Three soils were sampled on the Kohala climatic gradient in Hawaii, spanning a wide range in soil mineralogy, biological activity, and P availability (26). The three soils are all from grassland sites and correspond to three climatic domains, arid (soil D1), sub-humid (soil D2), and humid domain (soil D3), where P cycling was analyzed under steady-state conditions in previous work (27, 28, 31). Mean annual precipitation (MAP) on D1 is ~275 mm, on D2 ~1578 mm, and on D3 ~3,123 mm (36). Going from arid to wet, primary minerals, e.g., apatite, are lost while amorphous secondary minerals and soil organic matter accumulate (28, 37). With increasing rainfall, biological activity also increases, as reflected by microbial biomass, soil respiration, and potential phosphatase activity (27, 31). Soil on D1 was previously classified as Clayey-skeletal, isotic, isohyperthermic Sodic Haplocambids or medial-skeletal, ferrihydritic, isohyperthermic Typic Haplotorrands, depending on the exact values of density, oxalate ex-tractable Fe and Al, and P sorption parameters; soil on D2 as medial, ferrihydritic, isothermic Andic Haplustolls or medial, amorphic, isomesic Humic Haplustands, depending on the expression of a Mollic epipedon and Andic properties; and soil on D3 as Medial, amorphic, isomesic Hydric Fulvudands and Alic Epiaquands, depending on subtle differences in landscape curvature [see (37) for details]. In a 34-day soil incubation experiment at 60% max. water holding capacity, daily soil respiration increased 10-fold from 7.6 in D1 to 73.5 mg C g⁻¹ soil day⁻¹ in D3 (31). Phosphorus availability is lowest on the wettest site, where leaching drives P losses and P sorption capacity is highest (27).

Soils were sampled in February 2016 and in February 2017 from the A horizon of the three sites. The soil samples were kept at room temperature during shipment to Switzerland (1 week) and then stored at −20°C until the beginning of the experiment in February 2018. The combination of soil samples from two sampling dates was necessary to meet the soil dry mass requirements for the experiment (31).

Soil pH was measured in water using a soil:solution ratio of 1:2.5 and a 24-h equilibration time (38). Total soil C and N were measured on an TC/EA (Temperature Conversion Elemental Analyzer; Vario Pyro Cube, Elementar, GmbH, Germany). To assess changes in soil mineralogy, Fe, Al and Si concentrations were measured in extracts of sodium pyrophosphate, oxalate ammonium, and dithionite-citrate-bicarbonate (39). In general, pyrophosphate tends to extract metals associated to organic matter; oxalate ammonium extracts metals associated to amorphous compounds such as allophane, imogolite and ferrihydrite; and dithionite-citrate-bicarbonate extracts metals from crystalline phases (40). All soil properties were measured with four replicates.

For both the ³³P and the ¹⁸O experiments, soils were preincubated for 17 days at 60% maximum soil water holding capacity. Maximum water holding capacity was 0.62 g water g⁻¹ dry soil for D1, 1.18 for D2, and 3.36 for D3 (31). After preincubation, soil was divided into 30 g dry mass equivalent experimental units and labeled with isotopic tracer (Figure 1). The ³³P experiment had four replicates, while the ¹⁸O experiment had two replicates. Because of low resin-P concentrations in D3, more soil was needed to provide enough phosphate for purification (31). At the same time, we wanted to preserve 30 g dry mass experimental units to make the results comparable between soils. To meet these requirements, in the ¹⁸O experiment the D3 soil treatments were carried out with three experimental units in parallel. At each sampling time point, soils from the three sub-replicates were combined to make one replicate, yielding enough P for purification.

