, and redox conditions. Complementary field observations should focus on clay-sized aggregates across land-to-ocean transects under varying hydrodynamic and geochemical regimes. Expanding such investigations across diverse spatial and temporal scales will help evaluate the broader applicability of this mechanism. Ultimately, these insights should inform advanced theoretical frameworks capable of quantifying divergent mineral protection mechanisms and refining our understanding of terrestrial OM fate during land-to-ocean transport. Such refined frameworks are essential for predicting carbon cycling dynamics in response to global change across critical land–ocean interfaces

Methods

The islands of Luzon and Taiwan were selected as study areas due to their contrasting organic matter and clay mineral compositions, making their river sediments ideal for studies of organo-mineral interactions. Sediment samples were collected from deposits at the mouths of two Luzon rivers, Vigan (LZ22: 17°33′32.1′′(N), 120°28′1.0′′(E)), and Cagayan (LZ31: 18°7′22.0′′(N), 121°40′14.1′′(E)), two rivers on the west flank of Taiwan, Choshui (CK1: 23°50′08.2′′(N), 120°16′48.0′′(E)) and Kaoping (CK7: 22°28′41.8′′(N), 120°25′09.5′′(E))6,12,32 (Figure S1 in Supplementary information).

The sediment samples were collected from the rivers during rainy seasons in July 2007 (Luzon) and August 2008 (Taiwan). Samples were collected from the topmost ~1–2 cm of the riverbed, from shallow water sites (<1 m depth) close to the river mouths. The sampling sites were carefully selected to avoid contamination from human activities and upstream of tidal influences to ensure the sediment in the deposits originated from the upper rivers. At each site, duplicate sediment samples were collected and preserved prior to analysis. The samples were kept in sealed polyethylene bags in dark conditions at 4 °C in the State Key Laboratory of Marine Geology at Tongji University, with no disturbance prior to analysis.

Physical isolation of clay-sized fraction

To extract the clay-sized fraction of the bulk sediments, a combined size and density fractionation was conducted, following the protocols applied by Vogel et al. (2014) and tested by Inagaki et al., 2019. This method effectively isolates the organo-mineral associations at the micrometer scale, removing materials not directly associated with mineral particles.

During specimen preparation, duplicates from each site were mixed to ensure representativeness, obtain sufficient material for clay fraction extraction, and minimize the cost of NanoSIMS imaging. Prepared 2 mm-sieved sediment samples were dried at 40 °C, and then gradually and gently saturated with sodium polytungstate solution (density 1.8 g·cm-3). The resulting sediment suspensions were allowed to settle overnight, and any free-floating organic matter was gently removed with a water jet. Ultrasonic dispersion was used to fully disperse large micro- and macro-aggregates without causing damage to the primary particle structures. Subsequently, the dispersed samples were centrifuged (15 mins at 3019 g) and washed three times, to remove the sodium polytungstate solution and occluded particulate OM. The remaining fractions were wet-sieved to 20 μm to separate course-grained silty particles and the sediment fraction with diameters smaller than 20 μm was subject to sedimentation to extract the <2 μm clay-sized fraction. This clay-sized fraction was then prepared for NanoSIMS imaging and ratio of total organic carbon (TOC), total nitrogen (TN), and stable carbon isotope (δ13C) analysis.

Bulk analysis of OM composition within the clay-sized fraction

Subsamples of the clay-sized sediment fractions were oven dried at 40 °C, homogenized using an agate mortar and weighed. Inorganic carbon was removed through fumigation in the presence of HCL (72 hours) and drying over NaOH pallets (72 hours) in a desiccator at 60 °C57. TOC, TN, ratio of total organic carbon and total nitrogen (C/N) and δ13C content of the samples were measured using an elemental analyzer equipped with an isotope ratio mass spectrometer (EA-IRMS). Triplicate analyses of the samples show the reproducibility of the tests with standard errors in the range of 0.01–0.02‰.

