Legacy effect of elevated CO2 and N fertilization on mineralization and retention of rice (Oryza sativa L.) rhizodeposit-C in paddy soil aggregates

Rhizodeposits in rice paddy soil are important in global C sequestration and cycling. This study explored the effects of elevated CO2 and N fertilization during the rice growing season on the subsequent mineralization and retention of rhizodeposit-C in soil aggregates after harvest. Rice (Oryza sativa L.) was labeled with 13CO2 under ambient (400 ppm) and elevated (800 ppm) CO2 concentrations with and without N fertilization. After harvest, soil with labeled rhizodeposits was collected, separated into three aggregate size fractions, and flood-incubated for 100 d. The initial rhizodeposit-13C content of N-fertilized microaggregates was less than 65% of that of non-fertilized microaggregates. During the incubation of microaggregates separated from N-fertilized soils, 3%–9% and 9%–16% more proportion of rhizodeposit-3C was mineralized to 13CO2, and incorporated into the microbial biomass, respectively, while less was allocated to soil organic carbon than in the non-fertilized soils. Elevated CO2 increased the rhizodeposit-13C content of all aggregate fractions by 10%–80%, while it reduced cumulative 13CO2 emission and the bioavailable C pool size of rhizodeposit-C, especially in N-fertilized soil, except for the silt-clay fraction. It also resulted in up to 23% less rhizodeposit-C incorporated into the microbial biomass of the three soil aggregates, and up to 23% more incorporated into soil organic carbon. These results were relatively weak in the silt-clay fraction. Elevated CO2 and N fertilizer applied in rice growing season had a legacy effect on subsequent mineralization and retention of rhizodeposits in paddy soils after harvest, the extent of which varied among the soil aggregates.


Introduction
In rice (Oryza sativa L.) paddy soil, rhizodeposition is an important source of soil organic carbon (SOC) (Kuzyakov, 2002). Rhizodeposits consist of root exudates, sloughed-off cells, mucilage, tissues, cell lysates, and root debris (Kuzyakov and Larionova, 2005;Gregory, 2006), accounting for over 10% of photosynthesized C (Jones et al., 2009;Nguyen, 2009). They contribute 1.5-to 3-fold more C than shoots (Hooker et al., 2005) and represent approximately 26% of the total belowground C input (Liu et al., 2019b). This organic C is often readily available for microorganisms and has a high turnover rate (Kuzyakov, 2002). However, rhizodeposits can also be stabilized by physical (aggregation), chemical (e.g., SOC-mineral interaction), and microbial processes (assimilated into the biomass and humified) and preserved as SOC (Six et al., 2002;Cotrufo et al., 2013;Wei et al., 2019;Cui et al., 2020). Recent studies have reported significant increases in SOC in paddy soil (Pan et al., 2004;Wu, 2011), largely due to the practice of residue incorporation and enhanced belowground C inputs (i.e., root and rhizodeposits) (Lu et al., 2002(Lu et al., , 2003. Climatic changes such as CO 2 elevation and nutrient management such as N fertilization, can influence rhizodeposition and C retention in soils. Elevated CO 2 can increase root biomass and exudates, and augment the root to shoot ratio, enhancing rhizodeposition (Ziska et al., 1996;Bhattacharyya et al., 2013;Zhao et al., 2019). Elevated CO 2 has also been reported to decrease total free amino acids, proline and abscisic acid in root exudates of barley (Calvo et al., 2017) that have relatively great potential to be retained by soil minerals (Keiluweit et al., 2015). In addition, elevated CO 2 increases the C:N ratio of root exudates by decreasing N-rich and increasing C-rich metabolites (Tarnawski and Aragno, 2006). Therefore, rhizodeposits produced under elevated CO 2 conditions have greater potential to be preserved in the soil than to be lost via mineralization. N fertilization also stimulates plant C assimilation, root and shoot biomass production, and rhizodeposition (Kuzyakov, 2002;Ge et al., 2015;Atere et al., 2017;Liu et al., 2019aXiao et al., 2019). Additionally, N fertilization may positively affect N-rich metabolites and have a negative effect on C-rich metabolites, decreasing the C:N ratio of root exudates (Zhen et al., 2016); it may also accelerate the microbial mineralization of rhizodeposits by changing the quality of rhizodeposits and soil C:N ratios, resulting in less C preservation (Li et al., 2017;Zhu et al., 2018;Xiao et al., 2019;Wei et al., 2020). Previous studies have focused primarily on factors influencing the retention and turnover of rhizodeposits during the crop growing season without considering of the fate of the residual rhizodeposits after the harvest (i.e., legacy rhizodeposits) Xiao et al., 2019;Zang et al., 2019). However, the rhizodeposit-C remaining after harvest is essential for SOC preservation.