The experiment had two treatments: control, where soils were kept constantly moist at 60% max. water holding capacity, and DRW (drying-rewetting). Soils under DRW were wetted to a 60% max. water holding capacity and then subjected to drying until reaching 10% max. water holding capacity (corresponding ~ to 0.06, 0.12, and 0.34 g water g⁻¹ dry soil for D1, D2 and D3, respectively). At these water contents, the three soils were below their respective permanent wilting points (0.28, 0.75, 1.23 g water g⁻¹ dry soil for D1, D2, and D3, respectively; 37). The time to reach 10% max. water holding capacity was 9 days for D3, 4 days for D2 and 2 days for D1. To make results among the soils comparable, DRW cycles were standardized to 9 days for all soils. Once a soil reached 10% max. water holding capacity, a lid was placed on the container to prevent further drying until the start of the next cycle. After 9 days, the drying time of D3, all soils were rewetted. This was repeated for 10 cycles for the ³³P experiment and for 17 cycles for the ¹⁸O experiment, giving an experiment duration of 90 and 153 days, respectively. Soils were sampled the day after rewetting. The ³³P experiment was shorter due to the relatively short half-life (25 days) of ³³P. Experimental units were arranged in a random complete block design in an incubation oven set at 27°C.

Phosphorus exchange was studied for control and drying-rewetting by labeling soil with ³³P as phosphoric acid (carrier free, meaning that the tracer added no measurable mass of P) and measuring the incorporation of radioisotope into different pools. Labeling amount (R) was 29.2 kBq g⁻¹ dry soil for D1, 43.7 kBq g⁻¹ for D2, and 59.9 kBq g⁻¹ for D3.

At each time point, a modified Hedley extraction was used to determine the concentration of P in each of the following operationally-defined inorganic and organic P pools: resin-P, microbial P, NaOH-Pi, NaOH-Po, and HCl-P (27). First, P was extracted with anion exchange membranes (BDH-551642S, VWR International) with or without 0.54 M 1-hexanol, the residue of the hexanol treatment was then extracted with 0.25 M NaOH and 0.05 M EDTA, and finally in 1 M HCl. The sorption-corrected difference between P extracted with hexanol and with resin only was taken as a proxy for microbial P (41, 42). In each fraction, inorganic P was measured by the malachite green method (43). In the NaOH pool, the organic fraction (NaOH-Po) was determined by subtracting the inorganic fraction (NaOH-Pi) from the total P of the pool, determined after digestion with 2.5 M H₂SO₄ and 0.18 M K₂S₂O₈ at 110°C for 60 min (44). Because even after centrifugation soil residue contained significant amounts of solution, ³¹P and ³³P carry over from the NaOH to the HCl extraction was corrected for using mass-balance.

The incorporation of radioisotope into P pools was determined by measuring the recovery of ³³P in each pool. Radioactivity in the extracted P pools was measured by liquid scintillation using a beta-emission counter (Tri-carb 2,500 TR, Packard Instruments, Meriden, CT, USA) after thoroughly mixing the samples with Ultima Gold or Ultima Gold AB for acid extracts. In a pre-test, three methods mentioned in the literature for separating radioactivity in organic and inorganic fractions of NaOH were evaluated: anion exchange resin-trap (45), isobutanol (46), and acidification-centrifugation (47). Since acidification-centrifugation yielded the highest recovery of ³³P spike, this method was used in this experiment. Despite having higher recoveries than the other methods, there was still a considerable quenching effect (48), which was corrected for using soil-specific quenching factors calculated in the pre-experiment (Supplementary Table 1).

For all pools, recovery of radioisotope [%] was calculated as,

where r(t) [Bq g⁻¹ soil] is the amount of radioactivity in the sample at time t and R is the total labeling amount, corrected for radioactive decay.

Specific activity sets recovery of isotope in relation to the P pool size and was calculated using Equation 2 (42). The recovery of ³³P is divided by the label amount (r(t)/R) in order to account for soil specific labeling. This is because best-practice in radiosafety requires only using the minimum amount of radioactive label still giving a signal, which is dependent on soil P sorption properties. Hence, in this study, as in earlier work, SA has the units [% of ³³P (mg kg⁻¹)⁻¹] (42, 49, 50).

