A database search was conducted until 17 November 2017, to identify published papers that accurately determined soil P compounds through one-dimensional liquid state 31P NMR on NaOH-EDTA extracts. According to McDowell et al. (2006) and Cade-Menun and Liu (2014), we consider as accurate the papers that used an adequate delay time prior to the NMR analysis, thus enabling the production of quantitative data on the NMR instrument. We used two platforms and specific search terms for each one. The first platform was the Web of Knowledge. The following terms were used: “soil* phosphorus or P or 31P* nuclear-magnetic-resonance or NMR* naoh or sodium hydroxide* edta or ethylenediaminetetraacetic” from which 129 results were obtained. The second platform was Google Scholar. The following terms were used: “soil* phosphorus* “nuclear magnetic resonance”* naoh* edta”, which yielded 2190 results (excluding patents and citations).

We followed pre-defined eligibility criteria to consider the papers and then to select or reject them. The first criteria was that only native growth media were considered (manure, pot soil, soil leachate, and sediment samples were excluded). In studies focusing on changing natural conditions, only the control (unchanged) samples were used, (e.g., litter removal in Vincent et al., 2010 was excluded). Next we only considered studies where the one-dimensional liquid state 31P NMR method was used with the following features: (1) NaOH-EDTA extractor without pretreatment (0.5 or 0.25 M NaOH and 0.1 or 0.05 M EDTA); (2) delay times > 2 s (i.e., quantitative data, see Cade-Menun and Liu, 2014); (3) NMR features or explanations according to 31P NMR principles (see Cade-Menun and Liu, 2014); and (4) total NaOH-EDTA extracted P and total P. Top mineral and organic layers were both considered. From selected papers we compiled the following data: total P, total NaOH-EDTA P, and NaOH-EDTA organic P, as well as the Po compounds inositol hexakisphosphates (myo-, scyllo-, neo-, and D-chiro-IHP, when available), deoxyribonucleic acid (DNA), phosphonates, NaOH-EDTA inorganic P, and the Pi compounds orthophosphate, pyrophosphate and polyphosphate. No duplicity was found in the selected papers.

Soil texture, total C, total N and pH, short range ordered Al and Fe minerals (poorly crystalline) estimated with oxalate extraction, climate characteristics (mean annual precipitation and mean annual temperatures), as well as soil age, when available, were also collected from the papers. When the total C was unavailable, the organic C was assumed to be the total C. This assumption only occurred for non-calcareous soils. Some variables were unavailable for some results, and the number of experimental used for each analysis is presented in the results section. Missing texture and total C data (representing 12 sites and 1 site, respectively) were extracted from a global soil dataset, SoilGrids, which is now at 250 m resolution (0–20 cm topsoil, Hengl et al., 2017). The resulting dataset is available in Sect. S1 int the Supplement (Deiss et al., 2018). We used the Whittaker's diagram (Whittaker, 1975) and the “BIOMEplot” package (Kunstler, 2014) to determine the biomes of our sites (Supplement S2).

Soil weathering stages were derived from the soil type according to Cross and Schlesinger (1995) and Yang and Post (2011) as well as from chronosequence positions. A low weathering stage was attributed to Entisol, Mollisols and Inceptisols forming the first stages of chronosequences and gleyed Acrisols. An intermediate weathering stage was attributed to Alfisol, Aridisol, Mollisols and Inceptisols forming the intermediate stages of chronosequences and orthic Acrisol. Finally, a high weathering stage was attributed to Oxisol, Spodosol, Ultisol, and humic Acrisol.

Statistical analyses were conducted on R Version 3.1.0 (©2014 The R Foundation for Statistical Computing) using mixed-regression models including edaphic and climatic variables as continuous and categorical fixed effects. The latitude, the percentage of P extracted with NaOH-EDTA and the soil sampling depth were considered as random effects. Latitude was used to control for the spatial auto-correlation (Maestre et al., 2005). The percentage of P extracted with NaOH-EDTA was used because the NaOH-EDTA extraction process varies according to soil characteristics and experimental conditions (i.e., pretreatment, soil-to-solution ratio and soil characteristics) (Cade-Menun and Liu, 2014, see Fig. S5). Sampling depth was used because of potential differences in organic matter concentration along the soil profile. The bivariate effects of latitude, percentage of P extracted and sampling depth on the soil P composition are presented in Supplement S3–S5.