The retention of rhizodeposits in different soil aggregates may vary due to differences in size, stability, capacity to preserve C (Six et al., 2002), and the accessibility of C to microorganisms (von Lützow et al., 2007;Brookes et al., 2017). This can be crucial for rhizodeposit-C sequestration (von Lützow et al., 2007;Schmidt et al., 2011). Several studies and models have suggested that macroaggregates contain fresh C inputs with short residence times (Elliott, 1986;Six et al., 2002), while microaggregates and silt-clay size aggregates accumulate SOC with longer residence times (Oades, 1988;Denef et al., 2007); others have reported that C input is preferentially sequestered in microaggregates and silt-clay within macroaggregates (Kong et al., 2005;Smith et al., 2014). Pan et al. (2008) reported that in paddy soils, SOC preferentially accumulated in the macro-and microaggregates, while Atere et al. (2017) found a greater deposition of photosynthesized C in silt-clay size aggregates. This divergence may be associated with the effects of different field practices on aggregate formation (Brye et al., 2012;Choudhury et al., 2014). Despite the conflicting results, these studies highlight the importance of soil aggregates in C retention. However, information on the turnover and retention of legacy rhizodeposits associated with soil aggregates is quite limited.
The objective of the present study was to reveal the legacy effect of elevated CO 2 and N fertilization on the fate of legacy rhizodeposit-C by analyzing the decomposition and retention of legacy rhizodeposit-C derived from rice grown under conditions with and without elevated CO 2 and N fertilization. It was hypothesized that (1) legacy rhizodeposits remaining after the rice growing season are preferentially retained in macroaggregates and are less mineralizable than those retained in microaggregates and silt-clay fractions; and (2) legacy rhizodeposits derived from rice grown under elevated CO 2 conditions are less mineralizable and are preferentially retained in soil aggregates, whereas legacy rhizodeposits derived from rice grown with N fertilization should exhibit the opposite characteristics.
2 Materials and methods 2.1 13 C-CO 2 continuous labeling Typical paddy soil was sampled for the pot experiment at the Changsha Research Station for Agricultural and Environmental Monitoring, Hunan, China (113°19′52″ E, 28°33′04″ N; 80 m above sea level). The area is characterized by a subtropical climate, with a mean annual temperature of 17.5°C and an annual rainfall of 1300 mm. The soil (sand 16.5%, silt 73.5%, and clay 10.0%) was collected from 0 to 20 cm depth. The soil (pH 6.4, 1:2.5 w:v) contained 27.10 g kg -1 total organic carbon, 2.60 g kg -1 total nitrogen, and 0.74 g kg -1 total phosphorus. Rhizodeposit labeling was conducted as described by Ge et al. (2013). Each pot (17.2 cm diameter, 16.7 cm height) was filled with soil, equivalent to 1.2 kg dry weight. After applying base fertilizers (KH 2 PO 4 20 mg P kg -1 soil, KCl 120 mg K kg -1 soil), three rice seedlings were planted in each pot, and the soil was flooded with water 3 cm deep. The pots were then transferred to automatically-controlled, gas-tight growth chambers (150 cm length, 100 cm width, 150 cm height).
There were four treatments: ambient CO 2 level (400 ppm) and no N fertilizer (AN 0 ); elevated CO 2 level (800 ppm) and no N fertilizer (EN 0 ); ambient CO 2 level and 100 mg N kg -1 (equivalent to 225 kg N ha -1 ) (AN 100 ); and elevated CO 2 level with 100 mg N kg -1 (EN 100 ). CO 2 was generated by acidifying Na 2 CO 3 (in beakers placed inside the growth chambers), with 0.5 M H 2 SO 4 , and the concentration was monitored using a CO 2 analyzer (Shsen-QZD, Qingdao, China). When the CO 2 concentration in the chamber fell below 100 µL L -1 , CO 2 was generated to maintain a concentration of 380-400 or 780-800 µL L -1 ; when it rose above 400 or 800 µL L -1 , a switch diverted the gas to pass through CO 2 traps (1 M NaOH solution) to absorb the excess CO 2 . During the tillering and elongation stages (a 43 d period), the CO 2 was continuously labeled with 2.8 atom% 13 C, and the beaker was filled with Na 2 13 CO 3 every 6 d (1.0 M, 99 atom% 13 C; Cambridge Isotope Laboratories, Tewksbury, MA, USA). The urea-N fertilizer was separated into three portions in a ratio of 5:3:2 to apply at the seedling, tillering, and heading stages. After the rice harvest, the soil was collected, and the total C content and 13 C atom% of the different soil particle fractions and bulk (unfractionated) soil were determined. The natural abundance was obtained by the same process but without 13 C labeling.