Treatment impact on SA_(t) of a pool was taken as an indicator for treatment effect on P exchange that pool. If specific activity changes faster under DRW than control treatment, this would imply that DRW accelerates cycling of P in that pool. For several replicates, specific activity could not be calculated due to either negative recovery or too small P pool size.

The effect of repeated DRW cycles on biologically driven P cycling was studied by tracing the incorporation of O from enriched water into phosphate for control and drying-rewetting treatments.

Soils from both treatments (control and DRW) were labeled with 0.6, 1.1, or 3.0 ml water per experimental unit with a δ¹⁸O_H2O of 387‰ (labeling volume depended on soil water holding capacity). The labeled water was prepared by diluting 98 atom % ¹⁸O (Sercon Limited, UK) with double distilled water. The DRW treatment received additional labeled water (δ¹⁸O_H2O of 52‰) with every rewetting. Soil water was sampled shortly before the first labeling and at every sampling time point, and the isotopic signature of water was measured after quantitatively extracting water with cryodistillation (51).

Oxygen stable isotope ratios in phosphate were measured in the resin, hexanol, and HCl-extractable pools. The pools were extracted in the same way as for the ³³P experiment, just with larger soil amounts, to meet the phosphate needs for purification. Phosphate in the resin, hexanol and HCl extracts was purified to Ag₃PO₄ (52), which was measured in three analytical replicates using a TC/EA (Vario Pyro Cube, Elementar GmbH) in pyrolysis mode, coupled in continuous flow to an isotopic ratio mass spectrometer (Isoprime 100, Elementar GmbH). Two benzoic acid standards (IAEA 601: δ¹⁸O = 23.1‰, IAEA 602 δ¹⁸O = 71.3‰), an internal Ag₃PO₄ standard (Acros Organics, Geel, Belgium, δ¹⁸O = 14.2‰), and in-house made standards were used for instrumental drift correction and calibration. The δ¹⁸O_P of microbial P was determined by mass balance considering the pool concentrations. If the difference between δ¹⁸O_P of resin-P and hexanol P was less than twice the standard deviation of the analytical replicates, then δ¹⁸O_P of microbial P was set equal to δ¹⁸O_P of hexanol P. In the HCl extraction, we controlled for inorganic hydrolysis of the P-O bond in phosphate or any organic phosphate compound carried over from previous steps by using ¹⁸O-labeled and unlabeled HCl extractants (52). Since there was no systematic difference between the two splits, data did not need to be corrected.

Even though the experiment was conceived to have enough phosphate yield for purification, resin-P concentrations decreased throughout the course of the experiment, so that for later time points phosphate yield was too low to measure δ¹⁸O_P for some experimental units. For three samples (D1 res B control t27; D2 res A control t27; D2 Res B DRW t27) we suspected organic matter contamination in the Ag₃PO₄. Accordingly, these samples were cleaned with H₂O₂ prior to analysis.

We compared the change in ¹⁸O in phosphate that we measured with the change in ¹⁸O in water to determine the relative exchange of oxygen in phosphate (Equation 3).

Where f is the value at the time of sampling and i is the value at the beginning of the experiment (before labeling). Since only biological processes affect O incorporation into phosphate, the higher the % of O exchanged in phosphate, the higher the proportion of biologically mediated P cycling. If treatment affected % of O exchanged between water and phosphate, this would suggest that there was a treatment effect on biological P cycling. While it is more common to interpret δ¹⁸O_P relative to a theoretical equilibrium of the isotopic composition of O in phosphate with O in water (53), this approach was not deemed appropriate for this experiment where soil water was constantly changing due to drying-rewetting. While the isotopic signature of soil water was only measured during sampling (shortly after re-wetting), the changes in δ¹⁸O in water during drying is predicted from earlier work on the same soils (Supplementary Figure 1; (54)). Measuring % of O exchanged in phosphate allowed us to compare results from different soils, which because of different labeling amounts and drying times, had different O isotopic ratios in water.

