Every paddy manager knows the rule for cadmium: flood the field. Standing water pushes soil redox potential (Eh) negative, precipitates Cd as sulfide, raises pH, and keeps the metal out of the grain. It is the cornerstone of safe rice production on contaminated land — and a team at the Hunan Academy of Agricultural Sciences has now systematically inverted it. In a pot study plus two years of field trials at Yonghe Town, Liuyang City, Hunan Province, Yin, Huang, Liu and colleagues (Plant and Soil, 2026; CC-BY preprint) show that the same physiology, run backwards and timed to the plant’s own cadmium-uptake clock, turns an ordinary rice crop into an extraction crop.
The mechanism: oxidation on a schedule
The treatment is agronomically simple. From seedling recovery to late grain-filling the fields run alternate wetting and drying — each drying phase goes to visible cracks, each wetting phase holds a 1–2 cm flood — and from late grain-filling to maturity the field is simply drained. The chemistry is not. Oxidation of Fe²⁺ and Mn²⁺ releases protons (O₂ + 2Fe²⁺ + 4H₂O → 2Fe(OH)₃ + 2H⁺), so pH fell (by 0.32 units on average versus conventional irrigation, per the abstract) while Eh rose (by ~350 mV, from around −164 mV under flooding to positive values averaging 242–292 mV across treatments). The oxidative conditions convert insoluble CdS to soluble Cd²⁺ plus sulfate, and Fe/Mn-mineral oxidation releases sorbed and co-precipitated Cd. The measured result: DGT-labile Cd averaged 0.08 µg kg⁻¹ under conventional irrigation versus 0.67 µg kg⁻¹ under managed drying — a 7.94-fold increase. Within each wet–dry cycle the effect was sharpest early: at tillering, drying-phase DGT-labile Cd ran 61.70% above the wetting phase. Porewater Cd peaked at heading at up to 7.88 µg L⁻¹. Cadmium was not just mobilised; it was mobilised in pulses that the plant could be positioned to intercept.
Timing beats intensity
The non-obvious design choice is when, not how hard. Comparing their three water-management regimes, the authors found that extending drainage through the late grain-filling-to-maturity window mattered more than adding extra wet–dry cycles: post-filling drainage lifted DGT-Cd by 44.62% over continued alternate wetting and drying, and the regime with the longest dry tail (WMG1) produced the highest plant Cd in the pot study — tissue concentrations of 29.96 mg kg⁻¹ in roots, 8.35 in stems, 3.73 in leaves and 5.51 in spikes, 7.4- to 19.4-fold above controls. Daily accumulation rates confirmed the window: under managed drying, Cd influx during heading–flowering ran up to ~170 times the control rate in leaves and ~117 times in stems, with the stem dominating during grain-filling. In the pots this took removal efficiency from 4.91% to 23.73%.
The plant outsources part of the job to its microbiome
Water regime, not cultivar, was the dominant driver of the rhizosphere community — about 27% of variation along the first PCoA axis, with PERMANOVA R² = 0.31 (P < 0.001) — and functional profiling points at who is doing the mobilising. Managed drying enriched Acidobacteria (including Candidatus Koribacter, whose abundance correlated with total plant Cd accumulation at R² = 0.37, P < 0.001) and FAPROTAX-predicted functions for sulfide, thiosulfate and iron oxidation, while the Fe/S-reducing phyla (Bacteroidota, Firmicutes, Desulfobacterota) fell. The regime effectively cultivates an oxidising rhizosphere consortium that dissolves Cd-bearing sulfides and iron oxides — a reminder that in phytoextraction the plant is often the logistics network for a microbial mining operation, a theme we have seen before with Sedum alfredii-derived synthetic communities.
Field numbers and the honest bounds
In the field, the high-Cd-accumulating cultivar Pokkali under the optimised regime accumulated 2.64 mg Cd per plant in its shoots — 474.56 g ha⁻¹ per single cropping season at standard planting density (512.76 g ha⁻¹ for the whole plant including roots). Non-rhizosphere soil total Cd fell from 0.80 to 0.50 mg kg⁻¹, a measured 37.5% reduction in one season. That is in the league of dedicated hyperaccumulators the paper cites — Sedum alfredii and Sedum plumbizincicola are quoted at ~27.6% Cd removal per growing season — but delivered by a crop farmers already grow, with no new cultivation infrastructure. The comparison comes with caveats both scientific and regulatory, and the authors are unusually candid about the latter: under managed drying, grain Cd rose 1.63-fold (Pokkali) and 2.44-fold (9311) in 2024, the grain fails food-safety standards, and they explicitly call for policy keeping remediation rice out of the market plus safe-disposal protocols for the Cd-laden biomass. That is the real risk profile of the method: every drained window is also a leaching window (porewater Cd spiked above 7 µg L⁻¹, and downward Cd flux was not measured), and a single harvest-segregation failure puts high-Cd grain back into the food chain.
The numbers also need bounding in a way the paper itself does not fully manage. Its “theoretical” remediation efficiency for the same 474.56 g ha⁻¹ extraction is printed as 43.01% (abstract and results), 43.04% (discussion) and 46.79% (conclusions); we verified all three against the preprint and report the measured before/after reduction (37.5%) as the load-bearing figure. The 7.94-fold DGT increase is likewise loose arithmetic (0.67/0.08 is 8.4-fold), one discussion sentence states the mechanism with the redox directions backwards, and the microbiome phylum declines include percentages over 100% that cannot be literal. And this is one mining-contaminated granite-derived paddy in Hunan, two seasons, three replicates — a strong demonstration of mechanism at field scale, not yet a regional recipe. As a route to decontaminating paddies where grain can never be safe anyway, though, inverting the flood rule is a genuinely new lever.
Source: Yin, Huang, Liu, Chen & Ji (2026), Plant and Soil, DOI 10.1007/s11104-026-09050-0 (CC-BY preprint via ResearchSquare, posted 2026-05-07). Every load-bearing number was verified verbatim against the full text by three independent reviewers via two independent retrieval paths (direct preprint PDF; independent web re-retrieval); Springer’s published-version page was not fully machine-readable, so figures were bounded to the preprint. Noted caveats: the theoretical efficiency is reported inconsistently within the paper (43.01/43.04/46.79%), the 474.56 g ha⁻¹ figure is per cropping season rather than per year, and the stated field longitude (113°89′E) is not a valid coordinate as printed. See /methodology/ for how we source and check analysis pieces.