Soil particle fractionation and incubation setup
After 150 d, the rice was harvested, and the soil was collected. Plant roots and leaves in the soil were carefully removed by handpicking. The soil was sequentially wet-sieved with distilled water into macroaggregate (>250 µm), microaggregate (53-250 µm), and silt and clay ( < 53 µm) fractions as described by Atere et al. (2017) and dried overnight at 50°C. The bulk soil and fractionated particle fractions collected from the four treatments (AN 0 , EN 0 , AN 100 , and EN 100 ) were used to conduct an incubation experiment. Approximately 20 g soil was transferred into a 500 mL glass bottle, flooded with water 2 cm deep, sealed with a rubber stopper, and kept for 100 d at 25°C. Using a gas-tight syringe, approximately 30 mL headspace gas was sampled every day for the first 15 days and then every 2-10 days thereafter, until the end of the experiment. The samples were stored in pre-evacuated Exetainer glass bottles (Labco, High Wycombe, UK); the glass bottles were kept closed between the sampling dates and were only opened for~30 min each sampling day to be aired with atmospheric air for aerobic maintenance. The CO 2 concentration and 13 CO 2 atom% were determined. The soil was destructively sampled on days 5, 30, and 100 for further measurements.
2.3 CO 2 efflux, microbial biomass carbon and 13 C isotope analysis CO 2 concentrations of the headspace gas samples were measured using a gas chromatographer (Agilent 7890A; Agilent Technologies, Palo Alto, CA, USA) equipped with a thermal conductivity detector. The stable C isotope composition of CO 2 in the headspace gas samples was analyzed using an isotope ratio mass spectrometer coupled with a GasBench (Thermo-Fisher Scientific, Waltham, MA, USA). The soil microbial biomass C (MBC) of the bulk soil and aggregate samples was measured following the fumigation extraction method (Wu et al., 1990). The equivalent of 15 g dry soil was extracted with 60 mL of 0.05 M K 2 SO 4 . Another 15 g dry soil sample was fumigated with alcohol-free CHCl 3 for 24 h and extracted in the same manner. The MBC was calculated based on the difference in dissolved organic C (DOC) between the fumigated and non-fumigated samples, using a K EC factor of 0.45 (Wu et al., 1990). The mineral N (NH 4 + and NO 3 -) content in the same non-fumigated extract was measured. Extracts of fumigated and non-fumigated samples were freeze-dried and subsequently subjected to d 13 C analysis. Soils were mixed thoroughly, and a subsample (10 g) was air-dried and ball-milled before total C and d 13 C analyses. The stable C isotope composition of the soils and freeze-dried extracts was analyzed using a MAT253 isotope ratio mass spectrometer (Thermo-Fisher Scientific, Waltham, MA, USA) coupled with a FLASH 2000 elemental analyzer (Thermo-Fisher Scientific). Total C was analyzed using a TOC-VWP total C analyzer (Shimadzu, Kyoto, Japan).

Calculation and statistics
The d 13 C values of the soil with labeled rhizodeposits and CO 2 were converted into d ( ‰) relative to the Pee Dee Belemnite (PDB, 0.0111802) standard and further expressed in atom% as follows: The 13 C content (mg kg -1 ) of DOC, SOC and CO 2 levels were calculated using the following equation: where atom% sample and atom% natural are the atom% 13 C in the labeled and unlabeled sample, respectively, and C sample is the total C content (mg kg -1 ) of the sample. The 13 CO 2 content was calculated using an analog equation. The SOC-13 C content measured before incubation was assumed to be all sourced from rhizodeposits and is referred to as the initial rhizodeposit-13 C content. The allocation of rhizodeposit-13 C of the bulk soil and aggregates (macroaggregates, microaggregates, and siltclay) into different pools (CO 2 , DOC, MBC, and SOC) was calculated as follows: DOC Proportion bulk ¼ 13 DOC bulk =Rhizo-13 C bulk Ã 100%: (3) DOC Proportion bulk (%) is the proportion of initial rhizodeposit-13 C content transformed into 13 C-DOC in the bulk soil; 13 DOC bulk (mg kg -1 ) is the 13 C-DOCcontent of the bulk soil.
Rhizo-13 C bulk (mg kg -1 ) is the initial rhizodeposit-13 C content of the bulk soil before the start of incubation. The proportion of DOC in macroaggregates, microaggregates, and silt-clay was calculated using the same equation. The proportion of rhizodeposit-13 C content transformed into 13 C-MBC, 13 C-CO 2 , and 13 C-SOC was calculated using the analog equation.