Since different soils had different drying times, the three soils were statistically treated as separate experiments. For each soil, significance of differences in P concentrations, ³³P recovery, specific activities, pH, total C and N, and extractable metals was tested using one-way ANOVA followed by Tukey's Honest Significant Difference Test (55). One-way ANOVA was used since sampling was destructive. For each linear model, we carefully performed model diagnostics by visually examining residuals vs. fitted values to test the assumption that the expected error = 0, normal q-q plot to test the assumption of normality, scale-location plot to test homoscedasticity (equal variance) and constant leverage to test the effect of potential outliers. All statistical tests were conducted at the significance level p < 0.05.

Soil property response to repeated drying-rewetting differed by soil type. The largest pH effect was observed in D3, the soil from the most humid site on the climatic gradient, where pH increased from 4.56 to 5.30 in the control, while it dropped to 3.94 under DRW (Table 1). Carbon concentration in the soil tended to decrease for both treatments compared to t0, though this effect was not statistically significant. Pyrophosphate-extractable Al, Fe, and Si tended to be higher in DRW than control for D1, though this too was not significant. There was no significant effect of DRW on pyrophosphate-extractable metals in D2. In D3, pyrophosphate-extractable Fe and Si were lower under DRW. Oxalate-extractable Al responded differently for the different soils. While in D1 DRW increased oxalate-extractable Al, it decreased in D2 and tended to decrease in D3. There were no significant differences in oxalate-extractable Fe, dithionite-citrate-bicarbonate (DCB)-extractable Al, and DCB-extractable Fe between DRW and control for any soil. However, there was a trend to lower DCB-extractable Si in D1 and significantly lower DCB-extractable Si under DRW for D3.

Concentrations of P in the resin pool were highest in D1 and lowest in D3. Concentrations of resin-P decreased for all soils during the course of the experiment, though only significantly for D2 and D3 (Table 2, Figure 2A). Recovery of ³³P and specific activity of the resin pool decreased significantly with time for all soil and treatment combinations (Figures 2C,E). Despite large differences in pool sizes between the different soils, specific activity of resin-P in all soils was similar, ~0.10–0.17 % (mg P kg⁻¹)⁻¹ 90 days after labeling. DRW treatment significantly affected specific activity for both D1 and D2 (Table 2). While specific activity of resin-P decreased in D1 compared to the control, for D2 (and D3, though not significant), DRW led to an increase.

Concentration of microbial P was highest in D3 and decreased under DRW compared to control treatment (Figure 2B). Microbial P concentration in D1 and D2 DRW treatment at later time points was not significantly different from 0. Recovery of ³³P in microbial P also decreased with time, hence specific activity did not show significant treatment or time effects (Figure 2D). While specific activity in microbial P displayed high variability, in general specific activity was in a similar order of magnitude as that of resin-P (0.10–0.20 % (mg P kg⁻¹)⁻¹. For D2 there was a significant time × treatment interaction effect on specific activity because under DRW microbial P decreased while under the control it stayed constant or increased slightly.

Concentrations of NaOH-Pi were ~2000 mg P kg⁻¹ for D1 and D2 and ~500 mg P kg⁻¹ for D3. Overall, NaOH-Pi showed no clear trend with time or treatment (Figures 3A,D,G). For all time points, NaOH-Pi was the main sink of ³³P in D1 and D2, with 50–65% of added ³³P recovered in this pool. Recovery of ³³P in NaOH-Pi of D3 was lower, ~20–25%. Despite high recovery, specific activity was lower than for resin or microbial pools, at ~0.03–0.05% (mg P kg⁻¹)⁻¹. Specific activity was not significantly affected by treatment or time for any soils (Table 3).