We used variation partitioning and Venn diagrams (Legendre and Legendre, 2012) (the “vegan” package) to partition the total variation explained uniquely by the matrix of either soil variables, climate variables or soil weathering stages as well as the variation explained by the combined effect of these matrices. The unique effect of soil, climate or soil weathering stages was calculated as the adjusted r2 value (ra2) difference between the full model and unique model. The joint effect of these matrices was calculated as the difference between the summed ra2 of unique models and the ra2 of the full model.

Structural equation modeling (path analysis, the “lavaan” package) was used to explore how variations in Pi and Po compounds are driven by both direct and indirect effects of key environmental drivers (soil, climate, and parent material). We first established an a priori model that is based on our knowledge and is presented in Supplement S10. Then, we tested for the most parsimonious model among many alternative ones, i.e., the one that differed the least from the observations and presented the greater p value. Parent material, which was unknown for our sites, was considered as a latent variable in the model to explain the remaining coordinated variations in total P, clay, and pH variables that were not explained by climate and weathering. As mainly determined by the biota, total C was not considered as being constrained by the parent material.

Different units were used across statistics to analyze soil P composition. The bivariate relationships (Figs. 2, 3, and 5, and Figs. S3–S8) considered are as follows: (i) total Pi or Po concentration in NaOH-EDTA extracts (mg kg-1 soil), (ii) proportion of total Pi or Po as percentage of total NaOH-EDTA P (% of NaOH-EDTA P), and (iii) proportion of soil P compounds as percentage of their respective pools (% of NaOH-EDTA Pi or Po). In contrast, in both Venn diagrams (Fig. 6) and structural equation modeling (Fig. 7) soil P compounds were in mg kg-1. In bivariate relationships, we used percentages was to compare the relative composition of P along environmental variables that are linked with the weathering of soils. In both the Venn diagram and path analysis, the objective was to explain soil P composition either partitioning the variation among state factors or accounting for the causal structure of environment. For that, we used the mg kg-1 unit so that the distribution of our variables was not constrained as a proportion.

All surveyed edaphic properties affected soil Pi and Po pools and compounds. These results are summarized in Figs. 2 and 3. Total Pi (Fig. 2a) and Po (Fig. 3a) concentrations in NaOH-EDTA extracts (mg kg-1 soil) both had a quadratic response to soil pH, with higher values occurring at an intermediate pH, although this effect was constrained to the mineral layers. In mineral layers, the proportion of NaOH-EDTA P in the form of Pi increased with pH (Fig. 2b), whereas the proportion in the form of Po decreased with pH (Fig. 3b). However, there was no pH effect on either pool in the organic layers. The distribution of compounds in both Pi (Fig. 2c–e) and Po (Fig. 3c, d) pools responded dynamically to the pH. In the Pi pool (% of Pi) of mineral layers, orthophosphate decreased, and pyrophosphate accounted for the remaining Pi as the pH decreased. The pH had no effect on these Pi compounds in the organic layer (even though there is an apparent trend, these relationships became insignificant after including sampling depth as a random effect on models; Supplement S3 shows the sampling depth effect over the soil P composition). In the Po pool (% of Po), both inositol hexakisphosphates (mineral layer) and DNA (mineral and organic layers) proportions increased as the pH decreased. Phosphonates response to edaphic properties (insignificant) is presented in Supplement S6.

Both total Pi and Po concentrations in NaOH-EDTA extracts (mg kg-1 soil) responded quadratically to the clay concentration, with higher values occurring at intermediate textural classes (Figs. 2f and 3e). Clay impacted on neither Pi and Po, nor on the proportions of P compounds (% of NaOH-EDTA P) (Figs. 2g–j and 3f–h).

