Multidecadal persistence of organic matter in soils: multiscale investigations down to the submicron scale

Minerals, particularly clay-sized minerals, protect soil organic matter (SOM) from decomposition by microorganisms. Here we report the characterization of SOM and the associated minerals over decades of biodegradation, in a French long-term bare fallow (LTBF) experiment started in 1928. The amounts of carbon (C) and nitrogen (N) in the study area declined over time for six fractions (sand, coarse silt, fine silt, coarse clays, intermediate clays, and fine clays). The C : N ratios of SOM associated with silt fractions remained constant, whereas the ratios significantly decreased in clays, reaching very low values in intermediate and fine clays (C : N<5) after 8 decades of LTBF conditions. Xray absorption spectroscopy revealed the following: (i) bulkscale SOM chemical speciation remained almost constant; (ii) submicron particulate OM was present in coarse clays, even after 79 years of LTBF conditions; and (iii) illite particles became progressively SOM-free with time, whereas mixed-layer illite/smectite and smectites were always associated with OM throughout the bare fallow treatment. In summary, these results suggest that clay-sized minerals preferentially protect N-rich SOM and that smectites and mixedlayer illite/smectite seem to protect associated OM more effectively than pure illites.


Introduction
Soils represent an important carbon reservoir on the global scale: they store 3 times more carbon than the atmosphere (Batjes, 1996) and are currently considered as one of the solutions for climate change mitigation and adaptation in addition to food security as highlighted by the "4 per 1000" initiative (Soussana et al., 2017). Soil organic matter (SOM) encompasses compounds with residence times ranging from days to millennia (Trumbore, 2000), and the mechanisms controlling SOM turnover are actively debated (Dungait et al., 2012;Lehmann and Kleber, 2015;Schmidt et al., 2011).
Except to some extent for pyrogenic C , the current thinking considers organomineral interactions (by adsorption or coprecipitation) as a dominant factor, rather than intrinsic chemical recalcitrance, for the longterm persistence of otherwise labile organic compounds in soil (Baldock and Skjemstad, 2000;Balesdent et al., 2000;Lehmann and Kleber, 2015;von Lützow and Kögel-Knabner, 2010;Schmidt et al., 2011;Sollins et al., 2006). However, the exact mineralogical nature of soils remains barely documented, especially at the submicrometer scale, and in particular for soils where the clay fraction is dominated by phyllosilicates (Barré et al., 2014).
Soil phyllosilicates are very diverse (Wilson, 1999). Among phyllosilicates, smectites are considered to have higher protective capabilities than illite and kaolinite, be-cause of their higher specific surface area and cation exchange capacities (Bruun et al., 2010). Nonetheless, in situ experimental demonstrations are still lacking (Barré et al., 2014). To date, the influence of phyllosilicate mineralogy on the chemical composition of persistent SOM has only been documented in model systems (Mikutta et al., 2007).
Here, we studied samples from long-term bare fallow (LTBF) experiments. These vegetation-free experimental plots offer the unique opportunity to naturally concentrate persistent SOM (Barré et al., 2010;Rühlmann, 1999): with time, biodegradation occurs in the LTBF plots and the carbon content gradually decreases as there is no input of fresh organic carbon.
The present contribution addresses the following fundamental questions: 1. How much pluri-decadal persistent SOM is in the different fractions?
2. What is the chemical nature of pluri-decadal persistent SOM in the different fractions?
3. What is the long-term protective capabilities of the different phyllosilicates?
2 Materials and methods

Soil description and sampling
We used archived samples of the "42 Parcelles" LTBF experiment to study the evolution of SOM towards more persistent forms of SOM over 79 years of biodegradation. The 42 Parcelles LTBF is an INRA (Institut National de la Recherche Agronomique) experiment which started in Versailles (France) in 1928 (Burgevin and Hénin, 1939). Since 1929, the soil in the area has not been cropped and has been weeded (by hand or with herbicides) and plowed (to the depth of 25 cm) twice a year. It is a non-carbonated brown soil (silty loam Luvisol: 16 % clay, 57 % silt, and 27 % sand), and the control plots had an average pH of 6.3 in 1929 and of 5.2 in 2008 (Grasset et al., 2009;Paradelo et al., 2013;Pernes-Debuyser and Tessier, 2002). We used archived samples (air dried and stored in the dark) from four plots (

