Reversed marine phosphorus gradient in the Ediacaran Ocean

Abstract

The Ediacaran Period (635–539 Ma) witnessed the emergence of early animals, rapid oceanic oxygenation, and rising oceanic P availability. However, the relationships among these developments remain intensely debated. Here, we present high-resolution carbonate-associated phosphate (CAP) data from seven Ediacaran sections spanning diverse depositional environments across South China. Contrary to the modern ocean, our measurements reveal an inverted phosphorus (P) gradient, with lower P availability at depth. By integrating these data with a quantitative biogeochemical model, we show that increasing shallow-water P inputs drove enhanced productivity and bottom-water anoxia on continental shelves, while P-starved distal oceans became progressively oxygenated from the top downward, providing an oxygenated niche—a deep-ocean cradle—for early metazoan evolution. Our study establishes a conceptual framework linking nutrient dynamics, marine oxygenation, and evolutionary innovation in Earth’s early history.

Carbonate associated phosphate indicates that Ediacaran oceans had more P in shallow waters but less in the deep sea, opposite to modern. A distal P shortage may have aided deep-water oxygenation and early animals.

Introduction

Phosphorus (P), the primary limiting nutrient for marine productivity on geological timescales¹, is thought to have played an important role in the biogeochemical transformations of the late Neoproterozoic², a time marked by rising atmospheric and oceanic oxygen levels and the rapid evolution of complex life, particularly early animals, during the Ediacaran (635–539 Ma)³. Increasing oceanic oxidant availability and/or declining extents of ferruginous conditions (i.e., oceanic oxygenation) in the Ediacaran^4–6 have been proposed to have driven both the rapid evolution of complex life and enhanced oceanic P availability, potentially facilitating the formation of large phosphorite deposits². Alternatively, increased weathering-derived P fluxes, and thus higher Ediacaran oceanic P levels, may have elevated rates of P-limited oxygenic photosynthesis, leading to oceanic oxygenation and, in turn, the radiation of complex life^6–8. However, these hypotheses remain largely untested and debated, mainly because of the limited availability of methods for reconstructing palaeoceanic P concentrations and the corresponding lack of spatiotemporal information on oceanic P levels at that time. In this study, we measured carbonate-associated phosphate (CAP)—a proxy for marine P concentrations (see Methods for CAP methodology)⁹—in seven carbonate sections of the Ediacaran Doushantuo and Dengying formations on the Yangtze Platform (South China). These sections record a known Ediacaran oceanic oxygenation event (see below) and represent depositional environments ranging from shallow to deep marine. This approach permits a more robust reconstruction of spatiotemporal variability in Ediacaran oceanic P availability (Figs. 1, 2) and therefore allows us to test the hypotheses above by examining potential spatiotemporal differences in P cycling under the proposed scenarios.

Fig. 1. Inorganic carbon isotope (δ¹³C_carb) and carbonate-associated phosphate (CAP) given as CAP/(Ca + Mg) data for the geological sections analysed.

a Geological map of the Ediacaran Yangtze platform in South China. b–h Huangliang, Lianghong, Sishang, Jiulongwan, Siduping, Xiushan, Fengtan sections. CAP values with arrow and asterisk in the Jiulongwan section come from diagenetic calcite concretions hosted within the black shale of Member IV in the Doushantuo Formation. Dark blue CAP data points are dolomite, light blue calcite, and white outliers (1.5 x interquartile range). Note, due to faults the middle to end Shuram Excursion are missing in the Sishang and Huangliang sections. Dark grey lines over CAP data represent locally weighted scatter plot smoothing. CAP and δ¹³C_carb data from Sishang and Jiulongwan from ref. ¹³. All other CAP and δ¹³C_carb data from this study, except Siduping δ¹³C_carb data which is from ref. ¹².

Fig. 2. Carbonate-associated phosphate (CAP) and carbonate carbon isotope data (δ¹³C_carb) across depth gradients in Ediacaran Yangtze Platform (South China).

a, b CAP before and during the Shuram Excursion, in dolomite (blue) and calcite (green), including median line and mean marker “x”; c cross-plot of CAP in dolomite (blue) and calcite (green) against δ¹³C_carb values before the Shuram Excursion. All data from before the Shuram Excursion comes from Member III. ED = Ediacaran dolomite, EC = Ediacaran calcite, MC = modern calcite. Note, modern carbonate preserves a negative correlation between CAP and δ¹³C_carb consistent with the Redfield slope, whereas Ediacaran carbonates do not. Modern carbonate data (magenta) come from Timor Sea and Equatorial Pacific with Mg/Ca < 0.01, see Supplementary text 1 and Extended data Figs. 1–2. Where error bars not visible, the data point exceeds the error margins.

The sections analysed are, in order from shallow- to deep-water facies, Huangliang (inner shelf), Lianghong (inner shelf), Sishang (mid-depth shelf), Jiulongwan (mid-depth intrashelf basin), Siduping (upper slope), Xiushan (slope-basin), and Fengtan (slope-basin) (Fig. 1a; see Supplementary Information text 1 for the geological settings of the study sections). In the upper Doushantuo Formation, all sections exhibit a shift toward large negative carbonate carbon isotope (δ¹³C_carb) values, representing the well-known globally synchronous Shuram Excursion (SE; Fig. 1b–h)¹⁰. This event is interpreted as a prominent oceanic oxygenation event that may have contributed to the radiation of the Ediacaran biota¹⁰, as well as to the increase in P availability noted above¹¹. We chose to analyse CAP across the SE interval because (1) oceanic oxygenation during the SE has been proposed either to have increased P availability or to have resulted from it⁶, and (2) stratigraphic correlation of the upper Doushantuo Formation based on the SE allows spatiotemporal analysis of CAP variation as a function of sedimentary facies¹⁰.

Results

CAP trends in the study sections

In all study sections, CAP values [given as CAP/(Ca+Mg)] generally increased at the onset of the SE, returned to pre-event values during the SE, and then reached a second CAP peak in some sections toward the end of the SE (Fig. 1). These CAP trends have been observed globally in SE sections and are proposed to reflect interactions among the Ediacaran marine C-O-P-S cycles¹¹. Here, however, we focus on spatial differences in CAP values of pre-SE and syn-SE intervals among the study sections. Because calcite and dolomite have different distribution coefficients for CAP^9,11, we compared CAP values according to sample mineralogy, with dolomite defined here as carbonate with an Mg/Ca ratio > 0.2 mol/mol.