Total Pi and Po concentrations in NaOH-EDTA extracts (mg kg-1 soil) increased as the soil C concentration increased in mineral layers, whereas in organic layers there was no C concentration effect on Pi and Po concentrations (Figs. 2k and 3i). As a percentage of NaOH-EDTA P, Pi decreased and Po increased (% of NaOH-EDTA P) as the soil C concentration increased in mineral layers, and there was no C concentration effect on either Pi and Po proportions in organic layers (Figs. 2l and 3j). In the Pi pool (% of Pi) of mineral layers, orthophosphate and pyrophosphate proportions decreased and increased, respectively, as the soil C concentration increased (Fig. 2m–o). As the soil C concentration increased in the organic layer, orthophosphate decreased, at a greater extent when compared to the mineral layer, pyrophosphate decreased (in contrast to the mineral layer, in which it increased), while the polyphosphate proportion increased, and gradually dominated the Pi pool at greater soil C concentrations. In the Po pool (% of Po), there was no C concentration effect on the soil Po composition (phosphonates, inositol hexakisphosphates and DNA) in either mineral or organic layers (Fig. 3k–l; Supplement S6).

Both total Pi and Po concentrations in NaOH-EDTA extracts (mg kg-1 soil) from both mineral and organic layers increased as the total soil P concentration increased (Figs. 2p and 3m). Only the DNA compound from the Po pool (% of Po) in the mineral layer was affected by the total soil P concentration (Fig. 3p). As the total soil P concentration increased, the DNA proportion in the Po pool decreased. It is important to note that the reported total P (x axis on Figs. 2 and 3) is the one obtained by digestion and it includes both the extracted P and the residual P. The recovery of the total P by NaOH-EDTA extraction varies depending on soil characteristics and laboratory procedures (Cade-Menun and Liu, 2014).

Total Pi and Po concentrations in NaOH-EDTA extracts (mg kg-1 soil) were only affected by the soil C : P ratio in organic layers. Total Pi and Po concentrations both decreased as the soil C : P ratio increased (Figs. 2u and 3q). As a percentage of NaOH-EDTA extract (% of NaOH-EDTA P), Pi decreased while Po increased, both exponentially, as the soil C : P ratio increased in mineral layers (Figs. 2v and 3r). As proportions in the Pi pool (% of Pi) of mineral layers, orthophosphate decreased and pyrophosphate increased as the soil C : P ratio increased (Fig. 2w–x). In the Pi pool (% of Pi) of organic layers, proportions of orthophosphate decreased and polyphosphate increased, gradually dominating the Pi pool as the soil C : P ratio increased (Fig. 2w–y). In the Po pool (% of Po), the DNA proportion increased as the soil C : P ratio increased, only in the mineral layer (Fig. 3t).

Climatic properties affected soil Pi and Po pools and their composition only through the mean annual precipitation. These results are summarized in Supplement S7 and S8. The mean annual temperature, ranging from -0.4 to 27 ∘C, did not promote any change in the soil P composition in natural ecosystems. There was no effect of climatic variables on total Pi and Po concentrations in NaOH-EDTA extracts (mg kg-1 soil) (Supplement S7a and S8a). As a fraction of the NaOH-EDTA extract (% of NaOH-EDTA P), Pi decreased and Po increased as the precipitation increased (Supplement S7b and S8b). As the precipitation increased, proportions of orthophosphate decreased and pyrophosphate increased as compounds of the Pi pool (% of Pi) (Supplement S7c–d).

Soil weathering stages determined from the soil type and chronosequence positions affected soil age and C : P ratios following an expected effect of pedogenesis (Fig. 4). As soil weathering stages increased, the soil age and C : P ratio also increased. Pi and Po pools were both affected by the soil weathering stage (Fig. 5). Total Pi and Po in NaOH-EDTA extracts (mg kg-1 soil) were more concentrated in soils at moderate weathering stages when compared to low and high weathering stages (n=79, Fig. 5a, f). As percentages in the Pi pool (% of Pi), orthophosphate decreased and pyrophosphate increased as the soil weathering stage increased (n=79 for all Pi compounds, Fig. 5c–d). In the Po pool (% of Po), the DNA (n=64) proportion was greater in more weathered stages, and there was no effect of weathering stages on phosphonates (n=79) and inositol hexakisphosphates (n=52) proportions (Fig. 5h–j). Using available data (n=49), we observed no effect of soil weathering stages on short range ordered (poorly crystalline) Al and Fe minerals estimated with oxalate extraction (p > 0.1, Supplement S9).