Microscale element distribution measured with NanoSIMS analysis

NanoSIMS measurements were conducted using a NanoSIMS 50 L (Cameca, France), located at the School of Earth System Science, Tianjin University, China. Subsamples of the clay-sized fractions suspended in aliquots of deionized water were diluted ~20 times in deionized water, and 100 μl of the diluted suspension pipetted onto a silica wafer and dried overnight in a desiccator at room temperature58,59,60. The homogeneous distribution of samples was confirmed through scanning electron microscopy (SEM), and the dry samples coated with Au/Pd that, combined with the use of an electron flood gun, compensated for potential charging effects. Potential contaminants and additional gold coating were pre-sputtered prior to analysis. Cs+ was applied with a primary ion impact energy of 16 keV. During the scanning, 12C-, 13C-, 12C14N-, 16O-, 30Si-, 27Al16O-, 56Fe16O- secondary ions were collected with an electronic dead time fixed at 44 ns. The field of view of the measurements were 40 × 40 μm (516 × 516 pixels) with a lateral resolution of 78 nm. For every sample, 6–15 spots were randomly selected to obtain a representative dataset for quantifying the spatial distribution of organic matter within mineral assemblages.

NanoSIMS image processing was performed by the following workflow, which depends on the spatial distribution of 12C-, 16O- and 12C14N- ion distributions provided by NanoSIMS, including image segmentation using machine-learning algorithm in the Ilastik software and OM classification using logical computation algorithm in the Avizo 9.3.033. NanoSIMS measurements were corrected for election multiplier dead-time (44 ns) by using the OpenMIMS plugin for Fiji. The composite images obtained from NanoSIMS measurements that contain the information of 12C-, 16O- and 12C14N- ion distributions are classified into three classes: background, mineral particles, and organic matter (OM) patches (Figure S2a). 16O- secondary ion images were used as a proxy for mineral surfaces and the OM was further revealed by a joint analysis of 12C- and 12C14N- ion distribution58. All the three ion distributions exhibit considerably significant variability in intensity, texture, and gradient (Figure S2a). The supervised machine-learning algorithm considers the variabilities in ion distributions33. The pixel classification module in Ilastik enables the segmentation algotithm to be trained by drawing trainlines in representative areas for each material class. The training lines were particularly drawn close to the boundaries of each material, to improve the machine-learning efficiency and reduce the number of trainlines for pixel classification. The number of trainlines depends on the level of uncertainties, a result produced by the training algorithm to evaluate the performance of machine-learning, until the hight uncertainties are limited to the borders of each material class. After segmentation, spots smaller than 5–10 pixels were removed from the images, to avoid the noise of inaccurate information (Figure S2b–c). To quantify the co-localisation of OM patches and mineral particles, we combined the segments of OM patches with mineral particles using logical computation algorithm in the Avizo software61. Two types of OM patches are classified depending on their localisation relative to mineral particles. The OM patches co-localized with minerals are detected and quantified (Figure S2d). In this case, the OM patches have their surfaces fully overlapped with mineral particles and the preservation is via coating mineral particles, classified as surface coating OM (mode I). By contrast, the OM patches that do not have pixels co-localize with that of mineral particles are quantified (Figure S2e). Whilst this proportion of OM patches might have some of the edges in contact with mineral particles, the preservation predominantly occurs in aggregate pores, classified as pore-occluded OM (mode II). 1091 organic patches and 4255 inorganic particles are quantified to obtain their chemical compositional features.

An internal calibration method was applied to correct NanoSIMS results60. The correction equations for the internal calibration method were obtained by fitting uncorrected 12C-/12C14N-, 13C-/12C-, 30Si-/56Fe16O-, 30Si-/27Al16O- measured by NanoSIMS on more than 500 μm2 per fraction to the corresponding EA/IRMS measurements from this work and X-ray Fluorescence (XRF) measurements from prior studies of the same samples5,32. This method allows us to calibrate elemental and isotopic ratios obtained from NanoSIMS measurements with the macroscale analysis using EA/IRMS and XRF5,32,60. The correction equations were obtained by fitting uncorrected NanoSIMS results on more than 50 µm2 per fraction to the corresponding values of the same samples obtained from EA/IRMS measurements (N/C and 13C/12C) of OM patches and XRF measurements (Si/Al and Si/Fe) of mineral particles, when the Pearson correlation coefficient (R2) was higher than 0.860. The correction equations for C/N and 13C/12C ratios of OM patches and Si/Al and Si/Fe ratios of mineral particles are listed in Table S1.

Statistical analysis

The normality was checked with Shapiro-Welch tests. Statistical significance of the compositional difference between the OM preserved via modes I and II, and between the mineral surfaces that have OM attached and exposed are tested using independent-sample T test, when the datasets apply to normal distribution. If not, we used nonparametric Mann-Whitney U test for the comparisons. Statistical significance was assessed using two-sided tests with a threshold of p  <  0.05 unless otherwise specified (e.g. p  <  0.001, p  <  0.01). All analyses were conducted in IBM SPSS statistics62.