CAP before and after the Shuram Excursion

Before the SE in Doushantuo Member III, CAP in both dolomite and calcite was generally higher in shallow-water sections than in deep-water sections (Fig. 2a). From shallowest to deepest, median CAP values are 0.56 (n = 12, Huangliang), 0.35 (n = 23, Siduping), and 0.09 (n = 4, Fengtan) for dolomite-dominated sections, and 0.09 (n = 18, Sishang) and 0.02 (n = 2, Xiushan) for calcite-dominated sections (Fig. 2a). By assigning each section an ordinal value to represent relative depth, we statistically evaluated the CAP-versus-depth trend. Dolomite samples before the SE yield a Spearman r-value of – 0.69 and a p-value of < 0.05, and calcite samples yield an r-value of –0.53 and a p-value of < 0.05, suggesting that CAP gradients with depth are statistically significant. Following the SE, only the Lianghong and Sishang sections can be compared in calcite of roughly equivalent age, yielding median CAP values of 0.14 (Lianghong) and 0.22 (Sishang), respectively. The higher CAP in the mid-depth Sishang section than in the shallower Lianghong section is consistent with a possible residual effect of enhanced dissolved organic matter (DOM) oxidation releasing P at mid-ocean depths during the SE, resulting in the weakened, non-significant CAP–depth gradient that characterises the SE interval (Fig. 2b; see below).

CAP during the Shuram Excursion

During the SE, variability in stratigraphic and mineralogical coverage through the event interval makes bulk comparison difficult. Because CAP values outside the CAP peaks are generally similar within individual sections and plot along a minimum baseline, we can compare calcite and dolomite CAP data using these baseline values. To reduce the influence of outliers and provide a more robust statistical measure, we use the lower quartile (Q1) values of each section during the SE to compare minimum baseline CAP values. For dolomite, Q1 values in order of increasing depth are 0.23 (Huangliang), 0.05 (Lianghong), 0.09 (Sishang), 0.20 (Jiulongwan), 0.13 (Siduping), and 0.06 (Fengtan). For calcite, Q1 values in order of increasing depth are 0.09 (Lianghong), 0.09 (Sishang), 0.12 (Jiulongwan), and 0.03 (Xiushan). Assigning each section an ordinal value to represent relative depth, we statistically evaluated the CAP-versus-depth trend using the Q1 values. Dolomite samples have an r-value of –0.64 and a p-value of 0.173, and calcite samples have an r-value of –0.32 and a p-value of 0.684, indicating non-significant trends with depth (i.e., p > 0.05). Based on these values, the pre-SE depth-related gradients in CAP, and thus in aqueous P, became less pronounced during the SE: the mid-depth sections (e.g., Sishang, Jiulongwan, and Siduping) have median CAP values from dolomite or calcite that are generally higher than those from both shallower and deeper sections, except for the shallowest section (i.e., Huangliang; Fig. 2b). The elevated P levels in the mid-depth sections during the SE (Fig. 2) are consistent with the previous hypothesis that enhanced oxidation of oceanic DOM intensified P release at mid-ocean depths, thereby weakening the P and CAP depth gradients as oxidant availability increased during the SE¹¹.

CAP in modern marine carbonates

In modern carbonates collected at various ocean depths, CAP decreases with depth despite increasing seawater P concentration (Supplementary Fig. 1). Carbonate Mg and Sr concentrations also decrease with depth (Supplementary Fig. 2a, b), implying that CAP and other trace-element concentrations are likely strongly influenced by mineralogy, with aragonite hosting higher CAP, Mg, and Sr values than calcite in deeper water masses. After correcting for these mineralogical differences by screening out carbonates with Mg/Ca > 0.01 mol/mol, we find that CAP values actually increase with depth and with increasing P levels (Supplementary Fig. 2c, d and Supplementary Information text 2). Therefore, the pre-SE CAP gradients observed in our Ediacaran sections are inverted relative to those in the modern ocean.

Discussion

CAP gradient inferring an Ediacaran oceanic P gradient

In theory, the pre-SE CAP gradients observed in our study may reflect either an oceanic P concentration gradient or spatially variable diagenesis and varying degrees of CAP loss (see Supplementary Information Text 2, 3). Although spatially variable carbonate alteration has been inferred during the SE based on Ca-isotope and Sr-concentration data¹², our petrographic observations, fluid-rock alteration models, and geochemical data suggest that this process is insufficient to explain the CAP differences between shallow- and deep-water sections (Supplementary Figs. 3–5 and Supplementary Tables 1–3; see Supplementary Information text 3). We outline this below.

First, our samples show little petrographic evidence of widespread recrystallisation or textures reminiscent of late-stage diagenetic alteration (Supplementary Fig. 3). Second, Sr and CAP concentrations in dolomite and calcite samples from before the SE show no statistically significant correlations (Supplementary Fig. 5a; see Supplementary Information Text 3), and Sr concentrations across the full dataset show no consistent section-level patterns expected from fluid-buffered diagenesis (Supplementary Fig. 5b); in contrast, carbonates that have undergone fluid-buffered diagenesis typically show coherent trends between Sr concentration and degree of alteration^13,14. Third, we compared CAP values with a suite of geochemical parameters commonly used to evaluate diagenesis, including redox-sensitive elements (Fe, Mn), total organic carbon (TOC), bulk-rock P, carbonate-associated sulfate (CAS), and other standard alteration proxies (δ¹⁸O_carb, δ¹³C_carb, Mn/Sr) (Supplementary Tables 1–3). To evaluate these data, we grouped samples from before and during the SE by both mineralogy (calcite versus dolomite) and age, ensuring that we were making equivalent comparisons. We further examined the data both by individual section (Supplementary Tables 1, 2) and across the full dataset (Supplementary Table 3). This allowed us to test whether any apparent diagenetic signature was an artefact of site-dependent chemistry or was consistently present within individual sections. Comparing CAP with diagenetic proxies across samples from different depositional settings and within individual sections provides a powerful way to distinguish signatures linked to depositional environment from those produced by diagenesis.