The cumulative 13 CO 2 emission was calculated as follows: where cumulative 13 CO 2 is the cumulative 13 CO 2 emission (% of the initial 13 C-SOC content); n represents the incubation time (days); 13 CO 2t is the 13 C-CO 2 efflux (mg kg -1 d -1 ) at day t; 13 C initial is the initial rhizodeposit-13 C content of each aggregate (mg kg -1 ) of the soil before incubation. The kinetics of the mineralization was described by fitting a first-order single exponential function: where b describes the amount of bioavailable labeled rhizodeposits pool, k is the mineralization rate of the pool of rhizodeposits, and t is time (days). The parameters obtained were used to calculate the mean residence time (MRT) as 1/k, and half-life as ln2/k . Statistical comparisons between treatments within the same aggregate fraction or between aggregate fractions within the same CO 2 and N treatment were performed using analysis of variance with Tukey pairwise post-hoc testing, and effects were considered significant at p < 0.05. A twofactorial analysis of variance (ANOVA) was conducted on rhizodeposit-13 C content, cumulative CO 2 emission, 13 C-MBC, 13 C-DOC and 13 C-SOC proportions with the factors of elevated CO 2 concentration and N fertilization. Pearson's correlation was conducted between daily 13 C-CO 2 , 13 C-MBC, and 13 C-DOC proportions and soil physicochemical parameters in each aggregate with data from three soil sampling dates. Statistical analyses were conducted using SPSS v22 (IBM Inc., Armonk, NY, USA).
In contrast to its mass distribution, the rhizodeposit-13 C content of each fraction declined with decreasing aggregate size ( Fig. 2A, B). The interactions between elevated CO 2 and N fertilization were observed in the rhizodeposit-13 C contents of the bulk soils and microaggregates (Online Resource 2). Elevated CO 2 increased the rhizodeposit-13 C content of bulk soil and all aggregate fractions by 10%-80% (for the comparisons between AN 0 vs. EN 0 and AN 100 vs. EN 100 ) (p < 0.05). In contrast, N fertilization reduced the content in bulk soils and microaggregates under elevated CO 2 conditions by 24% and 36%, respectively, (for the comparisons between EN 0 vs. EN 100 ) (p < 0.05) (Online Resource 2; Fig. 2B). Similarly, N fertilization under ambient CO 2 conditions reduced the rhizodeposits in microaggregates by 37% Fig. 1 Distribution of (A) mass, (B) soil organic carbon (SOC), and (C) rhizodeposit-13 C of the bulk soil in different aggregates after rice growth under ambient CO 2 concentration without N fertilization (AN 0 ), elevated CO 2 concentration without N fertilization (EN 0 ), ambient CO 2 concentration with N fertilization of 100 mg N kg -1 soil (AN 100 ), and elevated CO 2 concentration with N fertilization of 100 mg N kg -1 soil (EN 100 ) treatments. All data are means±SE (n = 3).
(for the comparisons between AN 0 vs. AN 100 ) (p < 0.05), but had no effect on the other two fractions.
3.2 Mineralization of rhizodeposit-13 C On day 5, when the most easily consumable C compounds existed and were being preferentially used, the rhizodeposit mineralization rate was fast, and 0.8%-1.0% of rhizodeposit-13 C in macroaggregates of all treatments was mineralized to CO 2 . The cumulative 13 CO 2 emission on day 5 increased with decreasing aggregate sizes (1.2%-2.2% of microaggregates, 2.1%-2.8% of silt-clay) (Fig. 3). The cumulative 13 CO 2 emission of all aggregate fractions showed greater mineralization in N-fertilized soil (AN 100 and EN 100 ) than nonfertilized soil (AN 0 and EN 0 ) on day 5 (p < 0.05). In the silt-clay fraction, cumulative 13 CO 2 emission from the treatment of elevated CO 2 (EN 0 and EN 100 ) was higher than the ambient CO 2 treatment (AN 0 and AN 100 ) on day 5 but was opposite in the microaggregates. At the end of the experiment (100 d), 12%-25% of rhizodeposit-13 C had been mineralized (Fig. 3), thus exceeding the level of SOC mineralization (p < 0.05) (Online Resource 1) and suggesting that rhizodeposit-C is more labile and available to soil microorganisms.
In the bulk soil, macroaggregates, and microaggregates, cumulative 13 CO 2 emission at the end of incubation (100 d) was higher in the N treatments (AN 100 and EN 100 ) than in treatments with no N fertilizer (AN 0 and EN 0 ) (p < 0.05); 3%-9% more rhizodeposits were mineralized to CO 2 in N-fertilized soil. In N-fertilized soil, the treatment with elevated CO 2 showed lower cumulative 13 CO 2 emission at the end of incubation than the treatment with ambient CO 2 in the bulk soil, macroaggregates, and microaggregates (p < 0.05), but it was the opposite in the silt-clay fraction (p < 0.05) (Fig. 3). In the non-fertilized soil, the cumulative 13 CO 2 emission at the end of the experiment was in the order of silt-clay>microag-gregates>macroaggregates, whereas in N-fertilized soil, it was highest in the microaggregates and lowest in macroaggregates (Fig. 3).
