The experimental site is part of the long-term Agro-ecosystems, Biogeochemical Cycles and Biodiversity Observatory on Environmental Research (SOERE ACBB), which is managed by the French National Institute of Agricultural Research and Environment (INRAE), in Lusignan, France (46°25′12.91″ N; 0°07′29.35″ E) ( Figure 1 ). The area was under managed grassland, grain cropping or ley arable rotations for at least 70 years before the experimental site was implemented in 2005 (INRAE archive). The climate in the region is characterized by a mean annual temperature of 12°C and by a mean annual rainfall of 750-800 mm. The soil profile is characterized by two main horizons: upper soil horizons, characterized by loamy texture and classified as Cambisol, and lower soil horizons, characterized by rubified clayey horizons rich in kaolinite and iron oxides and classified as Paleo-Ferralsol (34–36).

The experiment started in 2005 and followed a completely randomized block design with four repetitions. The treatments analysed included permanent cropland (CC) with maize–wheat-barley rotation and permanent grassland (GG) ( Figure 1 ). Zea mays L. (maize) was sown at a density of 8.5 seeds m⁻² in the permanent cropland, Triticum aestivum L. (wheat) was sown at a density of 150 seeds m⁻², and Hordeum vulgare L. (barley) was sown at a density of 165 seeds m⁻². A mix of Dactylis glomerata L. (cocksfoot), Festuca arundinacea Schreb (tall fescue) and Lolium perenne L. (ryegrass) was used in the permanent grassland.

The soil was conventionally ploughed and subsequently planted in CC treatments after each crop harvest. No P and K fertilization was applied over time, and N application followed plant demand and the growing season (37). Grasslands (GG) were mowed 3 to 5 times a year with a tractor mower in GG treatments, depending on climatic conditions; all aboveground plant biomass was removed. No animal grazing was carried out on the mowed grassland since the experiment was implemented in 2005. No P or K fertilization was applied over time. N fertilization was only carried out based on the amount of shoot biomass N exported at each mowing event. In addition, the timing and rate of N fertilization were guided by the nitrogen nutrition index (NNI) – between 0.9 and 1.0 – to provide the lowest N limits for potential plant production (38).

Plant biomass was assessed shortly before harvest (once a year for cereals and 3-5 times a year for grasslands) in an area of 7.5 m². Biomass was oven-dried at 60°C for dry matter quantification. Thereafter, the P concentration was determined according to the procedure described by Masson et al. (39). Briefly, 0.5 g of plant powder dry matter was calcined at 450°C for 4 h. Then, the ash was heated in 5 ml of concentrated HNO₃ (at 65%) on a hot plate until near evaporation. The ash was then filtered through P-free filter paper and placed in 50 ml vials, and total P was determined by a radial targeting inductively coupled plasma emission spectrometer (ICP–AES). To ensure the quality of the results, the introduction of blanks and 3 replicates of certified plant powder samples in the analytical run was applied. P export was computed based on the product of the element concentration and dry matter production.

Soil samples were collected in 2005, 2011, 2014, and 2017, when maize was present in the system (i.e., at the beginning of the experiment and 6, 9, and 12 years later, respectively) from the 0-30 cm layer to quantify C and N. For the P analysis, samples from 2005, 2011, 2014 and 2018 (i.e., at the beginning of the experiment and 6, 9 and 13 years later) were taken in a 0-30 cm topsoil layer. Five samples were randomly collected per field repetition; they were mixed to determine the composite sample representative of the field per plot. In total, four truly representative field replications were utilized. Samples were oven-dried at 60°C, sieved to 2 mm, and stored for chemical analysis after soil sampling.

The bulk density of the topsoil (0-30 cm) was determined in 2005, 2010 and 2016 through soil sampling with the aid of a cylinder of known volume. The soil was weighed after drying at 105°C for 24 h, and the density was determined based on the soil mass:cylinder volume ratio.

Total P (Pt) was determined through the ignition method at 550°C for 2 h (40), followed by extraction with 1 mol L^-1 H₂SO₄ for 24 h. EnviroMAT SS-2 standard soil (No. 140-025-002) was used as a reference for the method, and 96% recovery of total P was found. Total organic P (Po) was measured as follows: two subsamples were weighed; one subsample was ignited at 550°C for 2 h; and the other subsample was unignited. P was extracted into both subsamples with 10 mL of 0.5 mol L^-1 H₂SO₄ after the subsample was stirred for 24 h. Po represents the difference in P concentration in the two subsamples. Total inorganic P (Pi) was quantified as the difference between Pt and Po.