In our dataset, no statistically significant correlation (p > 0.05) is observed between CAP and Mn/Sr, δ¹⁸O_carb, or δ¹³C_carb in the majority of sections, nor is there a consistent relationship with TOC within individual sections (Supplementary Tables 1–3). Furthermore, La, a rare earth element commonly found in phosphate minerals, shows no significant correlation with CAP in any section (p > 0.05), arguing against widespread influence from phosphate minerals or phosphitisation. Reduction of iron oxides and oxidation of TOC in marine sediments can both release phosphate to porewaters and, therefore, could, in principle, generate correlations among Fe, TOC, and CAP if Fe is incorporated into carbonate after reduction or if preserved TOC content tracks phosphate release to porewaters and subsequent incorporation into carbonate. This process is exemplified in the Jiulongwan section, where known diagenetic calcite concretions in shale produced by biomass remineralisation carry CAP values 2-6 times higher than those of adjacent primary carbonate sediments (Fig. 1e)⁹. There is, however, no significant correlation between Fe and CAP (Supplementary Tables 1–3). TOC and CAP show a statistically significant correlation (i.e., p < 0.05) across the entire dataset (Supplementary Table 3; R = + 0.34 to +0.79), reflecting higher TOC in shallow-water sections, as expected where productivity tends to be highest. However, when CAP and TOC are compared within individual sections, this relationship is not statistically significant (Supplementary Tables 1-2; R < + 0.34 and p > 0.05). If TOC content were controlling CAP values, we would expect this relationship to continue within individual sections, with organic-rich horizons carrying markedly different CAP values from horizons with lower TOC. Therefore, correlations between Fe or TOC and CAP do not support the idea that P released during early diagenetic cycling consistently elevated CAP.

To complement these comparisons, we also evaluate CAP against bulk-rock P and CAS, both of which can respond to early diagenetic processes in porewaters (Supplementary Tables 1–3). Bulk-rock P may vary because of sediment phosphatisation or elevated porewater phosphate fluxes, potentially co-affecting CAP if phosphate released into porewaters is incorporated into carbonate during diagenesis. Similarly, CAS is a well-known diagenetically sensitive proxy, with concentrations typically reduced in carbonates that recrystallised or precipitated under fluid-buffered or methanic conditions. A shared diagenetic control would therefore be expected to produce correlations between CAP and either bulk-rock P or CAS. Correlations between CAP and bulk-rock P are observed in several sections. The strongest are during the SE in dolomite of the Siduping (R = + 0.94, p < 0.05, n = 11), Huangliang (R = + 0.81, p < 0.05, n = 24) and Jiulongwan (R = + 0.60, p < 0.05, n = 18) sections, and may reflect a shared influence from water-column phosphate concentrations, authigenic P enrichment, or porewater fluxes. The Huangliang section before the SE shows a moderate but non-significant correlation (R = + 0.56, R² = 0.31, p = 0.059, n = 12), which we interpret as a relationship limited by small sample size rather than a true absence. By contrast, the Xiushan section shows essentially no correlation, particularly during the SE (R = – 0.23, R² = 0.05, p > 0.05, n = 27), making it the clearest case of bulk-rock P not controlling CAP. Bulk-rock P was not measured in the Fengtan, Sishang or Lianghong sections, so these cannot be assessed for this proxy. Taken together, the relationship between CAP and bulk-rock P is present in some shelf sections (especially during the SE) but absent or reversed in the deepest calcite section, arguing against bulk-rock P, and the diagenetic processes it tracks, being a universal or facies-independent control on CAP. Likewise, CAP shows no significant correlation with CAS (R =  0.15; p > 0.05), and there is no consistent covariation among sections. Although CAS can be altered during fluid-buffered recrystallisation, it can also be inherently low in carbonates that formed, or experienced early burial, in methanic zones, meaning that low CAS alone does not necessarily indicate diagenetic overprinting of CAP. Finally, we note that some CAP–proxy relationships have moderate R values but do not reach significance at small sample sizes, and we report these explicitly rather than treating them as definitive absences; even where correlations between CAP and indicators of alteration or authigenesis are weak or non-significant, local diagenesis in individual samples can still occur without producing strong covariation among diagenetic indicators¹⁵.

Taken together, these comparisons do not eliminate the possibility of localised diagenetic influence in individual samples, but they do argue against a systematic, depth-dependent, or facies-linked diagenetic control on CAP values among the study sections. Therefore, although we cannot conclusively rule out a diagenetic effect on the observed CAP gradient, any diagenetic model would need to alter CAP values in a consistent way across sections that represent markedly different depositional regimes (i.e., environments conducive to calcite versus dolomite formation) and distinct diagenetic histories, as indicated by trace-element and calcium-isotope data (Supplementary Figs. 3–5 and Supplementary Tables 1–3; see Supplementary Information text 3, 4). Because such an explanation would require a substantial degree of coincidence and multiple assumptions, we prefer the explanation that requires the fewest assumptions—namely, that the pre-SE CAP data reflect a first-order water-column P depth gradient.

A final possibility is that the shallow-water sections were deposited in a restricted basin that favoured P accumulation in shallow environments, thereby creating an artificial P gradient in the Ediacaran ocean record. There is indeed evidence for partial restriction of the shallow Yangtze Platform in South China during deposition of the Doushantuo Formation¹⁶, although the degree of restriction remains uncertain¹⁷. Several elemental proxies [e.g., δ¹³C_carb, δ²³⁸U_carb, CAP, I/(Ca+Mg)] from shallow and deep sections of the Yangtze Platform preserve geochemical trends through the SE that are mirrored temporally in globally distributed sections^11,18,19. Notably, uranium has an ocean residence time on the order of hundreds of thousands of years, and δ²³⁸U_carb records associated with the SE are consistent across the Yangtze Platform and many other global localities, suggesting that the Yangtze Platform was well connected to the global ocean^19,20. CAP trends in the upper Doushantuo Formation of the Yangtze Platform are also similar to those recorded in Australia, the USA, Mexico, and northwestern China¹¹, implying that the Yangtze Platform P cycle was well connected to the global marine P reservoir. Reproducing such consistent CAP trends independently at each locality without reference to the global marine P reservoir would be unlikely. Therefore, we interpret the observed CAP gradient as representative of global marine P distribution, although further validation from other Ediacaran basins is still needed.