The variation partitioning of ecosystem properties governing the soil P composition (in mg kg-1 soil) was generally more pronounced for soil variables (pH, clay concentration, and total P and C concentrations) than climatic variables (precipitation and temperature) and soil weathering (Fig. 6). For the total Pi concentration and its compounds orthophosphate and pyrophosphate, the total variation explained by models ranged from 46 % to 89 %, and they were mostly explained by soil variables and combined effects of soil and weathering. Polyphosphates had a poorly defined response to the variation partitioning of ecosystem properties (< 0.01 % of the total variation explained).

In the Po pool, the total variation explained by models ranged from 41 % to 86 % (Fig. 6). The total Po, inositol hexakisphosphates and DNA had their total variation mostly explained by soil variables, and to a lower degree, but more pronounced for the DNA compound, by combined effects of soil variables and weathering. In contrast, most of the variation in phosphonates was explained by combined effects of climate and soil variables, followed by uniquely soil variables.

We used path analyses to explore the interdependences between edaphic and climatic variables and how they relate to the soil Pi and Po compounds (Fig. 7; Supplement S10). The parent material was used as a latent variable (set by the pH) in both models (Pi and Po). Climate and soil weathering drivers were independently related to soil variables (total P, pH, clay and total soil C), and soil variables were considered direct effects in the models. The most parsimonious path analysis model explained up to 78 % of Po compounds variation and 89 % of Pi compounds variation.

Following an expected effect of pedogenesis, the path analysis indicated that the parent material (latent variable) was positively related to the soil total P, clay and pH. Greater mean annual precipitation was negatively related to the soil total P, pH, and it positively influenced soil total C, while clay was negatively influenced by precipitation in the Po model only. In the Po model, precipitation promoted soil weathering, whereas in the Pi model, soil weathering was positively affected by temperature. The mean annual temperature positively affected the clay and pH. Soil weathering was negatively related to the soil pH, and positively related to the soil clay and total C. In the Pi model only, soil weathering negatively affected soil total P. There were also significant direct and positive effects between soil total C and clay, and total P, in both Pi and Po models, and there was a positive relationship between soil total C and pH in the Po model only.

In the Pi model, orthophosphate was negatively related to precipitation, and it was positively influenced by soil total P and total C. Pyrophosphate was positively influenced by precipitation, soil total P, and total C. Polyphosphate was negatively influenced by temperature, and it was positively related to soil pH. In the Po model, inositol hexakisphosphates were negatively affected by precipitation and temperature, but positively affected by the soil total P, total C and pH. In contrast, total P and total C positively affected DNA, and there were no effects of climatic variables over DNA. Phosphonates were negatively affected by temperature and weathering, but positively affected by precipitation and soil total C.

As soils aged, pyrophosphate and polyphosphate may have accumulated because of the incorporation and stabilization of these compounds (biological origin) into soil organic matter (Turner et al., 2007). The soil pH, total carbon and C : P ratio, as well as weathering stage had a major influence on the soil Pi pool composition. As the orthophosphate proportion decreased in more weathered, acidic, organic-rich, and P-limited soil environments (Figs. 2c, m, w and 5c), pyrophosphate and polyphosphate proportions increased and dominated the Pi pool (Figs. 2d, n, o, and y and 5d).

Even though pyrophosphate and polyphosphate are Pi compounds, they have a biological origin (Turner and Engelbrecht, 2011). Condensed forms of P (including pyrophosphate and polyphosphates) are found in every bacterial, archaeal, and eukaryotic cell, but in highly variable amounts (Kornberg et al., 1999). Bünemann et al. (2008) found a positive relationship between the proportion of fungi and the amount of pyrophosphate, and Reitzel and Turner (2014) found a positive link between the pyrophosphate proportion and soil microbial P. Polyphosphate can originate from ectomycorrhizal fungi (Koukol et al., 2008), and there are some ectomycorrhizal fungi specialized for P uptake in low P, acidified soil conditions (Wang and Qiu, 2006). Ectomycorrhizal fungi convert the orthophosphate that they take up from the soil into polyphosphates, and translocate the polyphosphate along fungal hyphae, sometimes at a great distance from where the orthophosphate is taken up (Bücking and Heyser, 1999; Plassard and Dell, 2010). Therefore, we believe that pyrophosphate and polyphosphate dominated the Pi pool in acidic, P-limiting (C : P ratio), and high organic matter (total C) soils because of the microbial origin of these P sources, but much information is still needed in regard to plant and microbial communities characterization in studies of P forms. These organisms could have helped to deplete and transform the bioavailable orthophosphate, turning it into more microbial biomass derived P compounds as pedogenesis progressed in these environments.