The cumulative 13 CO 2 emission fitted to the first-order kinetic model [Eq. (5)], and the MRT of different aggregate fractions decreased with reducing aggregate size ( Table 1). The MRT and bioavailable pool size of rhizodeposit-13 C in the aggregate fractions varied across the treatments with elevated CO 2 and N fertilization, which however, was not observed in the bulk soils (Table 1). The rhizodeposits derived from rice grown under elevated CO 2 treatments showed a 25%-51% shorter MRT and a 17%-45% smaller bioavailable C pool size than under ambient CO 2 treatments for all soil fractions (p < 0.05), except in macroaggregates without N fertilization and microaggregates with N fertilization (Table 1). Fig. 2 Rhizodeposit-13 C values (atom%) (A), and content (mg kg -1 ) (B) of bulk soil and soil aggregate fractions after rice growth in treatments AN 0 , EN 0 , AN 100 , and EN 100 . All data are means±SE (n = 3). Lowercase letters a, b, and c indicate significant differences (p < 0.05) of the measured parameters between different treatments within the same aggregate fraction, and x, y, and z indicate significant differences (p < 0.05) of the measured parameters between aggregate size fractions within the same treatment. The y-axis in (A) starts from the natural abundance value of 1.072% measured in the soil without 13 C isotope labeling.
Rhizodeposits from rice grown in the N-fertilized soil showed a 9%-52% shorter MRT than non-fertilized soil in all aggregates under elevated CO 2 conditions and 42% shorter in microaggregates under ambient CO 2 conditions (p < 0.05) ( Table 1). Rhizodeposits in macroaggregates from N-fertilized soil showed a larger bioavailable pool than the non-fertilized soil under ambient CO 2 conditions, and microaggregates under elevated CO 2 conditions; while in macroaggregates and siltclay under elevated CO 2 conditions, the non-fertilized soil showed a larger bioavailable pool of rhizodeposits (Table 1). Cumulative 13 CO 2 emission was correlated with soil MBC, NH 4 + , SOC, soil organic nitrogen (SON) content, and SOC/ SON ( Table 2). The daily 13 CO 2 emission rate was correlated with the DOC content and 13 C-DOC proportion; NH 4 + ; pH; SOC and SON content; and SOC/SON (Table 2).
On day 5, elevated CO 2 increased the proportion of 13 C-MBC in microaggregates relative to ambient CO 2 in nonfertilized soil (p < 0.05). In the silt-clay fraction, both elevated CO 2 and N fertilization had a positive effect on the 13 C-MBC proportion (p < 0.05) (Online Resource 5). N fertilization with elevated CO 2 had significant effects and interaction effects on the incorporation of rhizodeposit-13 C into DOC in all three aggregate fractions on day 5 (Online Resource 5). On day 5, the proportion of 13 C-DOC in N-fertilized soil under elevated CO 2 treatment was greater than the other treatments in all soil fractions (p < 0.05), and it decreased sharply from day 5 to day 30 (Fig. 4). In macroaggregates, only the 13 C-DOC proportion in the treatment with both elevated CO 2 and N fertilization significantly increased over time (Fig. 4). Along with the mineralization of rhizodeposit-13 C, its incorporation into SOC decreased over time in all treatments of the aggregate fractions and bulk soil (Fig. 4).
At the end of the incubation period (100 d), 5%-44% of the initial rhizodeposit-13 C of each aggregate was in MBC and 31%-83% and < 0.19% were in SOC and DOC, respectively (Fig. 4). At 100 d, N fertilization and elevated CO 2 had no Fig. 3 Cumulative 13 CO 2 emission (% of rhizodeposit-13 C) of bulk soil and aggregate fractions in treatments AN 0 , EN 0 , AN 100 , and EN 100 . All data are means±SE (n = 4). Lowercase letters a, b, and c represent significant differences (p < 0.05) between treatments within the same aggregate fraction on day 100 of incubation, and x, y, and z indicate significant differences (p < 0.05) between different aggregate fractions within the same treatment on day 100 of incubation. Two-factorial analysis of variance (ANOVA) was conducted with the factors of elevated CO 2 concentration (eCO 2 ) and N fertilization (N) on cumulative CO 2 emission at the end (100 d) of incubation. *, **, and *** indicate significance levels of p < 0.05, p < 0.01 and p < 0.001, respectively. Not sig. indicates the effect is not statistically significant. The curves are fitted to a first-order single exponential Eq. (5). effect on the proportions of 13 C-DOC and 13 C-MBC in macroaggregates or the silt-clay fraction (Online Resource 5). For the treatment with N fertilization only, in microaggregates, 2%-5% and 9%-16% increases in rhizodeposits were recorded in DOC and MBC, respectively, over the same in the treatment without N fertilization, especially under ambient CO 2 conditions (Fig. 4). These increases were in contrast to the 13 C-SOC proportion (Fig. 4). Rhizodeposit-13 C derived from elevated CO 2 showed 0.01%-0.14% and 0.1%-22.5% decreased incorporation into the DOC and MBC pools, respectively and 2.8%-23.3% increased incorporation into SOC (in aggregate fractions of both N-fertilized and nonfertilized soils) (p < 0.05), except for macroaggregates (Fig. 4). On day 100, elevated CO 2 and N fertilization showed positive interactions on the proportions of 13 C-DOC and 13 C-MBC but a negative interaction on the proportion of 13 C-SOC in the siltclay fraction (p < 0.05) (Online Resource 5).