Mechanisms and effects of P gradient in the Ediacaran oceans

Our pre-SE CAP data indicate a P gradient and, accordingly, lower P concentrations in the Ediacaran deep ocean. Two commonly cited scenarios could explain these lower P levels at depth: Fe-P scavenging^5,21,22 and inhibited organic P recycling^7,23. Ediacaran deep oceans are generally regarded as having been largely ferruginous, in which P may have been sequestered within the water column and sediments by ferrous and ferric minerals, and/or minerals containing both ferrous and ferric iron, thereby removing bioavailable P from the ocean^5,21,24,25. In addition, Ediacaran deep oceans have been proposed to have been depleted in sulphate and other oxidants relative to the modern ocean^26,27. This may, in turn, have suppressed rates of water-column organic matter oxidation and organic P recycling^6,7. Both processes are illustrated by data from the modern ferruginous lakes Matano (Indonesia) and Pavin (France), where Fe-P scavenging strongly controls surface-water P levels^28,29 and low oxidant availability allows up to 90% of organic matter to escape remineralisation³⁰. However, in these modern examples, water-column P is generally higher at depth than in surface waters because reductive dissolution of Fe oxides below the redoxcline releases P back into the water column^28,29.

Here, we tested the relative roles of Fe-P scavenging and inhibited organic P recycling using a well-established quantitative model of marine biogeochemistry (Supplementary Fig. 6)^31,32. The model tracks the global cycles of P, C, and O₂ in a four-box ocean system representing proximal shelf-ocean, distal shelf-ocean, distal surface-ocean, and distal deep-ocean environments (Figs. 3, 4 and Supplementary Fig. 6). Our modelling and modern-analogue results clearly indicate that the Fe-P scavenging scenario cannot adequately explain the CAP gradients observed in this study (see Supplementary Information text 5 and Supplementary Fig. 7). We therefore focus below on the inhibited organic P recycling scenario.

Fig. 3. Model results showing the effects of oxidant-limited P recycling and no oxidant-limited P recycling on ocean-atmosphere chemistry.

a Dissolved seawater P concentration in micromoles/ litre from the proximal shelf (magenta), distal shelf (green) and distal deep ocean (blue). The arrow points to the transition when deep ocean P concentrations rise above the proximal and distal ocean boxes. b Sedimentary P burial flux in mol of P/ year in the proximal (magenta), distal (green) and deep ocean (blue). c, d Fraction of seafloor anoxia in the proximal and distal zones. e Atmospheric O₂ level (PAL = present atmospheric level). f Dissolved O₂ in the deep ocean relative to the present day deep ocean dissolved O₂ concentration. Solid and dashed lines in a–f delineate model results for when oxidant-limited P recycling is not applied and applied, respectively, and show the median of a 1000-member Monte Carlo ensemble in which the background forcing parameters were randomly varied (see “Methods”).

Fig. 4. Conceptual ocean model for P, C and O cycling across the Ediacaran Yangtze Platform based on biogeochemical model results.

a, b Low marine P input during pre-Shuram/ non-oceanic oxygenation and high marine P input during Shuram/ oceanic oxygenation event, respectively. Note, when P input increases as shown in b, the nearshore proximal areas and distal shelf-surface ocean areas become increasingly oxygenated due to increasing input of weathering-derived sulphate and photosynthetic O₂ release to the former and mixing with an increasingly oxygenated atmosphere to the latter, allowing these areas to become populated by early animals. Conversely, between these extremities, the most shelf area remains anoxic due to elevated rates of primary productivity and, therefore, dissolved oxygen consumption in lower water column. In the distal deep ocean, limited quantities of dissolved sulphate result in the transition from anaerobic oxidation of methane to methanogenesis at depth. These processes drive deep water δ¹³C_DIC to more negative values through methane oxidation in the upper water column while also keeping low P levels in the deep ocean. See text for more details. [CH₂O]_x= fresh organic matter. DOM = dissolved organic matter (represented by black dots). Red boxes refer to quantitative biogeochemical model areas in this study. Green stars = primary producers.

To model inhibited organic P recycling within the water column, we made organic matter recycling (and thus organic P recycling) dependent on dissolved O₂ concentrations in each box, with suppression becoming strong below 2 μM dissolved O₂ (see Methods for a detailed description). Because we do not know how limiting oxidant availability may have been, we explored a range of limiter values from 0.1 to 1.0 relative to the original model remineralisation flux, where 0.1 represents a 90% reduction in organic matter recycling. We found that a 5% reduction relative to the original model recycling rate was sufficient to drive distal deep-ocean P levels below those of the proximal shelf and distal shelf boxes (Fig. 3a, dashed lines). Given the uncertainties, we chose this minimum reduction in organic matter recycling to replicate the CAP data, that is, lower P concentrations in the distal deep ocean than in the proximal and distal shelf oceans. We then ran the model by gradually increasing riverine P input to the proximal shelf ocean, following previous suggestions that weathering and riverine P fluxes may have increased during the SE¹¹ and over geological time more broadly^33,34. In these model runs, when organic P recycling is suppressed under anoxia, distal deep-ocean P concentrations remain below those of the distal and proximal shelf boxes until P inputs to the ocean approach 25% of the modern value (see arrow and dashed lines in Fig. 3a). By contrast, in model runs without limited organic P recycling, distal deep-ocean P concentrations always remain higher than those of the distal and proximal shelf boxes (see solid lines in Fig. 3a), a pattern mirrored by sedimentary P burial rates in the respective boxes (Fig. 3b).