P-lability was assessed through five-step sequential extraction (11) modified by (41). Briefly, 0.5 g of soil was weighed in a Falcon tube, and P was sequentially extracted by a) an anion exchange resin (AER) membrane (AR 103 Plate 434 QDP); b) 10 mL of 0.5 mol L^-1 NaHCO₃ through shaking for 16 h; c) 10 mL of 0.1 mol L^-1 NaOH through shaking for 16 h; d) 10 mL of 1 mol L^-1 HCl through shaking for 16 h; and e) 10 mL of 0.5 mol L^-1 NaOH through shaking for 16 h. Pi was quantified by means of the molybdate colorimetry method at 700 nm in alkaline extracts (NaHCO₃ and NaOH) (42), and Pt was determined after mineralization with ammonium persulfate (7.5%) + H₂SO₄ (1:1; v/v) in an autoclave at 121°C for two hours (43). Po was determined based on the difference between Pt and Pi. The Pt of mineralized alkaline extractions and Pi in acid reagents were measured by means of colorimetry at 882 nm after acid phosphomolybdate complex formation (44).

P lability analyses considered include:

The total soil C and N contents were determined through elementary analyses carried out in an elemental analyser (CHN NA 1500, Carlo Erba) after milling the soil samples at 100 μm. All C was considered organic C since there was no carbonate in the soil (25).

P, C and N were analysed as soil stocks; they were determined by considering the concentration and bulk density of the 0-30 cm soil layer. The following soil atomic ratios (C:Po and N:Po) were determined through an equation described by Xue et al. (45):

based on the following atomic weights: C: 12.0107, N: 14.0067 and P: 30.9738 g mol^-1.

One composite sample per treatment was used to perform ³¹P-NMR spectroscopy to investigate the structure of the alkali-soluble P form in samples from 2005 (initial state) and 2018, i.e., 13 years after the experiments were installed. To do so, suspensions with 1 g of soil from each repetition of treatments and 10 mL of 0.25 mol L^-1 NaOH + 0.05 mol L^-1 EDTA were prepared in centrifuge tubes. These tubes were shaken for 4 hours and centrifuged at 3,000 x g for 15 min. The extracts of the four repetitions were homogenized; Pt, Po and Pi contents in the extract were quantified based on the aforementioned methods and freeze-dried. Lyophilized extracts were redissolved in 2.7 mL 0.25 mol L^-1 NaOH + 0.05 mol L^-1 EDTA and 0.3 mL D₂O and stirred in a vortex for 5 min to perform the ³¹P-NMR analyses. The supernatant was separated by centrifugation (3000 x g for 15 min) after 120 min of contact, filtered (<45 µm) and transferred into a 10 mm NMR tube. The ³¹P-NMR spectrum was determined utilizing a Bruker Advance DPX 400 spectrometer operating at a frequency of 162 MHz with proton decoupling, a pulse angle of 90°, a temperature of 20°C, an acquisition time of 0.5 seconds and a relaxation time of 2 seconds. The number of scans ranged from 27,000 to 32,000. Peak area integration was performed electronically at a baseline of 10 Hz in Bruker TopSpin 3.6.2 software. The following phosphate compounds were identified: orthophosphate (7.10 ppm), monoester (6.33 to 5.09 ppm), diester (2.00 to -1.00 ppm), pyrophosphate (-3.14 to -3.58 ppm) and adenosine diphosphate (-9.36 to -9.41 ppm). The hydrolysis of diester compounds was not considered in the peak area integration. The monoester compounds were separated into myo-inositol hexakisphosphate (myo-IHP), scyllo-inositol hexakisphosphate (scyllo-IHP), α–glycerophosphate, β–glycerophosphate and adenosine monophosphate (46–49).

The variables were analysed based on a randomized complete block design, with split plots in time and four repetitions. Soil management was taken as the main factor (plot), and management time was used as the secondary factor (split plot). The adopted statistical model was described through Equation 5.

where Y is the observation; µ is the overall experimental average; B is the blocks (i = 1, 2, 3, 4); T is the soil management (j = 1 and 2); R is the management time (k = 1, 2, 3, 4); and error is the experimental error.