Concentrations of NaOH-Po were higher in D2 and D3 than in D1. Concentration of NaOH-Po was significantly affected by time in D1 and D2 (Table 3). However, there was no consistent decrease or increase of NaOH-Po concentration with time in any of the soils (Figure 3B). Recovery of ³³P in the NaOH-Po pool displayed high variability between replicates, but in general was low (<6 %) (Figure 3E). Though also displaying high uncertainty, specific activity of NaOH-Po in D2 and D3 tended to increase with increasing time (Figure 3H). For D3, specific activity of NaOH-Po was the lowest for all pools measured [~0.01 % (mg P kg⁻¹)⁻¹].

Concentration of HCl-P was ~2000 mg P kg⁻¹ for D1 and ~500 mg P kg⁻¹ for D2 and D3 (Figure 3C). These concentrations were significantly affected by time in D1 and D2 (Table 3). However, like for NaOH-Pi and Po, there was no consistent decrease or increase of concentration observed with time. Recovery of ³³P in HCl-P was low (<5%) in D1 and D2, but considerably higher in D3 (15–30%) (Figure 3F). Specific activity of HCl-P increased with increasing incubation duration for all soils (Figure 3I). For D1, specific activity in HCl-P had the lowest specific activity of any pool [<0.02 % (mg P kg⁻¹)⁻¹].

Total recovery of ³³P in D1 ranged from 73–86%, in D2 from 67–75%, and in D3 from 53–70%.

Stable oxygen isotopic value in soil water (δ¹⁸O_H2O) increased from −1 to −6‰ prior to labeling to +9–+15‰ in the control treatment and to +44–+54‰ in the DRW treatment (Supplementary Table 2). Water in the DRW treatment was heavier because of evaporation during the drying phase and because this treatment was continuously rewetted with ¹⁸O-enriched water.

δ¹⁸O_P values of all treatments and soil combinations in all three analyzed pools increased with time. The δ¹⁸O_P in resin-P increased from ~20 to 30‰ for all soils and treatments (Supplementary Table 2). The δ¹⁸O_P in microbial-P showed a similar pattern to resin-P for D1 and D2 soils. In D3, δ¹⁸O_P in microbial-P in the control treatment tended to be below δ¹⁸O_P in resin-P. δ¹⁸O_P of the HCl pool increased during the course of the experiment. After 153 days of incubation, δ¹⁸O_P in D1 had increased by about 4‰, D2 by ~6‰, and D3 by ~5–7‰. As for resin and microbial δ¹⁸O_P, there was no clear treatment effect on δ¹⁸O_P in all three analyzed pools.

Despite having similar δ¹⁸O_P values, the amount of O exchanged between water and phosphate was higher in the control relative to DRW treatment. Since water in the DRW treatment was heavier, the same δ¹⁸O_P value implies less actual exchange of O in phosphate (Equation 3). In resin-P, up to 50 % of O in phosphate was exchanged in the control treatment, compared to ~20% in DRW (Figure 4, left). A similar pattern was observed in the microbial-P pool, where the highest exchange was observed in the control treatment of D2 (85%). However, in D3 the proportion of O exchanged in phosphate in the microbial pool did not differ significantly between treatments (Figure 4, right). Finally, in the HCl pool, the relative proportion of O exchanged in phosphate was ~35-70% lower in the DRW treatment (Figure 5), with the clearest difference between control and DRW in D1.

Recently the sequential extraction method, which is widespread in P research and the basis of most P modeling frameworks, has come under increasing scrutiny (58–60). As a consequence, spectroscopic methods are becoming more widely used as an alternative and as a way to validate extraction procedures. The soils used in this study have already been characterized with spectroscopic approaches and the link between the sequentially extracted pools and the soil composition validated (27, 28). An important validation component is to test the mobility assumptions inherent in sequential extractions, to better understand the dynamics of each pool. Here we trace isotope incorporation into sequentially-extracted pools to better understand P dynamics. Our study confirms earlier findings that NaOH-Pi is the main sink of radioisotope label in the time frame of days-months covered by incubation experiments (45, 50, 56, 57, 61). However, unlike in previous studies, where recovery of ³³P in NaOH-Pi continued to increase even after 34 days of incubation (61), in our study specific activity of NaOH-Pi remained more or less constant through the course of the experiment (Figure 3G). The specific activity measured in NaOH-Pi was very close to a theoretical equilibrium specific activity. A common assumption in isotope exchange kinetics is that after enough time has passed, the radioisotope will be distributed equally among all the soil P pools in proportions relative to the pool size (62). Accordingly, the specific activity of any pool in this equilibrium state (SA_eq) would be (Equation 4),