In our model, as P input fluxes to the ocean increase, so too do global ocean productivity and, in turn, atmospheric and dissolved-ocean O₂ levels (Fig. 3e, f). Rising dissolved O₂ causes distal deep-ocean P concentrations to increase above those of the proximal and distal shelf boxes because oxidant limitation of organic P recycling is progressively alleviated (Fig. 3a). As riverine P input increases, the extent of bottom-water anoxia in the proximal shelf ocean also increases (Fig. 3c), whereas bottom-water anoxia in the distal shelf-ocean region remains limited (Fig. 3d). This outcome reflects increasing proximal shelf-ocean P availability (due to riverine input) and persistent distal shelf-ocean P starvation (due to limited P recycling from organic matter). Productivity therefore becomes increasingly concentrated in proximal shelf regions, where aerobic decomposition of sinking organic matter places a greater demand on dissolved O₂ in the water column, leading to more extensive bottom-water anoxia (Fig. 4b). Conversely, when organic P recycling is not limited, the proximal shelf ocean remains more oxygenated than the other ocean boxes (Fig. 3c, d). Additional P inputs, such as P released during possible DOM oxidation in mid-depth shelf water masses during the SE oxygenation event, as suggested previously¹¹, may also have contributed to higher productivity and more intense bottom-water anoxia on shelves, as well as to a less pronounced P gradient during the SE (Fig. 4a versus 4b), although this effect was not modelled separately here.

Furthermore, higher productivity and bottom-water anoxia on the proximal shelves would have increased O₂ production through enhanced organic burial, leading to higher atmospheric pO₂ (Fig. 3e) and increased dissolved O₂ levels in the ocean-surface mixed layer via equilibration with atmospheric O₂ (Fig. 4a versus 4b). We note that direct photosynthetic O₂ release from higher productivity in proximal shelf oceans, together with enhanced weathering-derived oxidant inputs (such as sulphate)³⁵, would also have favoured the development of a more oxygenated oasis in nearshore proximal-shelf environments suitable for complex eukaryotic life (Fig. 4b). By contrast, higher atmospheric pO₂ would have particularly favoured oxygenation of distal oligotrophic oceans because P starvation in the distal deep ocean would have suppressed productivity and thereby reduced water-column consumption of dissolved O₂ and other oxidants (Fig. 3f). Interestingly, when oxidant-limited P recycling is included, the deep ocean reaches higher dissolved-O₂ levels for a given atmospheric O₂ level than in the baseline model without oxidant-limited P recycling (Fig. 3e, f). Consequently, with higher P availability in proximal settings and deep-ocean P starvation, the P gradient observed in this study would have helped maintain more extensive oxygenation in distal shelf-deep oceans while promoting widespread bottom-water anoxia across much of the proximal continental shelf (Fig. 3c versus 3d, f), except for the nearshore oxygenated oasis noted above (Fig. 4b). This result may help explain the long-standing enigma of why some of the earliest unequivocal animal fossils are found in deep-water slope environments^36,37, such as the emergence of the deep-water Avalon biota during the SE oceanic oxygenation event³⁸ (Fig. 4b). This interpretation is further supported by iodine-concentration and cerium-anomaly data^39,40, which indicate preferential oxygenation of distal shelf environments relative to proximal shelf environments during the SE^18,19,41. Finally, we note that when organic P recycling is not inhibited, atmospheric O₂ levels are considerably higher (Fig. 3e) because overall oceanic P availability, and therefore primary productivity and O₂ production, are greater.

Coupling of CAP and carbon-isotope gradients

Accompanying the CAP gradients, large δ¹³C_carb gradients have also previously been observed between shallow- and deep-water Ediacaran Doushantuo sections, including the study sections⁴². For example, pre-SE dolomite δ¹³C_carb averages + 4.1‰ in the inner-shelf Huangliang section and + 2.5‰ in the upper-slope Siduping section (Fig. 1), whereas it averages – 3.7‰ in the slope-basin Fengtan section. In comparison, pre-SE calcite δ¹³C_carb averages + 1.8‰ in the mid-depth Sishang section and – 4.0‰ in the slope-basin Xiushan section. When δ¹³C_carb and CAP values from the same pre-SE samples are plotted together, the data show a positive trend for both dolomite and calcite, whereas modern (0.017 to 0 Ma) calcite δ¹³C_carb and CAP values show a negative trend consistent with the Redfield slope (Fig. 2c). Therefore, the mechanism we propose—oxidant-limited organic P recycling in the Ediacaran deep ocean—must also account for the accompanying δ¹³C_carb gradients.

In the modern ocean, ~ 96% of primary productivity and its nutrient content are recycled principally through aerobic oxidation as organic matter sinks through the water column, producing P-enriched and ¹³C-depleted deep waters because respiration releases ¹²C-enriched CO₂ and organic-bound P from biomass⁴³. This process is governed in part by the Redfield ratio of planktonic biomass in the surface ocean (i.e., C:P = 106:1), which underlies the negative correlation between δ¹³C of dissolved inorganic carbon (δ¹³C_DIC) and P concentrations throughout the ocean, often referred to as the Redfield slope (Fig. 2c)⁴³. The positive coupling between CAP and δ¹³C_carb in Ediacaran carbonate sections spanning different palaeodepths (Fig. 2c) therefore implies that oceanic P cycling and δ¹³C_DIC did not follow modern water-column cycling patterns.

We propose that the positive coupling between CAP and δ¹³C_carb likely reflects not only oxidant-limited organic P cycling but also elevated rates of methanogenesis within the water column of the anoxic distal deep ocean, driven by low levels of dissolved oxygen and sulphate (Fig. 4a). In this scenario, P contained in biomass, including DOM, was incompletely returned to the water column during oxidant-limited organic remineralisation, resulting in the low deep-ocean P levels outlined above. This mechanism may have been reinforced by Fe-P scavenging, which removed dissolved P through incorporation into ferrous-ferric minerals such as vivianite in deep ferruginous oceans (Fig. 4a)²². Given the scarcity of oxidants, methanogenesis and anaerobic methane oxidation would have dominated carbon recycling at depth. Because methanogenesis generates highly ¹³C-depleted CH₄ that is subsequently oxidised to DIC when it diffuses into surface waters containing oxidants (Fig. 4), a positive relationship between δ¹³C_carb and CAP could have developed in Ediacaran carbonates, opposite to that observed in modern carbonates (Fig. 2c). The presence of a deep-water methanic zone in the Ediacaran ocean is supported by the preservation of organic matter with strongly negative δ¹³C compositions (δ¹³C_org < − 32‰) in deep-water sections of the Ediacaran Yangtze Platform^10,42,44. Available estimates suggest that sulphate concentrations in Ediacaran seawater were 14 to 140 times lower than in the modern ocean, implying much lower sulphate abundance at depth^45,46 and thus supporting our interpretation of a deep-water methanic zone.