The soil management effects were tested by the Scott–Knott test (p ≤ 0.05) based on complex variance. Variables were subjected to regression analysis to test the time effects (p ≤ 0.05). The statistical analyses were carried out in SISVAR 5.6 software.

The mean dry matter export reached 8.0 and 7.9 Mg ha^-1 year^-1 in CC and GG, respectively. This led to cumulative exports of 112 and 110 Mg ha^-1 (for CC and GG, respectively) at the end of the assessed period (2005-2018).

The dry matter exported by GG during the early experimental years (2005-2008) was 40% higher than that recorded for CC in the same period. However, after this time, it was possible to observe a shift in the weight of the exported dry matter. A yield decrease was observed in both treatments, mainly in GG, over time ( Figure 2 ). The mean P exportation rates were 19.1 and 21.0 kg ha^-1 yr^-1 for CC and GG, respectively. Consequently, the accumulated amount of exported P was 10% higher in GG than in CC (295 and 268 kg ha^-1, respectively) ( Figure 2 ). However, there was no trend of P concentration reduction in the dry matter; the mean P concentration reached 0.23% and 0.28% in CC and GG, respectively.

The temporal pattern observed for the C and N stocks in GG and CC can be seen in Figure 3 . GG enhanced the C stock by 5.5% (an increase of 2.6 Mg ha^-1 within 12 years), whereas CC promoted a reduction of 8.5% in the C stock (a decrease of 4.2 Mg ha^-1 within 13 years). GG increased N by 3.7% (0.19 Mg ha^-1), and CC reduced N by 7.4% (0.38 Mg ha^-1) in the stocks.

There was a total P stock decrease in both CC and GG. A linear decrease of 0.0217 and 0.0159 Mg ha^-1 yr^-1 was identified in GG and CC, respectively ( Figure 4 ). The accumulated depletion of the total soil P stock reached 0.30 Mg ha^-1 (9.4%) and 0.25 Mg ha^-1 (8.0%) in GG and CC, respectively, after 13 years. In addition, the total Pi stock was more aggressively (2.3 times) depleted in GG than in CC ( Figure 4 ). This depletion represented -0.75 and -0.29 Mg ha^-1 in both treatments, respectively, at the end of 13 years. However, the Po stock increased by 42.6% in GG at a rate of 32 kg ha^-1 yr^-1; it did not change in CC ( Figure 4 ).

A significant reduction in the labile P fraction was observed in both GG and CC over time ( Figure 5 ). However, the moderately labile P ( Figure 5 ) and nonlabile P ( Figure 5 ) fractions were not changed by the adopted soil management systems. Moderately labile P represented 44% of the total P stock in the 0-30 cm topsoil layer, on average, although it was not statistically significant. There was a mean of 0.13 Mg ha^-1 decrease in both treatments, representing a 47% decrease in the total P stocks. However, nonlabile P, assessed based on the 0.5 mol L^-1 NaOH extraction product, represented 22% of the total P stock in the 0-30 cm topsoil layer on average. Labile P depletion was 0.13 and 0.17 Mg ha^-1 in CC and GG, respectively; it represented a 23% and 32% reduction of its initial content and 53% and 56% of the total P stock depletion in CC and GG, respectively. Labile Pi and labile Po showed opposing trends over time. There was a labile Pi depletion of 0.16 and 0.21 Mg ha^-1 at rates of 0.012 and 0.016 Mg ha^-1 y^-1 in CC and GG, respectively ( Figure 6 ), whereas labile Po stocks increased by 49% (from 27 to 69 kg ha^-1) under the GG soil management system, but they did not significantly change in CC ( Figure 6 ).

C, N and Po stoichiometry changed depending on the management systems due to changes in C and Po stocks ( Figure 7 ). There was a significant reduction in the C:Po atomic ratio in the GG (from 116 to 86) and CC (from 117 to 104) systems over time. The N:Po atomic ratio decreased from 11 to 8 in GG and from 11 to 10 in CC. The C:N:Po atomic ratios were 117:11:1 and 116:10:1 in CC and GG, respectively, at the beginning of the experiment; they were 104:10:1 and 86:8:1 in CC and GG, respectively, 13 years later.