Particle size fractionation
We subjected soil samples to physical dispersion and particle-size fractionation following a published protocol (Balesdent et al., 1998, Fernández-Ugalde et al., 2016 to separate the following fractions: sand fraction (> 50 µm), silt fractions (20-50 µm, 2-20 µm), and clay fractions (0.2-2, 0.05-0.2, and < 0.05 µm). Approximately 50 g of soil was shaken overnight with 20 glass beads in 180 mL of deionized water to break aggregates bigger than 50 µm. The suspension obtained was then passed through a 50 µm sieve, thereby preventing the disruption of particulate OM during the subsequent ultrasonic dispersion (Balesdent et al., 1991), and sonicated in an ice bath for a total input of 320 J mL −1 using a digital sonifier (Sonics model 500W operating at 20 kHz -max. output of 120 Wwith a probe with a flat tip of 2.5 cm diameter). The dispersion conditions allowed for the recovery of a clay-size fraction (< 2 µm) equivalent in proportion to that achieved during standard particle-size fractionation (Balesdent et al., 1991). At this stage, the solution was centrifuged at 79 g for 12 min to first isolate the clay fraction (0-2 µm) and then at 5 g for 1 min to isolate the silt fractions (2-20 and 20-50 µm). Before weighing, sand and coarse silt fractions were oven-dried at 60 • C, whereas the fine silt and clay fractions were freezedried. The total mass recovery was 99.1 % (SD = 0.44 %).
Clay subsamples (2 g) were suspended in water and sonicated at 320 J mL −1 . The sonicated suspension was then sequentially centrifuged at 23 056 g for 61 min to isolate the fraction < 0.05 µm and at 5764 g for 15 min 22 s to separate the smaller fractions (0.05-0.2 and 0.2-2 µm). All clay subfractions were then freeze-dried. Mass recovery for the clay fractionation was 98.3 % (SD = 1.70 %).

Clay mineralogy analysis
The mineral compositions of the total clay fraction and of the clay subfractions (CC, IC, and FC) were determined by X-ray diffraction (XRD) analyses. Air-dried, oriented de-Biogeosciences, 16, 1401-1410, 2019 www.biogeosciences.net/16/1401/2019/ posits prepared using the filter transfer method (Moore and Reynolds, 1989) were analyzed with a Cu Kα radiation Rigaku UltraX18HF X-ray diffractometer (Rigaku, Tokyo, Japan). The XRD patterns were accumulated at 0.05 • step intervals with a counting time of 3 s in the range from 3 to 35 • (2θ). One set of replicates was then exposed to ethylene glycol vapor over 16 h at 60 • C before being remeasured under the same conditions as the air-dried samples.

Elemental analysis
The total organic carbon (C) and nitrogen (N) contents of all particle-size fractions were measured by dry combustion in a CHN autoanalyzer (Carlo Erba NA 1500). The total C content is equivalent to the total organic carbon content, as the investigated soil samples do not contain carbonates.

Synchrotron-based NEXAFS spectroscopy
In the present study, synchrotron-based C-NEXAFS data were collected using beamlines located at the Canadian Light Source (CLS, Canada) where the storage ring is operated at 2.9 GeV and at a current between 250 and 150 mA.

C-NEXAFS spectroscopy at the "bulk" scale
The "bulk" carbon speciation of clay subfractions was investigated by NEXAFS spectroscopy using the CLS beamline 11-ID-1 Spherical Grating Monochromator (SGM; Regier et al., 2007). See Supplement for details regarding the method. Each spectrum reported in the present study corresponds to an average of about 50 measurements. Of note, only the first 250 nm of the sample surface are probed using the SGM setup. Spectra were averaged, background subtracted, and normalized using the Igor Pro software.

STXM-based NEXAFS spectroscopy
STXM-based NEXAFS data were collected using the CLS beamline 10ID-1 (SM beamline, Kaznatcheev et al., 2007), which operates in the soft X-ray energy range (130-2500 eV) using an elliptically polarized undulator. See the Supplement for details regarding the method. All clay subfractions were analyzed for each sampling time, i.e., 15 samples were analyzed. We first analyzed each sample at the millimeter scale to identify carbon bearing regions of a few square micrometers. We then collected STXM-based NEXAFS data over the 250-450 eV energy range covering the C K-edge (280-295 eV), the K L 2,3 -edges (295-305 eV), the Ca L 2,3edges (345-355 eV) and the N K-edge (395-405 eV). For each sample, we analyzed one or two regions that contained carbon-rich particles and presented a diversity of absorptions in order to qualitatively investigate the organomineral interactions present in each fraction on each date. The C-NEXAFS spectra shown here correspond to homogeneous areas of several hundreds of square nanometers. Extensive databases of reference C-NEXAFS spectra are available in the literature (Dhez et al., 2003;Solomon et al., 2009;Le Guillou et al., 2018). The compositional maps presented in this study are derived from the analysis of the selected micrometric regions, they are the visual representation of the assignment of each pixel of the image to the category of particle it belongs to, according to its associated NEXAFS spectra.