Finally, we note that debate persists over the existence of an Ediacaran sulphate gradient⁴⁷, and whether sulphate deficiency limited deep-ocean biomass remineralisation rates, as suggested above, also remains unresolved^6,7. Nevertheless, oxidation of organic matter is known to be more strongly limited under anoxic than oxic conditions⁴⁸ and even more so at low sulphate levels⁴⁹, both of which are thought to have characterised Ediacaran oceans²⁶.

Summary and perspective

Our CAP data from seven Ediacaran sections spanning diverse depositional environments across the Yangtze Platform demonstrate a seaward decrease in P concentration that is unlikely to be explained by diagenetic effects or basin restriction. Using a well-established quantitative model of marine biogeochemistry, we find that the lower P levels in the Ediacaran deep ocean were most likely caused by inhibited organic P recycling due to limited oxidant availability, rather than by Fe-P scavenging alone. Lower P availability in distal deep Ediacaran waters likely reduced productivity and oxidant demand, thereby promoting earlier oxygenation in distal oceans relative to proximal shelf settings, except in nearshore waters where higher inputs of weathering oxidants and higher photosynthetic O₂ release from higher productivity and turbulent mixing may have created oxygenated oases. This phenomenon would have become increasingly prominent as P concentrations rose in proximal shelf oceans, such as during the SE, providing a deep-ocean cradle for early animals in addition to the nearshore proximal oxic waters. More broadly, these results imply that P recycling drove spatially heterogeneous oceanic oxygenation and, in turn, influenced early radiations of complex life in the Ediacaran.

Furthermore, our CAP data show a positive coupling with accompanying δ¹³C_carb data along a palaeoceanic depth gradient, opposite to that observed in the modern ocean. We propose that this positive coupling between CAP and δ¹³C_carb was generated mainly by oxidant-limited biomass remineralisation at depth, which not only weakened organic P recycling and thereby lowered P levels, but also promoted methanogenesis and thus oxidation of ¹³C-depleted CH₄ for Ediacaran carbonate formation. Our preferred model for the observed CAP and δ¹³C_carb gradients is supported by existing evidence for chemical zonation in Ediacaran oceans^26,44,50.

Our interpretation is readily testable on other ancient continental platforms through spatial geochemical analyses that can capture marine chemical gradients. We further speculate that this spatial control on P cycling, and its effects on early oceanic oxygenation and complex life, did not disappear until widespread deep-ocean oxygenation and a resulting rise in seawater P to modern levels in the late Palaeozoic⁵¹.

Methods

Materials availability

All samples were collected and exported responsibly and in accordance with relevant permits and local laws. Requests for materials should be addressed to C.L.

Bulk-rock P and trace element geochemistry

Fresh pieces of carbonate rock and sediment were selected and crushed to a powder. Trace element measurements were conducted by dissolving 50 mg of sample powder using a standard HNO₃–HF digestion protocol, in which HNO₃–HF (1:1) and HNO₃ were sequentially added to a 15-mL Teflon bomb at 190 °C until complete digestion was achieved. Distilled HNO₃ and trace metal-grade HF reagents were used for all samples. The leachates were then measured using an Agilent 7700x inductively coupled plasma mass spectrometer (ICP-MS). Analytical errors were better than ± 7% for all elements based on duplicate analyses of two USGS standards (BCR-2, BHVO-2) and two Chinese national standards (GSR1, GSR5).

Carbon analyses

Total carbon (TC), total inorganic carbon (TIC) and total organic carbon (TOC) were measured using a CS-900 Carbon Analyser. TC was measured directly by weighing out ~ 0.1 g of rock powder with iron shavings and tungsten pellets to aid combustion in the analyser. A calcium carbonate carbon standard (AR1034) was analysed after every 10 samples to check reproducibility ( ± 0.1 wt% C; n = 30). To measure TOC, an aliquot of sample powder ( ~ 1 g) was reacted with 6-M HCl to decarbonate the sample, rinsed with deionised water to neutral pH, then centrifuged and dried; the same process was then repeated as for measuring TC on the analyser. TIC was then calculated as the difference between TC and TOC.

Carbonate carbon- and oxygen-isotope analyses

About 60–300 μg of sample powder were loaded into a vial after drying at 70 °C for 24 h in an argon atmosphere. The samples were then reacted with 100% phosphoric acid under a vacuum at 70 °C for 220 s using a Kiel IV device. The resulting CO₂ was subsequently introduced into a MAT 253 isotope ratio mass spectrometer for isotopic measurements. Delta values were calibrated relative to international reference standard NBS-19 (δ¹³C = + 1.95‰; δ¹⁸O = − 2.20‰) and Chinese national standard GBW04416 (δ¹³C = + 1.61 ± 0.03‰; δ¹⁸O = + 1.59 ± 0.11‰). Carbon-isotope and oxygen-isotope data for carbonates are reported relative to Vienna Pee Dee Belemnite (VPDB) with a precision better than ± 0.1‰ based on duplicate analyses of GBW04416 and the study samples.