Soil Po and Pi forms at the beginning of the experiment (2005 - initial) and, after 13 years of cultivation (2018) under different soil management systems, were assessed after extraction with a solution containing 0.25 mol L^-1 NaOH + 0.05 mol L^-1 EDTA and a thorough ³¹P-NMR analysis. This solution extracted 57%, 54% and 51% of the total soil P at Pi/Po ratios of 2.7, 2.7 and 2.8 in the initial, CC and GG management systems, respectively ( Table 1 ). The Pi/Po ratio in the extraction solution was higher than the rate in the soil (2.0, 1.6 and 1.0 in the initial, CC and GG management systems, respectively); this finding indicates greater Pi extraction than Po extraction.

Orthophosphate, orthophosphate monoesters, orthophosphate diester, pyrophosphate and adenosine diphosphate were identified through ³¹P-NMR ( Figure 8 ). Although there was a slight indication in the spectra, peaks assigned to diester compounds (2 to -1 ppm) in CC and GG samples were below the detection limit available in the device. Orthophosphate (296, 250 and 250 mg kg^-1 at 72, 74 and 75%, respectively) prevailed in all spectra, followed by monoester (111, 92 and 88 mg kg^-1 at 23, 25 and 24%, respectively) in the initial, CC and GG management systems, respectively. Pyrophosphate and adenosine diphosphate accounted for lower contributions in the three samples ( Table 1 ). myo-IHP, scyllo-IHP, α-glycerophosphate, β-glycerophosphate and adenosine monophosphate were identified in the monoester region ( Table 1 and Figure 8 ). The contribution of the last three represented 56%, 61% and 62% of the phosphate monoester, respectively.

C, N and P stocks were analysed by focusing on changes in soil P lability observed over 13 years of cultivation (2005-2018) under contrasting soil management systems in soil presenting a high initial P stock (3.2 Mg ha^-1, on average), without supplementary P fertilization. The C and N stocks in the 0-30 cm topsoil layer were differentiated by management systems: the C and N stocks increased in GG and decreased in CC. The constant input of plant residues (leaves, stems and roots) and lack of soil disturbance were fundamental factors for the increase in C and N stocks under GG. Similar results were recorded for the 0-10 cm layer in the same experiment (25, 50). Soil ploughing in surface soil layers under tillage management favours SOM mineralization and C and N losses to the atmosphere by potentiating microbial activity through greater residue-soil contact, system aeration and SOM physical protection reduction (22, 26, 51, 52). In addition, grassland-managed soils have significant organic matter input, mainly through underground rhizodeposition, which changes the soil C stock (53, 54).

The total P stock decreased in both treatments because of P export by plants (forage or grain) without P input due to fertilization ( Figure 4 ). This long-term negative balance can lead to great P stock reduction, mainly in P available in the soil, which may constrain plant nutrition (32, 55, 56). Despite the total amount of exported dry matter was similar in the two management systems (112 and 110 Mg ha^-1 for CC and GG, respectively), the amount of exported P differed between treatments due to the nature of the exported plant material. Grains were harvested (maize–wheat-barley) at a mean P concentration of 0.23% in CC, whereas leaves and stems of grasses (D. glomerata, F. arundinacea and L. perenne) were exported as forage at a mean P concentration of 0.28% in GG. This finding partly explains the difference of 27 kg ha^-1 in the total exported P and the subsequent reduction by 0.3 Mg ha^-1 in soil P in GG after 13 years.

Although total P stock depletion was more evident in the system managed with perennial pasture (GG), it was slow in both systems because annual dry matter export was low; it was 19.1 and 21.0 kg ha^-1 y^-1 in CC and GG, respectively, despite the 3-4 mowings per year and plant shoot material removal from the perennial pasture. However, according to the literature, nutrient export by plants in grazed systems (e.g., not studied here), tended to be lower, as some of the nutrients exported by plants were returned to the soil through animal urine and dung (57, 58), which is not the case in mowed grasslands. Although P exportation was low overall, it had a stronger impact on soils presenting low P stock, mainly due to low P stock availability.