Moreover, polyphosphates tend to occur in abundance only in soils where decomposition is slowed, such as acidified soil conditions, or cold and wet soils high in organic matter (e.g., Cade-Menun et al., 2000; Turner et al., 2004). Studying wetland soils, Cheesman et al. (2014) found that polyphosphates played a preeminent role in P-limited systems, predominantly in acidic, high-organic-matter systems. Adding to that, pyrophosphate hydrolysis was found to be more rapid with greater biological activity and higher agricultural soil pH (Sutton and Larsen, 1964), and this may have contributed to reducing the pyrophosphate proportion at a higher pH in mineral soils (Fig. 2d). As the C concentration increased in organic layers, polyphosphate dominated the soil Pi pool (Fig. 2m–o) possibly because of its lesser lability when compared to orthophosphate and pyrophosphate. Pyrophosphate is less polymerized and potentially more susceptible to hydrolysis than polyphosphate. According to Savant and Racz (1972), Subbarao et al. (1977) and Dick (1985), pyrophosphate is hydrolyzed more rapidly than polyphosphate because pyrophosphate is an intermediate product of polyphosphate hydrolysis until the final orthophosphate is produced. These are also conditions under which ectomycorrhizal fungi are found. These fungi produce hyphal mats in the forest floor, so an increase in polyphosphates could reflect an increase in ectomycorrhizal hyphal mats. However, caution must be taken when interpreting pyrophosphate and polyphosphate hydrolyzation results from 31P NMR analysis on NaOH-EDTA extracts. Polyphosphates can potentially degrade to pyrophosphates during extraction and NMR analysis of P (Cade-Menun et al., 2006), so they cannot be considered as fully distinct P forms. This potential degradation could be one explanation to why the polyphosphates results were poorly explained by the variation partitioning and structural equation models.

As time passes after the onset of pedogenesis, modifications in the soil Po composition were possibly related to the acidifying environment in soils, which may increase the charge of some Po compounds, and thus increase sorption. Soil pH affects the sorption of Pi and Po compounds by soils, but different P compounds respond differently to pH changes (Shang et al., 1992). Shang et al. (1992) verified that sorption of both orthophosphate and inositol hexakisphosphate by Al and Fe precipitates generally decreased as pH increased, whereas there was little pH effect on the adsorption of glucose 6-phosphate by both precipitates. Sorption of inositol hexakisphosphate to minerals surface is often stronger than orthophosphate, but both tend to be less sorbed to those minerals in neutral to alkaline soils (Berg and Joern, 2006; Xu et al., 2017). The presence of humic acids may affect the amount of inositol hexakisphosphate that sorbs on the minerals surface at lower pH values, but it cannot displace inositol hexakisphosphate from that surface (Ruyter-Hooley et al., 2016). Moreover, another study founds that inositol hexakisphosphate sorption in soils was unaffected by the presence of orthophosphate, β-D-glucose-6-phosphate or adenosine 5′-triphosphate (Berg and Joern, 2006). Following a similar pattern, the amount of DNA bound on clay minerals such as montmorillonite, kaolinite, hydroxyl aluminum species and variable charge soil colloids also increased by lowering pH of solution (Khanna and Stotzky, 1992; Cai et al., 2006, 2008; Wang et al., 2009; Saeki et al., 2010). At pH < 5, protonation of the amino groups of adenine, guanine and cytosine occurs and causes the increase of net positive charge of DNA and electrostatic attraction between negatively charged tetrahedral silica layer on the clay surface and DNA (Cai et al., 2006). Therefore, the increasingly acidic pH could have increased sorption, and thus facilitated inositol hexakisphosphates (Fig. 3c) and DNA (Figs. 3d, 7) accumulation in those soils.