Distribution of rhizodeposit-C in soil aggregates
Rhizodeposits derived from the early growth stage are primarily labile substrates such as exudates, while during the later growth stage, they are relatively more recalcitrant substances of root turnover products, such as sloughed-off cortical cells, fine roots, and senescent root fragments (Aulakh et al., 2001;Jones et al., 2004). Microbial utilization of those labile rhizodeposit substrates is very swift, i.e., within a few hours (Kuzyakov and Jones, 2006;, while for relatively recalcitrant substrates, this consumption could take days (Dormaar, 1992). Pausch et al. (2013) reported that 62% of rhizodeposition was mineralized within 16 days of maize growth. Therefore, the most labile forms of rhizodeposits may have Table 1 Mean residence time (MRT) and bioavailable pool size of rhizodeposit-13 C in bulk soil and particle size fractions in AN 0 , EN 0 , AN 100 and EN 100 treatments. Cumulative 13 CO 2 emission values were fitted to the first order kinetics: 13 C min = b (1e -kt ). MRT = 1/k. All data are means±SE (n = 4). AN 0 , treatment with ambient CO 2 concentration without N fertilisation; EN 0 , treatment with elevated CO 2 concentration without N fertilisation; AN 100 , treatment with ambient CO 2 concentration and N fertilisation at 100 mg N kg -1 soi; EN 100 , treatment with elevated CO 2 concentration and N fertilisation at 100 mg N kg -1 soil.Lowercase letters a, b, and c after numerical values indicate significant differences (p < 0.05) between different treatments within the same aggregate fraction, and x, y, and z indicate significant differences (p < 0.05) between different aggregate fractions within the same treatment. Table 2 Pearson's correlation coefficients between 13 C-CO 2 , 13 C-MBC, 13 C-DOC (% of rhizodeposit-13 C) and aggregates properties.  been lost during rice growth, and the legacy rhizodeposits we studied were the less labile forms. During the separation of the aggregates by the wet-sieving method, 22%-36% of rhizodeposit-13 C was lost. In contrast, the recovery rates of SOC and dry weight for the wet-sieving method were quite high (>85%) (Fig. 1). Previous studies have also reported approximately 90% SOC recovery after wet-sieving (Lopez-Sangil and Rovira, 2013;Zou et al., 2015). These suggest that the lost rhizodeposit-13 C may be mainly in dissolved forms, and this did not influence our study on the distribution, transformation, and fate of rhizodeposit-C within the aggregates. Rhizodeposit-13 C content was highest in macroaggregates, followed by microaggregates and silt-clay (Fig. 2B). 13 CO 2 emission was slower from macroaggregates than microaggregates and the silt-clay fraction (Fig. 3, Table 1). This result was consistent with our first hypothesis that rhizodeposit-C is preferentially retained in macroaggregates and is less mineralizable than in other finer fractions. Roots are preferentially located between and within large macroaggregates under flooding conditions during the rice growing season to allow for greater atmospheric O 2 exchange than if they were within and between microaggregates (Luo et al., 2018). However, due to O 2 conditions, more mineralizable organic materials in macroaggregates decomposed relative to the smaller aggregates during rice growth.
Moreover, the protection of organic matter in microaggregates is stronger than in macroaggregates (Six et al., 2002;von Lützow et al., 2007). This resulted in the preservation of relatively more recalcitrant/well-decomposed rhizodeposits in macroaggregates at the beginning of the incubation period. Furthermore, the macroaggregates' MBC content was generally the lowest (Online Resource 4) because of its lower surface area and reduced habitat for microorganisms, resulting in the slowest mineralization of rhizodeposits in paddy soil. Due to alternate flooding and drying, and tillage practices, it is difficult for macroaggregates to be formed and maintained. Smaller-sized fractions, i.e., silt-clay, accounted for a larger proportion of the soil mass; therefore, higher proportions of rhizodeposit-C and SOC were found in the siltclay fraction (Fig. 1C).
