Experimental design and plant cultivation
A loess soil was collected from the top 40-cm layer of an undisturbed site (34°51′30″N, 109°19′23″E) in a hilly-gully region at Ansai County, Shaanxi Province on the Chinese Loess Plateau, and used as the substrate for pot experiment in the present study. The pH of the soil was 8.7, and the field capacity was 33%. The concentrations of total N, total P, and bicarbonate-extractable P in the soil were 96 mg kg− 1, 493 mg kg− 1, and 3.3 mg kg− 1, respectively; the concentrations of total potassium (K) and organic C were 1.5 mg g− 1 and 1.6 mg g− 1, respectively. The soil was air-dried and screened through a 2-mm sieve before filling the pots. For each plastic pot (of 12 cm inner diameter and 15 cm height), a plastic bag was lined inside first, then 2.0 kg of the air-dried and sieved soil was filled into the bag; a total of 136 pots were filled in this way. Potassium was added at 50 mg kg− 1 soil as potassium chloride (KCl) aqueous solution to the soil in all pots.
For 68 out of the 136 pots, no P was added to the soil (hereafter referred to as 0P), while for another 68 pots, P was added at 20 mg P kg− 1 soil (hereafter referred to as 20P) as a monopotassium phosphate (KH2PO4) aqueous solution. Then four forms of N, i.e. calcium nitrate (Ca(NO3)2), ammonium nitrate (NH4NO3), ammonium sulfate ((NH4)2SO4), or urea (CO(NH2)2), were added to the soil at different rates. For both 0P and 20P, no N was added (hereafter referred to as 0N) to the soil in four out of the 68 pots, and the four pots without added P and N (0P0N) were treated as the control; for the rest of the 64 pots, N was added at four rates, i.e. 25, 50, 75, and 100 mg N kg− 1 soil (hereafter referred to as 25N, 50N, 75N, and 100N, respectively) as an aqueous solution of Ca(NO3)2, NH4NO3, (NH4)2SO4, or CO(NH2)2. There were four replicates for each N rate of each N form. After addition of the aqueous solution of K, P, and N, the soil in each pot was watered to 60% of the field capacity, then incubated for two weeks in a greenhouse at the Institute of Soil and Water Conservation (34°16′19″N, 108°04′20″E), Yangling, China. In order to obtain a substrate with homogenous distribution of the added K, P, and N, after incubation, the soil in each pot was air dried and screened through a 2-mm sieve once again separately, then mixed thoroughly and filled back to the pot that was lined with a plastic bag inside.
In the middle of September 2019, seeds of alfalfa (Medicago sativa L. cv Golden Empress) were surface sterilized by soaking the seeds in 10% (v:v) hydrogen peroxide (H2O2) for 10 min, then rinsed with deionized (DI) water three times and placed on moist filter paper in Petri dishes to germinate overnight (He et al. 2017b). Sixty seeds were sown in each pot at 0.5 cm depth, and the seedlings were thinned to 50 per pot two weeks after sowing. Soil water content was maintained at 60% of field capacity by weighing the pots and replenishing DI water every three days during the experiment, and no drainage was allowed. Plants were cultivated for a total of 100 days before being harvested in late December, 2019.
Collection of rhizosheath carboxylates and measurement of plant biomass
When plants were harvested at 100 days after sowing, the aboveground parts (hereafter referred to as shoots) of the plants in each pot were severed at the soil surface. The belowground parts (hereafter referred to as roots) of the plants and the soil in each pot were taken out of the pot together with the plastic bag; the roots were gently separated from the bulk soil, and the soil that was still attached to the roots after gently shaking was defined as rhizosheath soil (Pang et al. 2017). For each pot, rhizosheath carboxylates were extracted by soaking and gently stirring about 1.0 g fresh fine roots and rhizosheath soil in 20 mL of 0.2 mM CaCl2 in a glass beaker for 5 min. About 1 mL subsample of the extract was taken and filtered into a 1-mL HPLC vial through a 0.22-µm syringe filter; then a drop of concentrated phosphoric acid (H3PO4) was added to the vial to acidify the extract and prevent microbial degradation of the carboxylates. All extracts were stored at − 20°C until analysis (He et al. 2020).
The roots were soaked, collected and thoroughly washed with tap water to remove the rhizosheath soil as much as possible, then rinsed with DI water and oven-dried at 60°C for 48 h to obtain the dry mass. The roots that were not soaked were treated the same as those that were soaked, but weighed separately to obtain the dry mass. Total root dry mass (RDM) in each pot was calculated as the sum of the dry mass of the roots soaked and those not soaked. Shoots were also oven dried at 60°C for 48 h and weighed to obtain shoot dry mass (SDM). Root mass ratio (RMR) was calculated as the ratio between RDM to the sum of RDM and SDM.