Carbonate-associated phosphate (CAP) analysis

CAP was measured following previously established protocols⁹. Only samples comprising > 50% carbonates were chosen for analysis. Based on sample TIC and carbonate mineralogy, sufficient sample powder was weighed out to yield ~ 1 mmol of carbonate. Samples were then repeatedly washed for 24 h each time using 4 mL of 10% NaCl solution buffered to pH 8 with NaHCO3 to remove adsorbed P. Once adsorbed P was below 0.1 ppm in the wash solution, an appropriate amount of 2% vol/vol acetic acid was then added to the powder to dissolve up to 70% of the carbonate. After 30 min the sample was centrifuged, and the leachate was extracted. An aliquot of leachate was taken for major and trace element analysis and measured using an Expec 6000 Inductively Coupled Plasma Optical Emission Spectrometer (ICP-OES) for Ca and Mg and an Agilent 7700x ICP-MS for trace elements. Analytical errors were better than ± 2% for all elements based on duplicate analyses of two USGS standards (BCR-2, BHVO-2) and two Chinese national standards (GSR1, GSR5). Another aliquot was taken and mixed with H2SO4 to achieve a 1-M H2SO4 concentration and left for 1 hour to remove Ca as CaSO4. Then, P in the solution was measured by spectrophotometry using the molybdenum-blue method at a wavelength of 880 nm, with a relative standard deviation of better than ± 5%. The sample residue was then washed with 4 mL of 10% NaCl solution buffered to pH 8 with NaHCO3 for 24 h, and P in the wash was measured by the molybdenum-blue method. This sequence was repeated until P in the wash was below 0.1 ppm. The P in the leachate and washes were summed together to give CAP, which was then normalised to the Ca and Mg concentrations in the leachate. Three Ediacaran dolomite sediments of varying TIC and TOC were run alongside unknown samples to check reproducibility for quality assurance, with duplicate CAP measurements differing by < 10%.

Carbonate-associated sulphate (CAS) analysis

CAS was extracted using a modified method to measure CAS in small sample sizes for concentration only. For each sample with TIC > 50%, ~ 0.2 g of powder was weighed into 50 mL centrifuge tubes. To remove any adsorbed or easily exchangeable sulphate or sulphur compounds, each sample was washed in 30 mL of 0.5 M NaHCO₃ solution for 24 h and centrifuged, before discarding the solution and washing the residual powder with 18.25 MΩ distilled water. A 0.5 M NaHCO₃ solution was chosen instead of the 10% NaCl commonly used, as the hydroxyl generated from 0.5 M NaHCO₃ is far more effective at removing SO₄^2- than Cl^-. These steps were repeated 4 times, then the powder was washed with 5 mL of 18.25 MΩ distilled water for 24 h, and the solution measured for dissolved sulphur using an Expec 6000 ICP-OES. If the sulphur concentration was below 0.1 ppm, the samples proceeded to dissolution with 1 M HCl, if not the sample was subjected to further washes with 0.5 M NaHCO₃ and 18.25 MΩ distilled water until the sulphur concentration in the solution was < 0.1 ppm. Once the samples were determined to be clean of exchangeable sulphur, the samples were dissolved in 5 mL of 1 M HCl for 30 min. The samples were then centrifuged, and 0.1 mL of the leachate was taken and diluted in 4% HCl for Ca, Mg and Fe analysis by ICP-OES. The remaining leachate was passed through cation resin to remove Ca and Mg, and the solution was then diluted with 18.25 MΩ distilled water and measured for sulphur by ICP-OES. CAS was then calculated by normalising the sulphur concentration to CaCO₃ and MgCO₃ as follows: CAS(ppm) = (S(ppm) × 3 × 10⁶) / (Ca(ppm) × 2.5 + Mg(ppm) × 3.5).

Four-box biogeochemical model

The four-box model from ref. ³³ was used to calculate biogeochemical feedbacks between the P and O₂ cycles. This model contains the original Fe scavenging code from ref. ³². This model is solved in MATLAB using the Ordinary Differential Equation (ODE) suite. Here, we describe the equations and model modifications used in this study; for a full description of the original model, see ref. ³³.

Model setup

In the original model, several forcings modulate atmospheric O₂ over Earth history: a reducing-gas flux, continental emergence and weathering, a mantle-to-crust carbon flux, land-plant weathering, and terrestrial organic-carbon burial. Here, we removed the two terrestrial forcings (land-plant weathering and terrestrial organic-carbon burial), which are not central to the marine P–O₂ feedbacks examined here. The mantle-to-crust carbon flux was retained because it drives the secular increase in riverine P input, allowing us to express results against “P input to the shelf ocean relative to today.”

Because the remaining background forcings are poorly constrained for the Precambrian, we did not adopt a single deterministic parameterisation. Instead we ran a Monte Carlo ensemble of 1000 simulations in which the principal background parameters were drawn independently from uniform distributions across plausible ranges, as per the original model³³: the magnitude and timing of continental emergence (emergence factors EXP = 0.10–0.75 and EXP2 = EXP–1; emergence timing = 1.7–3.0 Ga), the mantle-degassing/reducing-gas scaling factor (D = 0.25–2), the crustal (mantle-to-crust) carbon-cycle exponent (C = 0.25–3), and the anoxic organic-matter C:P burial ratio (CP = 106–1100). Each ensemble member was integrated over Earth history and interpolated onto a common riverine-P-input axis; all model results in Fig. 3 and Supplementary Fig. 7 are reported as the ensemble median across the 1000 runs.

Based on the observed CAP trends over a range of water depths in the Doushantuo Formation, we infer that a mechanism(s) was maintaining lower P at greater ocean depths. As it is ambiguous what mechanism(s) could have maintained a low deep ocean P reservoir during the Ediacaran^5–7,21, we chose to model two scenarios where deep ocean P is lowered by either an iron-bound P scavenging flux^5,21,22 (Eqs. 1 and 2; see below) or oxidant-limited recycling of organic phosphorus in the water column^6,7.

Steady-state calculation for Fe scavenging of marine P

The iron-bound P scavenging flux is computed as below:

$${eScav}_{surf}=\frac{1}{1+e^{5000\cdot \left(O2\_DPconc-scavlim\right)}}$$ 1

$${Fescav}_{surf}=\left({eScav}_{surf}\cdot SRP\_DP\_S\cdot sigscav\right),\mathrm{upper}\mathrm{limit}=4.5\times {10}^{11}$$ 2

$$\mathrm{SRP}\_\mathrm{DP}\_\mathrm{S}=\mathrm{SRP}\_\mathrm{DPconc}\cdot \mathrm{Water}\_\mathrm{DP}\_\mathrm{S};$$ 3

Equation 1 computes a variable function for P scavenging dependent on the O₂ concentration in the deep ocean (O₂_DPconc) and a predefined scavenging limit (scavlim); we choose 0.01 for our model run for illustrative purposes only. A higher scavenging limit requires higher deep ocean O₂ concentrations to decrease the P scavenging flux. Equation 2 determines the quantity of P scavenged in the deep ocean, which is dependent on the amount of P being transferred from the deep to surface boxes via water mass mixing following Eq. 3, the concentration of O₂ in the deep ocean (eScav_surf), and the percentage of P that is scavenged (sigscav). In Eq. 3 SRP_DPconc is the concentration of soluble reactive phosphorus in the deep ocean, and Water_DP_S is the water mass flux from the deep to surface boxes and is a constant. The total scavenging flux was capped at an upper limit of 4.5 × 10¹¹ moles of P. The scavenging flux (${Fescav}_{surf}\left)\right)$ is subtracted from the deep ocean soluble reactive phosphorus reservoir, because the P is scavenged from the upwelling P flux before it enters the surface ocean box. In our model scenarios, we set the percentage of P scavenging at 5.5%, which was sufficient to achieve a roughly equitable concentration of P in the proximal, distal and deep ocean boxes, irrespective of riverine P input.