C-NEXAFS data deconvolution procedure
To obtain a more "quantitative" insight on the evolution of the molecular signatures of the investigated experimental samples with increasing bare fallow duration and to be able to compare the spectra, we performed (i)

Statistical analyses
Statistical analyses were conducted using the R free software environment for statistical computing (http://www.r-project. org, last access: 27 March 2019). The significance of the difference between C contents or C/N ratios was assessed using pairwise t tests. A level of significance of P = 0.05 was considered for all analyses.

Carbon and nitrogen declines in the fractions
Initial C and N concentrations were low and very low in the sand and coarse silt fractions, respectively, but much higher in the fine silt and clay fractions ( Table 1). The IC and CC subfractions displayed an initial OC content (35.9 ± 2.19 mgC g −1 fraction and 46.6 ± 2.95 mgC g −1 fraction, respectively) 3 to 5 times lower than that of the FC fraction (147.8 ± 16.9 mgC g −1 fraction). Of note, dissolved organic carbon may contribute to this high value (Balesdent et al., 1998;Tiessen and Stewart, 1983). The C and N contents of the bulk and of each fraction decreased with increasing LTBF duration and somehow stabilized after 52 years of BF conditions (Table 1; Fig. 1). The C and N declines in the coarse silt fraction were not that clear, but these fractions had very low N and C concentrations. The low C decline (53 %) compared with the N decline (74 %) in the sand fraction could be explained by the presence of C-rich sand-sized coal or pyrogenic carbon . Overall, the CC, IC, and FC subfractions lost 59 %, 49 %, and 67 % of their initial  C content and 52 %, 36 %, and 48 % of their initial N content, respectively. The C : N ratios of the different fractions evolved differently with bare fallow duration, while that of the bulk soil remained roughly constant. The C : N ratio of the sand fractions increased (due to the increased proportion of pyrogenic carbon), while those of the clay fractions decreased (the C : N ratios of IC and FC fractions reached values as low as 4.5).

Mineralogy of the clay subfractions
The XRD patterns of the three clay subfractions of a bare fallow sample (plot 21) collected at t = 0 showed that, in agreement with previous results obtained on similar soils (Fernández-Ugalde et al., 2016;Hubert et al., 2009), the CC fraction contained smectite, illite, kaolinite, and mixed-layer illite and smectite, whereas the IC and FC fractions only contained smectite and mixed-layer illite and smectite (Fig. 3). The mineral composition of each investigated clay subfraction did not evolve during the 79 years of bare fallow conditions as indicated by XRD (data not shown). The soils studied did not contain significant amounts of iron (Fe) or aluminium (Al) minerals as evidenced by Fernández-Ugalde et al. (2013) who analyzed soils from an adjacent field at the same experimental site in Versailles.

"Bulk" C-NEXAFS spectroscopy
The "bulk" C-NEXAFS spectra of the clay subfractions were very similar, with spectral features attributed to four main chemical moeities (Fig. 4): aromatic or olefin carbons (peak between 285 and 285.5 eV), carbonyl groups (shoulder between 286.1 and 287.1 eV), aliphatic carbons (shoulder at 287.7 eV), and carboxylic groups (intense peak at 288.6 eV). The peak at 290.3 eV is attributed to carbonates (Bernard et al., 2015). The features at approximately 283-284 eV are artefact features created by the presence of minerals that significantly absorb the beam throughout the carbon absorption energy range. Deconvolution of the data allowed for the semiquantitative estimation of the relative concentration of the four functional groups described above (aromatics/olefinics, carbonyls, aliphatics, and carboxylics) as a function of time and revealed no major evolution except  for a slight but statistically significant increase in carboxylic moeities contained in the CC subfraction and a slight but statistically significant decrease in aliphatics in the FC subfraction (Fig. S3 in the Supplement).