Where P_sum is the total amount of P involved in exchange processes (the sum of exchanging P pools). If we take P_sum to be the sum of sequentially extracted inorganic and organic P pools, SA_eq equals 0.024 ± 0.001 for D1, 0.027 ± 0.001 for D2, and 0.046 ± 0.002 % (mg P kg⁻¹)⁻¹ for D3. Our calculated theoretical SA_eq closely matched the observed specific activity of the NaOH-Pi pools (Figure 6), suggesting those pools were already in equilibrium at the first sampling (9 days). Hence, analysis of specific activity in NaOH-Pi showed that this pool is more dynamic in soils rich in chemically sorptive components such as amorphous minerals than in other soils investigated in earlier studies, where recovery of ³³P in NaOH-Pi did not plateau.

The fact that NaOH-Pi reached equilibrium before resin-P does not mean that NaOH-Pi turns over faster than the much smaller resin pool. Rather, resin-P, being the fastest to exchange with the pool that is labeled (the soil solution), has the highest specific activity at the beginning of the experiment. This shows the fast exchange of this pool, as resin-P very quickly goes into isotopic equilibrium with phosphate in soil solution, before exchanges with other P pools take place (63). Previous observations have shown that specific activity of resin-P continues to decline during the experiment (42, 56), and our rudimentary extrapolations suggest specific activity of resin-P continues to decline until the radioisotope is equally distributed in the whole system (Figure 6). Our observations suggest that for less labile pools (such as NaOH-Pi), specific activity increases with increasing exchange reactions, and dynamics can only be followed until specific activity reaches SA_eq, after which time influx can no longer be differentiated from outflux. The concept of SA_eq may be interesting to study P dynamics in future radioisotope tracing experiments. For example, if more points were measured and for a longer period, one might be able to extrapolate an equilibration time of P in the system with some confidence and determine how this differs between different soils and/or treatments. However, for longer experimental durations significantly higher labeling doses would be required.

Mean residence time of P in HCl-P has been approximated to be on the order of years to millennia, depending on soil pH (10). Our results confirm the low reactivity of this pool, especially in D1, which has higher pH and is known to contain apatite (28). However, in D3 specific activity of NaOH-Pi and HCl-P were similar, suggesting that these two pools are chemically more similar for high-organic matter, acidic soils. This may be explained by the redistribution of P from the NaOH-extraction and dissolution of Fe- and Al oxides by 1 M HCl (64). Over all three soils, 7–38% of O in phosphate of the HCl-P pool was incorporated from water during the 5-month experiment (Figure 5). Phosphate enters the HCl pool without cleavage of the P-O bond and therefore an increase in O incorporation can only occur through abiotic exchange with biologically cycled resin-P (65). This confirms the ³³P results that HCl-P may contain both minerals that are inert on short time scales, such as apatite, and secondary P forms that are exchanging readily on the time-scale of months. Interpretation of the HCl-P fraction is complicated because depending on soil properties it can contain multiple forms of P, not just apatite (60).