Rice rhizodeposition was greater in treatments with elevated CO 2 than with ambient CO 2 (Fig. 1) that can be attributed to increased photosynthesis (Bhattacharyya et al., 2013); this corroborates the results of studies on other plants (Pendall et al., 2004;Allard et al., 2006;Phillips et al., 2011;Calvo et al., 2017). Elevated CO 2 stimulates root growth, mucilage and associated fungal hyphae, and encourages the formation of macroaggregates from microaggregates, silt, and clay particles (Czarnes et al., 2000;Traore et al., 2000;Rillig and Mummey, 2006;Jastrow et al., 2007). As a result, we observed an increase in macroaggregate mass (Fig. 1). Regarding the rhizodeposit-13 C content, elevated CO 2 had a positive effect on bulk soil and all aggregate fractions, but to varying degrees, the least effect was observed in the silt-clay fractions. Such differences may be the result of particulate organic C derived from root residues being occluded in macroaggregates and microaggregates but not in silt-clay (Six et al., 2002). As a result, silt-clay accounted for a reduced proportion of rhizodeposit-13 C in bulk soil due to elevated CO 2 .
Although N fertilization stimulates plant photosynthesis, it can lower the root/shoot biomass ratio (Ge et al., 2015;Atere et al., 2017;Luo et al., 2018;Xiao et al., 2019) because fewer root exudates and less root biomass are required when the soil N limitation is mitigated (Britto and Kronzucker, 2002;Linkohr et al., 2002;Li et al., 2010;Ge et al., 2017). As a result, the rhizodeposit-13 C content in N-fertilized bulk soil was lower, especially under elevated CO 2 (Fig. 2B). Moreover, N fertilization has been reported to promote C loss via respiration during rice growth Xiao et al., 2019); therefore, more rhizodeposit-13 C may have been depleted in the N-fertilized soil before the start of incubation, especially in microaggregates. As a result, the proportion of microaggregates in rhizodeposit-13 C of bulk soil was less in the N-fertilized soil than the non-fertilized soil. N fertilization had no impact on aggregate mass distribution in rice paddy soil (Figs. 1, 2), consistent with Luo et al. (2018).
In general, rhizodeposits were preferentially retained in macroaggregates, and these legacy rhizodeposits in macroaggregates were less mineralizable. CO 2 elevation and N fertilization had opposite effects on each aggregate's initial rhizodeposit content; the silt-clay fraction was the least affected.

Effect of N fertilization on mineralization and retention of rhizodeposit-C associated with soil aggregates
The legacy effect of N fertilization on rhizodeposit mineralization and retention in paddy soil silt-clay was limited except for at the very beginning stage. On day 5, the silt-clay was the only soil fraction showing a larger 13 C-MBC proportion in the N-fertilized than non-fertilized soil, indicating microbial immobilization at this early stage. However, considering the entire 100-day incubation period of the silt-clay, the cumulative 13 CO 2 emission and its allocation to DOC and MBC were not influenced by N fertilization (Figs. 3, 4, Online Resource 5). This might be because in the silt-clay, C retention was mainly through binding between minerals and organic substrates that contain functional groups, such as hydroxyls and carboxyls (Cotrufo et al., 2013), without other mechanisms such as the occlusion of particulate organic C from macroaggregates (Six et al., 2002;von Lützow et al., 2007). Since the silt-clay is relatively stable in its content, structure and binding capacity in the soil for only one growing season, N fertilization could hardly alter the rhizodeposit retention processes occurring in it.
Similar to the silt-clay, N fertilization had little effect on the initial rhizodeposit-13 C content, and its incorporation in DOC and MBC in macroaggregates (Online Resources 2, 5), but had a positive effect on the cumulative 13 CO 2 emission and the bioavailable pool of its rhizodeposit-C (p < 0.05) (Fig. 3,  Table 1). These results indicated that N fertilization in macroaggregates might result in less recalcitrance of the rhizodeposit and greater mineralization for energy, but not for microbial growth. Besides binding between minerals and organic substrates, macroaggregates also occlude free particulate organic C that could explain the difference in the cumulative 13 CO 2 emission resulting from N fertilization.
Microaggregates were the soil fraction most influenced by N fertilization. N fertilization increased cumulative 13 CO 2 emission and the bioavailable pool of rhizodeposit-C in microaggregates (Fig. 3, Table 1). Under ambient CO 2 conditions, in microaggregates, N fertilization increased the allocation of rhizodeposits to DOC and MBC. Microaggregates also occlude free particulate organic substrates. However, due to lower porosity and O 2 of microaggregates relative to macroaggregates under flooding conditions during rice growth, less-decomposed substrates may remain in microaggregates after harvest, which led to greater microbial mineralization and immobilization. In general, N fertilization had a positive legacy effect on rhizodeposit mineralization and microbial immobilization but to varying degrees across different aggregate fractions.