In our study, we found that there was an increasing proportion of inositol hexakisphosphates and DNA in the Po pool (% of NaOH-EDTA P) as pH decreased, and there was predominance of inositol hexakisphosphates in acidic, more weathered soils (Figs. 3c–d, 5j). The hierarchy of investment for P acquisition through enzymatic activity may also be a factor that contributed to modifications in the soil Po composition as time passes since the onset of pedogenesis. According to Turner et al. (2018), Turner and Haygarth (2005) and Kunito et al. (2012), P limitation may stimulate increased phosphoesterase synthesis as a way to increase bioavailable P by the mineralization of Po. Fungi are well known for their capacity to secrete acid phosphatases (Rosling et al., 2016), and are usually the predominant microorganisms in acidic natural soils; while alkaline phosphatase and phytase genes are distributed across a broad phylogenetic range and display a high level of microdiversity (Zimmerman et al., 2013). Plants and microorganisms that breakdown orthophosphate diesters need a higher investment for the P acquisition (see Turner 2008a for the conceptual model) than orthophosphate monoesters, as they require hydrolysis by both phosphodiesterase and phosphomonoesterase to release available phosphate, whereas orthophosphate monoesters require only the latter (Halstead, 1964; Skujins, 1967; Tabatabai and Bremner, 1969). Although inositol hexakisphosphates are also classified as part of orthophosphate monoesters, they need a higher investment in Po acquisition than other orthophosphate monoesters and diesters because they can be strongly bounded by metal oxides, clays and organic matter (Turner et al., 2007). Organic acids are required to desorb inositol phosphates, so that phytases can hydrolyze them to release a free orthophosphate. This would suggest that as soil gets more weathered, inositol hexakisphosphates may accumulate more than DNA, and the latter more than other Po compounds.

Nonetheless, inositol hexakisphosphates can persist as a main Po compound up to a certain point of pedogenesis, and then decline in more weathered soils. Some authors described that inositol hexakisphosphates declined to lower concentrations in older soils of non-tropical regions (Turner et al., 2007, 2014), and most tropical soils had negligible inositol hexakisphosphates contribution (e.g., Turner and Engelbrecht, 2011). In our results, soil weathering had no effect on inositol hexakisphosphates concentrations in mg kg-1 in non-tropical environments (Fig. 7), but in fact, all compiled results from tropical soils had no inositol hexakisphosphates. Inositol hexakisphosphates have been found in very weathered soils (e.g., Oxisols), but under agricultural management that included well-known sources for that P compound such as plant seeds (Turner, 2006; Smernik and Dougherty, 2007; Deiss et al., 2016). These results suggest that inositol hexakisphosphates could occur in tropical soils under native vegetation, but they are either being rapidly turned into bioavailable compounds (by plants and microorganisms) or inputs of inositol hexakisphosphates, which are abundant in seeds and pollen (Raboy, 2007), are lower in lowland tropical forests compared to temperate ecosystems (Turner and Engelbrecht, 2011). In P limited soil environments, the acquisition of inositol hexakisphosphates may be strongly improved by root exudates, which may increase the solubility of these compounds in soil (Gerke, 2015). In addition, mineralization of myo-inositol hexakisphosphate by ectomycorrhizal fungi (Chen et al., 2004) may also have contributed to its decline in more weathered, acidic soils, due to fungi predominance in these environments. Finally we believe that other P compounds such as DNA (Figs. 3d, 5j) and pyrophosphate (Figs. 2d, 5d) will possibly prevail in more weathered systems from tropical regions because they are intrinsic components of the microbial biomass (Kornberg et al., 1999). Mature soils are known for having the microbial P as the main component of the P pool (Turner et al., 2013).

Soil clay concentration affected both soil total Po and Pi concentrations (Figs. 2f and 3e), but had a minor association with soil P compounds. Recent investigations have found that organic compounds stabilization may be mainly driven by factors that are typically minor constituents of the clay-sized fraction by mass, but highly reactive components. Vogel et al. (2014) showed that organic matter is preferentially stabilized in certain hot-spot zones (i.e., rough surfaces), and that only a limited portion of clay-sized surfaces contributed to soil organic matter stabilization. This concept was further tested for P by Werner et al. (2017) who found that microscale spatial heterogeneity influences P accessibility and bioavailability in soil aggregates, depending on soil substrate and depth. They also found that in P-rich areas of soil aggregates, the P was predominantly co-located with Al and Fe oxides, while in low-P topsoil aggregates, most of the P was organically bound (Werner et al., 2017). Yang et al. (2016) also showed that only limited portions of fine mineral surfaces contributed to soil organic matter stabilization. So, these facts could justify the minor association between bulk soil clay concentration and soil P compounds as observed in our study.