NEXAFS spectra of organomineral particles
Four different types of assemblages were observed in the clay subfractions (Fig. 5): (1) SOM-poor K-rich minerals, for which the abundance increased with bare fallow duration; (2) organomineral complexes rich in C, N, and K; (3) organomineral complexes rich in C, N, K, and Ca; and (4) K-poor particulate OM. Of note, pure OM without any signal from K minerals was not identified in the present samples (a possible explanation could be that the signal corresponds to the average of at least 10 pixels of 40 nm×40 nm in size, it is therefore unlikely that pure OM of pure K would be isolated at this submicron scale), which is why there is a distinguishable contribution of K in the OM-rich reference spectra. Still, these K-poor OM particles can be described as particulate OM (Keiluweit et al., 2010). Similarly, pure mineral particles were not identified in the samples investigated: there was always a distinguishable contribution from C (same possible explanation).

Evolution of organomineral particles in the three clay subfractions with time
Over the course of the bare fallow treatment, CC subfractions displayed particles ranging from isolated OM (particulate OM) to mineral-rich, OM-bearing particles (Figs. 5 and 6). virtually undetected in the IC and FC subfractions. These subfractions exhibited a homogeneous signal similar to the spectra of OM+K+Ca organomineral particles, i.e., assemblages involving SOM and smectite or mixed-layer clay particles (Figs. 5 and 6).

Persistent SOM in clay fractions is N-rich
After 79 years of bare fallow conditions, SOM remained in all of the soil fractions. Apart from the coarse silt fraction which contained very low amounts of SOM, the amount of OC and N remaining after 79 years of bare fallow conditions were, as expected, higher in the clay fractions (Table 1). Of note, the relatively high amount of C remaining in the sand fractions could be explained by the presence of pyrogenic carbon in these fractions (Table 1, Fig. 2).
The percentage of C and N remaining increased with decreasing particle-size for fine silt, CC, and IC; however, the same was not observed for FC, where much lower percentages of C and N remained, suggesting a higher labile SOM content in the FC fraction. This is likely due to the fractionation procedure which can favor the accumulation of labile dissolved OM in the FC fraction (Laird et al., 2001).
Although it was not possible to determine the nitrogen speciation, our results show that persistent SOM associated with clays is highly enriched in N. Indeed, particulate OM with high C : N ratios is present in coarse fractions, whereas smaller particle-sizes typically have lower C : N ratios (Balesdent et al., 1987;Christensen, 1992;Fernández-Ugalde et al., 2016). Interestingly, while the C : N ratios of SOM associated with silt fractions did not evolve, the C : N ratios of SOM associated with clay fractions significantly decreased with bare fallow duration (down to values as low as 4.5 for the IC and FC fractions, which has, to our best knowledge, never been reported before). Accordingly, it appears that compounds with nitrogen moieties have a strong affinity for mineral surfaces (Kleber et al., 2007).
NEXAFS showed no major shift in the chemistry of SOM after several decades of biodegradation under bare fallow conditions besides a slight increase in carboxylic moeities in the CC subfraction and a slight decrease in aliphatics in the FC subfraction, supporting the fact that persistent forms of carbon are slightly more oxidized than the initial forms of carbon (von Lützow and Kögel-Knabner, 2010). The present study also confirmed that persistent SOM is mainly composed of microbial material: on average, all spectra displayed typical patterns of SOM strongly enriched in microbial material (Keiluweit et al., 2012;Kleber et al., 2011).

Particulate organic matter persists in clays after decades of biodegradation
Particulate OM could still be observed in the CC subfraction even after 79 years of bare fallow conditions. The NEX-AFS spectra of these particles highlighted their polyphenolic nature (Keiluweit et al., 2010), suggesting that these were pieces of lignin-rich plant debris, possibly physically protected in submicron aggregates as shown in similar temperate Luvisols (Chenu and Plante, 2006). Thus, the present results show that pluri-decadal persistent SOM is made of N-rich oxidized SOM adsorbed to mineral surfaces and, to a lesser extent, of particulate OM, in agreement with the recently proposed "Soil Continuum Model (SCM)" (Lehmann and Kleber, 2015). A simple mixing model allows for a rough estimation of the amount of C associated with minerals in the CC subfractions. Assuming a C : N ratio of 10 for pure particulate OM (corresponding to C : N observed in silt fractions and referred to as C : N particulate OM ) and a C : N ratio of 4.5 for OM bound to minerals (referred to as C : N OM-minerals ), i.e., without particulate OM (corresponding to the C : N ratio observed in IC and FC subfractions after 79 years of bare fallow conditions), and knowing that the C : N ratio of CC after 79 years of bare fallow conditions is 6.7 (referred to as C : N CC, 79 years ), we can calculate the proportion of OM associated with minerals in CC by solving the following simple equation: where C : N CC, 79 years = 6.7, C : N OM-minerals = 4.5, C : N particulate OM = 10, and x represents the proportion of particulate OM in the coarse clay fraction after 79 years of bare fallow conditions. Solving the equation leads to x = 0.4. Thus, it can be estimated that the CC subfractions contain (at the end) approximately 40 % particulate OM and 60 % SOM associated with minerals. Therefore, in addition to particulate OM, the CC subfractions contain approximately 11.6 mgC g −1 , in the form of OM bound to minerals, i.e., significantly less than the C content of the IC subfractions. This might be explained by the relatively small specific surface area of relatively coarse clay minerals compared with that of finer clay minerals.