Effect of elevated CO 2 on mineralization and retention of rhizodeposit-C associated with aggregates
Elevated CO 2 resulted in lower cumulative 13 CO 2 emission than ambient CO 2 levels, especially in N-fertilized soil (except for silt-clay fraction) (Fig. 3), and a reduced bioavailable C pool size of rhizodeposits (Table 1). This result indicated that elevated CO 2 might result in more recalcitrant forms of rhizodeposit-C. Consistent with cumulative 13 CO 2 emission, elevated CO 2 resulted in a lower proportion of rhizodeposit-C incorporated into the DOC and MBC pools; however, it increased incorporation into the SOC in soil aggregates (Fig. 4). This result may be attributed to a change in the recalcitrance of rhizodeposits due to elevated CO 2 .
A significant interaction of elevated CO 2 and N fertilization was observed with the cumulative 13 CO 2 emission and bioavailable C pool size of rhizodeposits in macroaggregates and microaggregates (p < 0.05) (Fig. 3, Table 1). This interaction may be associated with the C:N ratio of rhizodeposits being altered by elevated CO 2 and N fertilization. Elevated CO 2 can increase the plant:root ratio (Norby et al., 2001;Yang et al., 2011), resulting in nutrient limitation and reduce microbial activity and respiration rate (Recous et al., 1995); while N fertilization has the opposite effect (Leuschner et al., 2013;Wei et al., 2017). It could be speculated that elevated CO 2 and N fertilization had a similar effect on the C:N ratio of rice rhizodeposits, although we could not directly obtain evidence of it. When the C:N ratio of rhizodeposits is optimal for microbes, microbial biomass would grow, benefiting C retention (De Sosa et al., 2018). Thus, CO 2 elevation negatively affected rhizodeposit mineralization and microbial immobilization by altering the quality of the rhizodeposits.
In general, as illustrated in Fig. 5, elevated CO 2 resulted in higher rice rhizodeposition after the growing season because of increased photosynthesis (Bhattacharyya et al., 2013). N fertilization resulted in lower initial rice rhizodeposition due to Fig. 5 Conceptual illustration of the effect of CO 2 concentration and N fertilization during rice growing season on the mineralization and allocation of the legacy rhizodeposits in paddy soil aggregate fractions. Thickness of the arrows and squares represents the mass or proportion. lower allocation of root biomass and exudates and greater mineralization loss during the growing season (Linkohr et al., 2002;Ge et al., 2017;Zhu et al., 2018). Regarding the allocation of legacy rhizodeposits after harvest, rhizodeposits from the elevated CO 2 treatments tended to transfer more to SOC and less to CO 2 , DOC and MBC. In contrast, N fertilization tended to result in more rhizodeposits incorporated into CO 2 , DOC and MBC, and less into the stable SOC pool. This distribution pattern was mainly associated with the change in the recalcitrance of rhizodeposits caused by elevated CO 2 and N fertilizer during the growing season (Mooshammer et al., 2014). However, the effects of elevated CO 2 and N fertilizer on rhizodeposit mineralization and allocation to different pools was not always significant across the different soil aggregates. The silt-clay had the largest quantity and fastest turnover of rhizodeposits. However, in siltclay, the change in rhizodeposit mineralization and allocation to different pools due to elevated CO 2 and N fertilizer was quite small, while the change in microaggregates was the largest. This difference was associated with different C preservation mechanisms between the different soil fractions (Six et al., 2002;Kong et al., 2005;von Lützow et al., 2007).

Conclusions
The rhizodeposit-13 C content decreased with increasing aggregate size, but it was opposite for its distribution proportion in aggregates in all treatments because smallersized fractions accounted for a larger proportion of the soil mass. The effect of N fertilization and elevated CO 2 on rhizodeposit mineralization and incorporation into soil organic C pools varied across soil aggregates. N fertilizer increased rhizodeposit mineralization and its incorporation into the soil labile C pools of DOC and MBC, while elevated CO 2 tended to have the opposite effect, although not always significant. The silt-clay was the fraction least affected by the mineralization and retention of legacy rhizodeposit-C, while microaggregates were the most influenced fraction. Therefore, the silt-clay fraction in paddy soil plays a critical and stable role in rhizodeposit-C preservation, while microaggregates have the potential to be regulated and retain more rhizodeposit-C. CO 2 elevation can result in increased preservation of rice rhizodeposit-C, and N fertilizer can cause the legacy rhizodeposit-C to be more mineralizable and less stable.