Soil organic matter stabilization could be facilitated in more weathered soils by the potential increase in amorphous Al and Fe oxides (Albrecht, 1957; Walker, 1965), and consequently more reactive surface area availability to absorb and stabilize soil organic matter; however, we did not observe a significant overall soil weathering effect on Al and Fe oxide concentrations (Supplement S9), suggesting that it may depend on other factors such as the parent material or specific soil orders. Moreover, contrasting soil organic matter responses to short-range-ordered (amorphous) Fe or Al oxides have been found in the literature. Some investigations found a pronounced role promoted by Al oxides (Heister 2016; Kaiser et al., 2016), whereas others found Fe oxides as the main soil organic matter stabilizing mechanism (Wilson et al., 2013; Catoni et al., 2016; Deiss et al., 2017). Investigations also found no apparent relationship between soil organic matter and both Al and Fe oxides (Cloy et al., 2014; Rumpel et al., 2015; Vogel et al., 2015). Therefore, we could not confirm the role of Al and Fe oxides as influenced by soil weathering stages over the soil P composition.

Climatic variables exerted an important role on the soil P composition but to a lesser extent when compared to soil variables (Fig. 6). Contradicting what we expected, our results showed that temperatures ranging from -0.4 to 27∘C had no effect on both Pi and Po pools and their compounds (Supplement S7 and S8). It was expected that the soil Po concentration would decrease with increasing temperatures because higher temperatures are optimal for the breakdown of the soil Po compounds by the microbial biomass through phosphatase enzymes release (Hui et al., 2013). Hui et al. (2013) confirmed that greater maximum phosphatase activity occurred at incubation temperatures > 25 ∘C when compared to 20 ∘C, but no differences were observed among temperatures greater than 25 ∘C (Hui et al., 2013). Therefore, phosphatase activity may depend on the range and magnitude of temperatures; and our results covered a greater range of markedly lower temperatures, which may reduce microbial activity variability even more due to a slowdown in the microorganisms' metabolism.

In contrast, precipitation affected several variables in the soil Pi pool. This result was also expected based on the classic paper of Walker and Syers (1976), which suggested that pedogenesis depends predominantly on the volume of water leached through soil. In our results, the soil total P concentration, pH (Fig. 7), and orthophosphate proportion (Supplement S7c) were negatively related to precipitation. As precipitation increased (Supplement 7c) and the soil was in a higher weathering stage (Fig. 5c), the orthophosphate proportion (% of NaOH-EDTA P) possibly decreased because of increased leaching. However, caution needs to be used when discussing changes in orthophosphate extracted by NaOH-EDTA for 31P NMR, because NaOH-EDTA will preferentially extract Po rather than orthophosphate. As such, studies that analyzed the residual P after NaOH-EDTA extraction have shown that it is mainly composed by orthophosphate (e.g., Cade-Menun et al., 2005). Feng et al. (2016) evaluated P fractions along a climosequence and observed that greater precipitation (in soils with no impeded drainage) reduced the inorganic concentration of P linked to Ca, corresponding to a marked decline in soil exchangeable Ca and suggesting an enhanced leaching of P along with weatherable cations. Moreover, greater soil water availability, and consequent greater primary productivity, may have increased the demand for P in its bioavailable form and contributed to the orthophosphate depletion.

As the orthophosphate percentage and concentration (Supplement 7c and Fig. 7, respectively) decreased following greater precipitation, the pyrophosphate percentage and concentration (Supplement 7d and Fig. 7, respectively) increased suggesting that this compound predominates under these environmental conditions. As previously described, this may be due to the incorporation of these compounds into recalcitrant soil organic matter (Turner et al., 2007). Moreover, given the microbial origin of pyrophosphate and its association with the microbial P biomass (Koukol et al., 2008; Turner and Engelbrecht, 2011; Reitzel and Turner, 2014), pyrophosphate (Figs. 7, S7d) possibly mirrored the response of the total Po (Fig. S8b) to climatic variables, which may have resulted from greater soil organic matter accumulation following greater productivity (i.e., plants and organisms) in these ecosystems with greater water availability. Evaluating the P budget of the whole ecosystem, Turner et al. (2013) demonstrated the dominance of microbial P in mature soils. Wang et al. (2014) found that greater Po concentrations were associated with increasing biomass production (i.e., primary production and microbial biomass) because plants and microbes incorporate P into biomass and return it to the soil. However, it is important to note that the majority of P in plant biomass is as orthophosphate (e.g., Noack et al., 2012) and not as Po compounds. However, we believe that with higher orthophosphate inputs through plant biomass, soil Po concentrations would increase altogether with orthophosphate P concentrations, and also at expense of soil orthophosphate due to the greater bioavailability to plants and organisms of that latter P compound.