Steady-state calculation for oxidant-limited recycling of organic P

To represent oxidant control on water-column P recycling, we modified the existing particulate organic phosphorus (POP) remineralisation scheme so that remineralisation depends smoothly on local dissolved O₂ but does not go to zero under anoxia, due to the continuation of remineralisation via anaerobic metabolisms. For each ocean box i (proximal, distal, surface, deep) we first define an oxic baseline POP remineralisation flux, which is identical to the original model (Eq. 4):

$${R_i}^{original}={kprel}_i{OP}_i$$ 4

(where (OPi) is the POP inventory in box i). We then multiply this baseline by an O₂-dependent limiting function (Eqs. 5 and 6):

$${lim}_i=fan+\left(1-fan\right)\frac{\left[O_2\right]i}{k+\left[O_2\right]i},k=2\times {10}^{-3}\left(\mathrm{i}.\mathrm{e}.2\mathrm{\mu }\mathrm{M}{\mathrm{O}}_2\right),fan=0.95$$ 5

$${R_i}^{new}={R_i}^{original}{lim}_i$$ 6

Here ([O2]i) is dissolved oxygen in mol/m³ in box i; (K) is a scale parameter corresponding to the O₂ level at which the transition between anaerobic and oxic behaviour occurs (set here at 2 µM); and fan (0 < fan < 1) is the minimum fraction of the oxic remineralisation rate that is sustained by anaerobic metabolisms (e.g., nitrate, Fe, Mn, sulphate reduction) when O₂ is effectively exhausted. Given we do not have strong constraints on oxidant availability in the Ediacaran oceans, and to what degree oxidant availability may have limited organic matter remineralisation, we varied the fan between 0.95 and 1, which was sufficient to cause P to become depleted in the deep relative to proximal and distal ocean boxes. In these cases, 0.95 results in a 5% reduction of P recycling in the water column when there is ≪ 2 µM dissolved O₂ in that box, and 1 representing the original model P recycling efficiency when dissolved O₂ ≫ 2 µM. This smooth, two-plateau function avoids the unrealistic behaviour of fully shutting down remineralisation at low dissolved O₂, while still enforcing weaker water-column P recycling in oxidant-poor ocean regions. It also allows us to demonstrate that the key qualitative features of the model (lower deep-ocean P compared to the proximal and distal boxes, and delayed coastal oxygenation relative to offshore boxes as marine P input is increased) are robust across a range of oxidant supply, rather than being an artefact of a particular oxidant species or a hard O₂ threshold.

This formulation has two limits:

• High-O₂ (oxic) limit, [O₂]_I ≫ k:

lim_i → 1

R_i^new = R_i^original

so POP remineralisation increases to the original rate where remineralisation is not oxidant limited.

Low-O₂ (oxidant-limited) limit, [O₂]_I ≪ k: lim_i → fan

R_i^new ≈ fan R_i^original

So, remineralisation is capped at a fixed fraction of the oxic rate, rather than being switched off entirely, to capture uncertainty in the extent of organic matter remineralisation under anoxic oxidant-limited conditions.

The fraction of POP that escapes remineralisation is buried in the sediments with its original C:P stoichiometry (Eqs. 7 and 8). We close C and P mass balance by converting the buried POP in box i into an extra organic carbon burial/export flux using the Redfield C:P ratio and subtract it from the POC pools in each box (Eq. 8), while also including it in the total organic burial equations so that it is included in the O₂ flux and carbon isotope computations. Note, for the surface ocean box, it is assumed C and P are both directly buried in the deep ocean box. In summation, under low dissolved O₂ ( < 2 µM), the model remineralises less POP in the water column and instead buries more organic C and P, and that extra burial is internally consistent across the carbon, oxygen and phosphorus reservoirs.

$${\mathrm{POP\; buried}}_i={R_i}^{original}\mathrm{-}{R_i}^{new}$$ 7

$${\mathrm{POC\; buried}}_i={\mathrm{POP\; buried}}_i\mathrm{x}\mathrm{CP}\mathrm{redfield},$$ 8

Full model description

See ref. ³³ for full model equations and parameters.

Author contributions

C.L. led the research. C.L. and M.S.D. designed the study. M.S.D., M.C., Z.Z., H.G., L.Z. and J.F. performed analyses. C.L., Z.Z., M.C., H.G., J.H., M.H. and A.Y.S. provided samples. M.S.D and B.J.W.M. performed biogeochemical modelling, and T.J.A. contributed to the interpretation of the results. M.S.D. and C.L. wrote the manuscript, with important discussion and contributions from all authors.

Peer review

Peer review information

Nature Communications thanks Bing Shen and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by the National Natural Science Foundation of China (grants #42130208 and #42425002) and the National Key Research and Development Programme of China (2022YFF0800100) awarded to C.L. and a Forrest Research Fellowship awarded to M.S.D. A.S. acknowledges support for RV Sonne cruises funded by the German Federal Ministry of Education and Research (BMBF grant 03G0185A).

Data availability

The carbonate-associated phosphate, trace-element and isotope data generated in this study are provided in the Supplementary Information and have been deposited in Figshare https://doi.org/10.6084/m9.figshare.30121363⁵². Global coordinates, location information and drill-core names are provided for all collected samples in the Supplementary Information.

Code availability

The modified four-box biogeochemical model code used in this study is available from the authors upon request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-77093-z.