Determination of plant N and P concentrations, and calculation of plant N:P ratios
Concentrations of N ([N]) and P ([P]) in shoots and roots were determined. Each oven-dried sample was finely ground using a stainless pulverizer, about 0.1 g subsample of each ground sample was weighed and digested in a hot sulfuric acid (H2SO4)-H2O2 mixture. The concentration of N in the digestion solution was determined using a Kjeltec 2300 Automatic Kjeldahl Apparatus (Foss, Höganäs, Sweden) (Baker and Thompson 1992), and the concentration of P in the digestion solution was determined using the vanadium molybdenum yellow colorimetric method (Gupta et al. 1993). The N:P ratios in shoots and roots were calculated as the mass ratios of N to P, based on the determined [N] and [P] in shoots and roots, respectively.
Analysis of rhizosheath carboxylates
Carboxylates in the rhizosheath extracts of 0N, 25N, and 100N at both 0P and 20P were analyzed using High Performance Liquid Chromatography (HPLC). The apparatus used included a Waters E2695 HPLC, a Waters 2998 detector, and a Waters Symmetry C18 reverse phase column (Waters, Milford MA, USA). Carboxylic acids including tartaric acid, malonic acid, citric acid, malic acid, succinic acid, and acetic acid were used as the standards. The mobile phase included 20 mM monopotassium phosphate (KH2PO4) and 100% methanol, the KH2PO4 solution was pre-adjusted to pH 2.5 with concentrated H3PO4 and flowed at a rate of 0.6 mL min− 1, and the methanol flowed at a rate of 0.01 mL min− 1. Each sample was run for 13 min, and carboxylates were detected at 210 nm (He et al. 2020). The amounts of carboxylates in the rhizosheath were expressed in mmol per unit dry mass of the roots used for extraction.
For the two P rates (i.e. 0P and 20P), all four forms of N (i.e. Ca(NO3)2, NH4NO3, (NH4)2SO4, and CO(NH2)2), and all N rates except 0N (i.e. 25N, 50N, 75N, and 100N), a three-way analysis of variance (ANOVA) was carried out to investigate the effects of P rate (Pr), N form (Nf), N rate (Nr), the interactions between any two of the three factors (i.e. Pr × Nf, Pr × Nr, and Nf × Nr), and the interaction among the three factors (i.e. Pr × Nf × Nr) on parameters of plant biomass, [N] and [P], N:P, and the amounts of rhizosheath carboxylates. The three-way ANOVA was carried out using the general linear model in the SPSS 25.0 software package (IBM, Montauk, New York, USA), and the effects were determined to be significant at P < 0.05. Because 0N was shared by all four forms of N, it was not included in the ANOVA. The criterion to determine the effect of a treatment to be significant if P < 0.05 has been and still is widely used. However, it is getting increasingly challenged by statisticians, and presenting the effect size is encouraged when discussing the effect of a treatment (Goodman et al. 2019). Therefore, in addition to the effects determined based on the P-values of three-way ANOVA, the effect size of each treatment relative to the control (i.e. 0P0N) was calculated, and the interaction between P and N was determined for each treatment with added P and N. Here, we presented both the P-values of three-way ANOVA and the effect sizes, but relied more on the effect sizes for description and discussion of the results, as low replication or statistical power in experiments may obscure the ability to detect biologically meaningful responses using the P-value criterion, while the effect sizes are the log response ratios representing the proportional response to experimental treatment and tend to be distributed normally (Harpole et al. 2011).
In this study, the effect sizes were calculated according to the method described by Harpole et al. (2011), but with some modifications. Briefly, the effect size of a treatment for a parameter was calculated as the log response ratio, i.e. the log value of the ratio between the value of the parameter in the treatment to that in the control. Therefore, the effect sizes of P and N were calculated as follows:
Effect size of 20P = Ln(20P0N/0P0N)
Effect size of xN = Ln(0PxN/0P0N)
Then the simple addition of the effect sizes of P and N for each treatment with added P and N was calculated as:
Effect size of (20P + xN) = Effect size of 20P + Effect size of xN
and the effect size of the interaction between P rate and N rate was calculated as:
Effect size of (20P × xN) = Ln (20PxN/0P0N)
where x = 25, 50, 75, and 100, respectively, and the effect size of each N rate for each N form was calculated separately. As the effect size of a treatment was 0.05 and − 0.05 when the response ratio was 1.05 and 0.95, respectively, an effect size > 0.05 indicates that the value of a parameter increased by more than 5% compared with the control and the effect was positive, while an effect size <–0.05 indicates that the value of a parameter decreased by more than 5% compared with the control and the effect was negative; an effect size between − 0.05 and 0.05 indicates that the effect was negligible. Furthermore, intuitively, the interaction was considered additive when the effect size of the interaction was equivalent (a less than 10% difference was considered equivalent) to that of the simple addition, while it was considered super-additive when the effect size of the interaction was at least 10% greater than that of the simple addition, and sub-additive when the effect size of the interaction was at least 10% less than that of the simple addition.
Bivariate Pearson correlations were used to determine the correlations between the mean amount of each rhizosheath carboxylate and mean shoot [N], root [N], shoot N:P, root N:P, SDM, RDM, and total plant biomass (the sum of SDM and RDM) in the control and each treatment under 0P and 20P separately. The correlation analyses were performed using the SPSS 25.0 software package, and the correlations were considered significant at P < 0.05.