Changes in vegetation are expected to occur during pedogenesis, and climatic variables may govern magnitudes of these alterations along with soil changes. Vitousek et al. (1995) showed that as ecosystems develop, the pattern of P concentration in plants leaves follows a non-linear response to time, in which lower concentrations occur at either early or late stages of pedogenesis, and a maximum is reached at an intermediate stage of pedogenesis. In addition, precipitation can affect the magnitude of that maximum response (intermediate stage), where the P concentration in plant leaves is higher in mesic gradients when compared to more wet gradients (Vitousek et al., 1995). Moreover, as described earlier, the soil available P, along with other climatic variables, governs maximum photosynthetic rates, but a trend that is expected to gradually decline in more weathered soils, due to a lower P availability (Maire et al., 2015). Phosphorus limitation can become sufficiently intense in the late stages of ecosystem development (also known as the retrogressive phase) to cause a decline in forest biomass, and productivity (Wardle et al., 2004). The exception seems to be tropical forests (Turner et al., 2007), which exhibit very diverse tree communities on old, infertile soils (Losos and Leigh Jr., 2004). Moreover, Turner et al. (2018) showed that in lowland tropical ecosystems, P limitation affects individual species, but species-specific P limitation does not translate into a community-wide response, because some species grow rapidly on infertile soils despite extremely low P availability.

Many efforts have been made to explain soil P composition during pedogenesis; however, a clear picture on how specific plant species, plant functional traits, and their communities can influence the soil P composition is still lacking, especially with results obtained with 31P NMR. For example, why are inositol hexakisphosphates not found in tropical soils under native vegetation, i.e., is it because the rapid turnover promoted by plants and/or organisms, or exclusively due to lack of inputs from plants? Do the changes in forest biomass and plant species diversity as soil P turns scarcer contribute to soil P composition in non-tropical environments, either by inputs or P compounds consumption, or the soil per se governs both the soil P composition and vegetation dynamics? Therefore, we point out that studies aiming to disentangle confounding effects among soil biotic and abiotic components, climate and vegetation are required to enable a better understanding of soil P composition in natural ecosystems.

Moreover, studies coupling 31P NMR with other important techniques (e.g., Liu et al., 2013, 2015) could contribute to a better understanding of P cycling and composition in terrestrial ecosystems. A clear understanding of how orthophosphate species such as Fe-, Al- and Ca-phosphate (Hesterberg, 2010; Kizewski et al., 2011) respond to pedogenesis could be elucidated with XANES (see Prietzel et al., 2013; Hashimoto and Watanabe, 2014). Studies that quantify specific-P-related enzymes activity (see Turner et al., 2018) in native vegetation soils could help understand if the hierarchy of investment for the P acquisition actually contributes to different degrees of accumulation of inositol hexakisphosphates and DNA as pedogenesis progresses in non-tropical environments, and if phosphatases are leading to a rapid turnover of inositol hexakisphosphates in tropical environments. This can be achieved through determining the presence and abundance of microorganisms and enzymes, and how these changes affect soil P composition. Turnover, exchange kinetics, and mineralization rates could be assessed using isotopic techniques (see Frossard et al., 2011), and enable a separation between different sources of P compounds, and their dynamics in soils, organisms and plants.

Finally, we expect future research to provide results of as many soil P compounds as they can find rather than broad compound classes only (i.e. orthophosphate diesters and monoesters), even when compound concentrations are low (and describe when main soil compounds are not detected), which may enable future analyses to avoid possible confounding effects of P compounds inside functional groups (e.g., inositol hexakisphosphates and orthophosphate monoesters) and to make a more precise correction for potential degraded peaks occurring during the alkaline extraction and reading process. We also urge researchers to determine variances or standard errors for soils with distinctive properties. Then, as stated by Stewart (2010), future analyses could use the different information provided by studies of different scopes and quality in a meta-analytical approach.