The Pi stock decreased in both treatments simultaneously with the total P stock reduction. This process resulted from plant uptake of inorganic P and from further P export (2) or return to soil organic P forms. The Pi stock decrease in GG stood out ( Figure 4 ) because, in addition to Pi exportation, there was Pi transformation and accumulation into Po. The increase of 42% in Po indicates the high potential of grassland to transform Pi into Po. This transformation occurs when P is absorbed by plants; incorporated into essential plant metabolism; returned to the soil through leaves, stems and roots; and remains in the soil as organic matter (59).

Po accumulation in grassland is the consequence of an increase in SOM due to organic matter input to the soil and to nonploughing during the experimental period (25). Overall, pasture soils have increased organic P forms and C and N due to SOM accumulation (27, 29). SOM represents the coupling of P, C and N cycles in the soil organic phase (Rumpel et al., 2015). Moreover, Pi incorporation into organic molecules in rangelands or croplands can be larger if mining plants are used (e.g., Lolium sp. and Lupinus sp.), as well as a mix of plants or rotation, with more nutritionally demanding plants (i.e., leguminous crops) (60; Crème et al., 2016; 55).

The C:N:Po atomic ratio was also changed by the management systems because of changes in C and Po stocks ( Figure 7 ). The stoichiometry values (104:10:1 and 86:8:1 for CC and GG, respectively) were below the means of 166:14:1 (61) and 169:11:1 (62) recorded for pasture and above the mean of 64:5:1 (62) observed for cropland values reported in the literature for bulk soil at a global scale. The lower values observed in the present study were mainly caused by high soil Po contents, which resulted from the history of P input before the experiment and from the soil Po stock increase caused by permanent grassland cultivation. The temporal reduction in the C:N:P stoichiometry reveals that there is no decrease in OM input and thus in the amount of P even after 13 years without P fertilization, suggesting that there is internal P turnover (see below). Alternatively, a limitation of N in the SOM can be addressed when there is a reduction in the atomic N:P ratio. However, this likely did not occur in our treatments, as there were increases in N stocks over time.

A decrease in total P stocks was first observed through changes in labile and moderately labile P stocks, since turnover in these fractions can last from minutes to months (13). Moreover, labile P is the most available P fraction for plant uptake. Based on the current results, decreases of 0.13 and 0.17 Mg ha^-1 in labile P recorded for CC and GG, respectively, represented most of the reduction in total P stocks (mean of 55%); P was higher in GG due to greater P export under this soil management system. With respect to the total Pi and Po stocks, labile Pi and Po presented opposite patterns. Labile Pi decreased in both treatments due to the absorption of inorganic P forms (i.e., HPO 4 − 2 and H 2 PO 4 − ) (2). On the other hand, labile Po significantly increased under the GG treatment due to SOM accumulation in the soil. These results are consistent with the total C, N and Po soil stock increases ( Figures 3 , 4 ) and indicate an increase in C, N and Po coupling in the P pool, which showed greater availability. Labile Po corresponding to P from organic compounds was changed as a result of soil management systems (and thus via soil microbial biomass, soluble organic molecules, and organic orthophosphates less adsorbed by soil colloids). Therefore, it was possible to identify more significant changes in the labile P pool over the analysed period (13 years).

The mean decrease observed in the moderately labile P (0.13 Mg ha^-1) represented the remaining fraction of the total P stock reduction (47%, on average). This behaviour was similar in both soil management systems and showed the buffering effect of moderately labile P on labile P (12). These authors highlighted that Po presenting moderate lability (extracted by 0.1 mol L^-1 NaOH) in cropping systems with tillage accounts for a higher contribution (94%) to labile Pi buffering as ploughing enhances mineralization. However, this effect was not observed in the current research, since moderately labile P reduction was similar in both treatments and moderately labile P reduction was not significant in the CC treatment (data not shown).

The nonlabile P stock was not changed by the soil management systems throughout the evaluated 13-year period. Nonlabile P extracted with 0.5 mol L^-1 NaOH represented a more recalcitrant organic (inside soil aggregates) and inorganic P pool (inner sphere absorbed through iron oxihydroxides and clay minerals) (63). Therefore, there was low sensitivity and minor P pool turnover, regardless of the adopted soil and management system (13). Furthermore, the high labile and moderately labile P stocks that had a history of fertilization were able to fulfil crop demand without the need to access the nonlabile P pool.