In addition, our study provides information on exchange of P in organic P, which has rarely been tackled (66). We extracted organic P with 0.25 M NaOH and 0.05 M EDTA, the predominant method used for liquid ³¹P NMR speciation of organic P (67) and for stable oxygen isotope analysis of organic P (68). Earlier work has shown that organic P in this pool is not readily-degradable by enzymes (69). While there was large variability in the data within replicates and between time points, the order of magnitude of specific activity of NaOH-Po provides interesting qualitative insights into the dynamics and nature of this pool. In D2 and D3, the specific activity of NaOH-Po was the lowest of all measured pools, suggesting that the organic fraction of this pool is more inert than the inorganic pools studied (Figure 6). This confirms earlier ³³P tracing studies, where the specific activity of NaOH-Po was also very low, comparable to or lower than that of HCl-P (45, 56). It is well known that organic matter is heavily stabilized due to organo-mineral reactions in Hawaiian soils (70–72). Even in the organic horizon of P-poor forest soil, specific activity of NaOH-Po was lower than that of HCl-P (50). This suggests extremely long mean residence time of P in NaOH-Po, likely in a range similar to HCl-P (years to millennia).

While studies on the limitations and opportunities of fractionation techniques have mostly focused on inorganic P forms (58–60), much less is known about limitations of organic-P forms as determined by sequential fractionation. The fast exchange of the inorganic soil P pools (NaOH-Pi and HCl-P) compared to the relatively inert NaOH-Po pool suggests that physicochemical fluxes dominate soil P cycling and P availability in these soils. However, it is likely that some organic P fractions within the large NaOH-Po pool have a shorter mean residence time, but their dynamics are masked by the bulk of organic P, which is stabilized in organo-mineral complexes (73). Analysis of organic P with NMR or stable oxygen isotopes in NaOH-Po extracts needs to consider the relatively inert nature of this pool to avoid drawing false conclusions about P availability and nutrition. The dynamics of P in other operationally-defined organic P pools (e.g., 0.5 M NaHCO₃-extractable) or of single Po compounds need to be assessed to determine if there are meaningful substitutes to NaOH-Po for studying organic P cycling in soil.

Our results suggest that repeated DRW may reduce mean residence time of inorganic P in soil by accelerating physicochemical P exchange processes. A recent study on sandy soil tracing ³³P incorporation into water-extractable and microbial P reported reduced mean residence time of P in microbial-P under DRW as compared to control treatment (25). In our study, we complement these findings by investigating P dynamics in other inorganic P pools and analyzing O incorporation from water in phosphate. Our results show that specific activity was lower under DRW in resin-P of D2 and D3 soils, suggesting faster dilution of ³³P from resin-P into less mobile P pools (Figure 2E). This may be because drying-rewetting disrupts aggregate structure, exposing new sites to exchange P with the soil solution (12). Likewise, more ³³P tended to be recovered in the HCl-P pool, generally considered to be the least mobile inorganic pool, of DRW than control treatments (Table 3). As soil dries, ionic strength increases, and may lead to precipitation of phosphate with cations (74). The DRW effect on specific activity of resin-P and HCl-P thus provides evidence that DRW may accelerate the exchange of inorganic soil P by repeatedly disturbing the soil chemical environment.

Unlike inorganic P, turnover of organic P was likely reduced by DRW, because DRW hampered biological P cycling. Pool size, ³³P recovery, and specific activity of NaOH-Po showed no significant treatment effect, due to high variability between replicates and time points (Table 3). However, several indicators point to reduced biological P transformations. Firstly, soil microbial P decreased in DRW treatments, whereas it remained stable under control treatments (Figure 2B). This confirms earlier studies reporting the negative effect of DRW on soil microbial biomass (23, 24) and slowing down of microbial P turnover (25). ³³P recovery in microbial P followed the same pattern as concentration. Hence, the decrease in microbial P pool size resulted in a concomitant decrease in microbially-mediated P cycling.