Smectitic clays appear more efficient at protecting SOM than pure illites
After 79 years of bare fallow conditions, the FC and IC subfractions, mostly composed of swelling clays, had a higher N content than the CC subfraction which contained both illites and swelling clays (Fig. 3). The CC and IC subfractions had a similar apparent C content after 79 years of bare fallow conditions, which was lower than that of the FC subfraction (Table 1). However, as seen above, TOC (total organic carbon) contents might be misleading in this particular case as CC contains significant amounts of POM (particulate organic matter) (estimated to be around 40 % of TOC in this fraction). Therefore, if we focus on C associated with clay minerals by organomineral interactions (either through adsorption or coprecipitation), we find that at all times, there is more C associated with IC and FC than to CC. These results may be due to the predominance of swelling clays in IC and FC or to the fact that finer clay minerals have a higher specific surface area and can therefore interact with more SOM. These results also suggest that swelling clays may better protect N-rich SOM in particular. Spatially resolved observations at the submicron scale with STXM-NEXAFS clearly showed that mineralogy influences SOM stabilization. Indeed, several illite particles (identified by the presence of K and the absence of Ca) were devoid of OM (Figs. 5 and 6), and the relative abundance of OM-depleted illites increased over time. Conversely, smectite layers (identified by the presence of Ca, as interstratified illite/smectite in the presence of K or pure smectite in the absence of K) were always found in association with OM over the chronosequence. This demonstrates that among phyllosilicates, smectites might have higher SOM protective capabilities than illites. Our results do not allow us to make conclusions regarding the protective capability of kaolinite; thus, additional experiments are needed to confirm the present findings and investigate clays with other mineralogies.
The suggested higher capacity of smectites and mixedlayer illite/smectite to protect OM compared to illites could be due to the presence of calcium which might facilitate the formation of persistent bounds between clay surfaces and OM. Organomineral assemblages rely on different physicochemical interactions, depending on the chemical nature of the OM and of the mineral phase, which include covalent bonding, ligand exchange, or weaker interactions such as Van der Waals for instance (Barré et al., 2014;Kögel-Knabner and Amelung, 2014). Similarly to previous studies (Chen et al., 2014), we observed a colocation of C and Ca (contained in smectites). Although the exact mechanism responsible for this colocation is not clearly identified, one hypothesis is that Ca could facilitate the binding of negatively charged or polarized organic compounds to negatively charged mineral surfaces via cation bridging (Lützow et al., 2006;Mikutta et al., 2007;Rowley et al., 2018). It has been shown that the cation bridging mechanism can promote organomineral interaction and reduce the bioavailability of adsorbed organic molecules for negatively charged 2 : 1 clay minerals (vermiculite) (Mikutta et al., 2007). In contrast to smectites, the negative charges of illite surfaces are mostly compensated for by K + ions. K + ions are monovalent and have a lower charge/radius ratio and, as a result, are much less efficient as bridges between organic compounds and mineral surfaces. Additionally, smectites have a higher specific surface area and could adsorb more OM. These points may explain the lower capability of illite particles to protect SOM as evidenced here.
Data availability. Underlying research data are available upon request from the corresponding author.
Supplement. Materials and methods on C-NEXAFS spectroscopy at the "bulk" scale, STXM-based NEXAFS spectroscopy, and the C-NEXAFS data deconvolution procedure. Additional figures of normalized NEXAFS spectra of coarse and intermediate clay subfractions from all sampling dates (Figs. S1 and S2) and corresponding deconvolution (Figs. S3 and S4). The supplement related to this article is available online at: https://doi.org/10.5194/bg-16-1401-2019-supplement.