The phytoremediation and biofortification literatures both quote chemical speciation — which molecular form of an element a plant holds — to argue safety and efficacy. Duan, Du, Li, Zhang, Zhang and Yuan (Xi’an Jiaotong-Liverpool University and collaborators) have published a methods paper with an uncomfortable corollary for that habit: for selenium, the extraction step decides what you see, and the conventional protocol is the wrong one for hyperaccumulators (Food Chemistry: X 39:104410, 2026, gold open access).
Seven species, two columns, one cheaper detector
The team built an orthogonal LC-AFS system — strong anion exchange (SAX) for the inorganic oxyanions SeO₃²⁻ and SeO₄²⁻, strong cation exchange (SCX) for five organic species (SeLan, SeMet, SeCys₂, MeSeCys, γ-GluMeSeCys) — with atomic fluorescence detection as an alternative to ICP-MS. The orthogonality is not cosmetic: on SAX, SeLan, SeCys₂ and selenomethionine oxide co-elute at one retention time, and the literature-cited behaviour of SeCys₂ is elution near the column void (~2.5 min) with a high co-elution risk — an SCX separation under 90% buffer A resolves all five organics within 13 min. The explicit motivation is access: “the high cost of ICP-MS restricts its routine application in many food analysis laboratories.”
The mechanism: two plants store selenium in opposite ways
The paper’s plant comparison is why it matters beyond analytics. In wheat — the non-hyperaccumulator — selenate and selenite are assimilated into selenomethionine and selenocysteine that are non-specifically incorporated into proteins; liberating them for measurement needs protease digestion. In Cardamine hupingshanensis — the serpentine-outcrop Brassicaceae whose shoots carry Se far past the 1,000 mg kg⁻¹ hyperaccumulation line (the authors cite up to 8,000 mg kg⁻¹ dry matter from Wang et al., 2024) — Se is kept instead as non-protein-bound free selenoamino acids, chiefly methylselenocysteine and selenolanthionine (SeLan), in what is effectively a storage-plus-detox strategy. Roughly 72.06% of the plant’s total Se was water-soluble. Apply the wheat protocol to the hyperaccumulator and the measurement collapses: the one-step enzymatic method returned just 5.06 ± 0.28 µg Se g⁻¹ of SeLan, because the protease environment — pH shifts, ionic strength, enzyme-substrate interactions — degrades or transforms the labile free pool. A two-step protocol (water first, then enzymatic hydrolysis of the protein-bound residue) recovered 143.74 ± 2.70 µg Se g⁻¹, with SeLan found almost exclusively in the aqueous fraction. Even the buffer choice matters: Tris-HCl at pH 4.5 cut SeLan from 157.30 ± 8.14 to 42.72 ± 3.74 µg Se g⁻¹ versus plain water. For wheat the logic inverts: total recovery climbed from 34.11% (probe sonication) to 52.21% (water bath) with gentler one-step enzymatic handling, and moving hydrolysis into a nitrogen glove box (<5 ppm O₂) to block SeMet oxidation raised wheat SeMet from 98.73 ± 4.50 to 112.17 ± 7.87 µg Se g⁻¹ (a 13.61% gain) and the hydrolysate recovery from 80.79% to 90.02%.
What this means for the field
The non-obvious implication is about the literature behind the literature. If a hyperaccumulator’s diagnostic compounds — precisely the free selenoamino acids used to argue these plants are safer Se sources than SeMet-loaded supplements — are largely destroyed by the standard extraction, then published speciation profiles made with one-step enzymatic digestion are not merely imprecise; they may be structurally blind to the compounds that define hyperaccumulator metabolism. This is commercially live, not just academic: C. hupingshanensis is already being commercialized as an edible Se source and has been trialled as a feed additive for Se-enriched meat and eggs, and the paper notes that contradictory speciation reports exist in exactly this product space. A recovery below 35% — the authors’ own starting point for both matrices — means two-thirds of the speciation story is extrapolation. The same caution extends to the selenium-hyperaccumulator relatives in the database, Stanleya pinnata and the Astragalus accumulators, whose public profiles deserve the same matrix-specific scrutiny before their speciation is quoted.
The genuine risk, bounded
Three honest limits. First, the scale: this is one hyperaccumulator species, one crop, and one element, in a lab methods paper — a demonstration that the bias exists and matters, not a census of how big it is across the hyperaccumulator literature. Second, the headline recovery pairs need precise attribution: the abstract’s “20.82% → 80.65%” for C. hupingshanensis is a worst-to-best span whose 20.82% start point appears only in the abstract; the Results’ own stepwise improvements are 30.20% → 70.97% for the two-step (aerobic) procedure and 71.57% → 80.65% once anaerobic handling is added. For wheat, 34.11% is the probe-sonication baseline, not the direct “before” of 90.02%. Third, the authors are candid that why SeLan is lost “requires further investigation” — the mechanism of destruction is inferred from recovery gaps, not demonstrated, and a <5 ppm O₂ glove box is a meaningful barrier to routine adoption in the food labs this method targets.
For context, see the site’s selenium hub and the Cardamine hupingshanensis species page — where our database record (maximum 1,965 mg Se kg⁻¹, after Yuan et al. 2013) deliberately reports the measured primary record rather than the higher secondary figure the authors cite.
Source: Duan, Du, Li, Zhang, Zhang & Yuan (2026), Food Chemistry: X 39:104410, DOI 10.1016/j.fochx.2026.104410 (gold open access, CC BY-NC; full text retrieved 2026-09-23 from Europe PMC, PMC13572759). Every load-bearing number — the SeLan pair 5.06 ± 0.28 → 143.74 ± 2.70 µg Se g⁻¹, the 72.06% water-soluble fraction, the Tris-HCl depression 157.30 → 42.72 µg Se g⁻¹, the wheat SeMet pair 98.73 ± 4.50 → 112.17 ± 7.87 µg Se g⁻¹ and its 13.61% increase, the 34.11 → 52.21% and 80.79 → 90.02% recovery steps, the 30.20 → 70.97% and 71.57 → 80.65% cardamine steps, and the <5 ppm O₂ glove-box spec — was verified verbatim by three independent reviewers on three different model vendors, each retrieving the primary source independently; no numeric discrepancies were found. The reviewers’ flags are disclosed in the text: the abstract’s 20.82% starting figure appears only in the abstract (not traceable to a Results sentence), the abstract recovery pairs are cross-protocol composites (the stepwise pairs above are quoted instead), the 8,000 mg kg⁻¹ maximum is a cited secondary figure (Wang et al., 2024, with Wu et al., 2020 and Yuan et al., 2013) rather than the primary record, and the ~2.5 min SAX void-elution figure is literature-general (the study’s own co-eluting peak is at 2.3 min). See /methodology/ for how we source and check analysis pieces.