Drying-rewetting also reduced the amount of O exchanged between water and phosphate, an effect observed in all three pools analyzed (Figures 4, 5). Although soils were below their permanent wilting point under dry conditions, microbes are expected to be active also at soil water potentials well below [−5 mPa; (75)], thus we expect some O exchange. Since we used the δ¹⁸O of soil water after re-wetting as a reference to calculate exchange, our observation can be considered a conservative estimate. Under dry conditions, soil water may be up to 10‰ higher (Supplementary Figure 1), which would translate to even lower calculated values of O exchanged between water and phosphate under the DRW treatment (Equation 3). The lower amount of oxygen exchanged in phosphate in DRW could either be due to decreased O incorporation associated to increased proportion of phosphate being released by mineralization (34) or to stress-induced inactivity of microbes. It has been shown that under P-limitation, microbes rely more heavily on organic P mineralization (30, 76), which would incorporate only one O atom. It has also been observed that under drought conditions, microbial metabolism requires higher quantities of C, which translates in higher production of extracellular enzymes targeting C compounds (77). Although not targeting organic P, phosphate is often released as a by-product [i.e., 5′-nucleotidase; (78)] and would incorporate one O from water. In our experiment, DRW had lower O incorporation in all three soils, independent of their P and C availability. Rather, D1, which has the highest P and lowest C availability, respectively [Tables 1, 2, Figure 2A in (31)], had the largest reduction in O exchanged, and this soil also experienced the longest dry periods. Hence, the reduced O incorporation was likely caused by stress-induced inactivity or dormancy of a portion of the microbial biomass. When microbes are dormant, exchange with the soil solution is limited and intracellular processes, which could promote a complete exchange of O in phosphate, are reduced to a minimum (79).

D1 had a low water-holding capacity and thus a fast drying-time, 2 days compared to 4 and 9 days, of D2 and D3, respectively. Hence, treatment effect on O exchanged in phosphate might be the largest for D1 because the DRW soils dried out more quickly than D2 or D3. Interestingly, however, the long dry phase did not translate into a treatment effect on ³³P recovery in inorganic P pools (Tables 2, 3), even though physicochemical exchange processes also need water. This may be because aggregate breakdown and changing soil mineralogy (Table 1), as discussed above, increased physicochemical fluxes shortly after rewetting to a point that this compensated for reduced exchange under the dry phase. Thus, the treatment effect on O exchanged in phosphate was less strong in D3 because these soils stayed moist longer. Furthermore, we found evidence for unanticipated anoxic conditions in D3 control treatment. While pH in the DRW treatment of D3 decreased from 4.6 to 3.9, under control treatment pH increased to 5.3 (Table 1). The pH increase under the control treatment came along with a strong sulfur smell, suggesting that the sealed caps led to anoxic conditions and the reduction of Fe³⁺ to Fe²⁺, consuming protons, and sulfate to hydrogen sulfide (80). The unintended anoxic conditions likely resulted because high microbial activity of these soils consumed oxygen in the closed containers. The depletion of oxygen likely restricted biological activity in D3 and explains the smaller treatment effect on O-exchanged in phosphate in D3 compared to D2 and D1.

Another limitation of this study was the low ³³P recoveries in D3. While total ³³P recoveries of D1 and D2 were in the range of those reported in earlier work (50, 56), recoveries of D3 were only 53–70%. Buehler et al. (56) measured total recoveries of 68–90% in Ferralsols from an agricultural field experiment, with lower recoveries on soils with low P availability. D3 had a considerably lower resin-P concentration than the other two studied soils (20 compared to 70 and 90 mg P kg⁻¹), hence total recovery was also negatively related to P availability in our experiment. Part of the incomplete recovery may be due to color quenching or self-absorption of scintillations by soil particles (48), though this effect was at least partially accounted for with our quenching study for the NaOH-extraction (Supplementary Table 1). However, low recovery may also be a sign that the modified Hedley extraction misses large P fractions in organic-matter rich, highly-weathered Andisols. Buehler et al. (56) found additional 0–6% of added radioisotope label in the residual pool, which they extracted by digesting in concentrated H₂SO₄ at 360°C under stepwise addition of H₂O₂. It is possible that the residual P pool may also play an important role in P cycling even on a timescale of weeks to months for the soils studied and may have been a sink of ³³P in our experiment.