The extraction solution NaOH + EDTA extracted more than 50% of the total P from the soil. This observation is common in soils from temperate regions, given soil mineralogy and moderate pedogenic evolution. The soil in the present study was classified as Cambisol with loamy texture in the upper soil horizon with high P content, mainly Pi (67%, 62% and 49% of total P in Pi in the initial, CC, and GG management systems, respectively), due to the history of phosphate fertilization used prior to the beginning of the experiment. Doolette et al. (64) extracted 21% to 89% of the total soil P by using the same NaOH + EDTA solution (with prevalence > 60%) in a study carried out with soils presenting different uses and textural characteristics in New South Wales, South Australia and Tasmania. Abdi et al. (65) extracted 51% of the total soil P, on average, from soils collected in Canada (from temperate regions), presenting clayey texture and cultivated with grassland. On the other hand, Rheinheimer et al. (66) found much lower P recovery (15.1%, on average) in subtropical soil. These authors attributed the P recovery rate recorded for subtropical to its clayey texture and a mineralogy dominated by soil kaolinite and iron oxides.

Similar to other current studies, and in compliance with all previously reported cases, in this study there was a prevalence of orthophosphate and monoesters and a small contribution of diesters and pyrophosphate in the analysed extract. More complex organic compounds, such as myo-IHP, scyllo-IHP, α-glycerophosphate, β-glycerophosphate and adenosine monophosphate, were identified in the spectrum shift attributed to the monoester. These compounds have also been identified in soils cultivated with annual crops and grassland in temperate regions (65, 67, 68). The present study stands out for its evaluation of the temporal effect of the conventional crop rotation system and permanently mowed grassland on the chemical nature of soil organic P. This approach contemplated the need to analyse the temporal change in organic P forms, as evidenced by Cade-Menun (68).

The permanent cultivation of cropland and grassland for 13 years did not significantly change the orthophosphate concentration. The total Pi stock was identified, despite the decrease in the total P stock ( Figure 4 ). Likewise, no relevant temporal increase in organic P forms in GG was observed through ³¹P-NMR (extracted by NaOH + EDTA), most likely because of the high initial total Po stock and C and N contents. Moreover, no relevant changes in Po forms between GG and CC were identified, mainly because the Po rate in the NaOH + EDTA extract was similar between treatments (GG and CC - 27%, on average). Pi extraction selectivity was observed for the initial, CC and GG samples. Lower Pi rates (67%, 62%, 49% of Pi in the initial, CC and GG samples, respectively) were observed in the soil than in the analysed NaOH+EDTA extract (73%, 73%, 74% of Pi in the initial, CC and GG samples, respectively). Selectivity for Po compounds was more often observed with the use of the NaOH + EDTA extraction solution; this was also reported by Rheinheimer et al. (66) for soils from subtropical regions.

Dry matter was removed annually on the order of 7.9 Mg ha^-1 y^-1 in the GG treatment; this reached 110 Mg ha^-1 within 13 years. Dry matter export from pasture tends to reduce Po content in the soil in the long term, mainly the monoester forms detectable through ³¹P-NMR. Boitt et al. (32) found decreased rates of monoester compounds in the NaOH+EDTA extract when they simulated pasture dry matter grazing and exporting for 20 years in temperate pastures in New Zealand. Similarly, Deiss et al. (67) observed decreased monoester compounds in NaOH+EDTA extract due to animal grazing in experiments that lasted 7 to 12 years in winter forage under a subtropical climate in Brazil.

The main changes resulting from soil cultivation with different management systems were decreased α-glycerophosphate, diesters, pyrophosphate and adenosine diphosphate forms and increased myo-IHP and adenosine monophosphate forms in both treatments ( Table 1 ). Significant differentiation between treatments was observed only in monoester compounds ( Table 1 ). Therefore, cultivation with grassland had the strongest effect on α-glycerophosphate reduction and on myo-IHP and adenosine monophosphate increases. Higher values of α-glycerophosphate and lower values of myo-IHP and adenosine monophosphate in permanent crop rotation are linked to greater availability of substrate in the soil for microbiological development (69). In this case, soil rotation favours the contribution of organic material and its incorporation into the soil; consequently, the microbial activity is high. In this case, there is marked production of α-glycerophosphate, with temporal accumulation of myo-IHP and adenosine monophosphate.