Subašić, Selović, Dahija, Demir, Samardžić, Bonomo, Rigano, Giosa and Karalija — University of Sarajevo with collaborators in Italy — report in Plants (January 2026, open access) that priming seeds of the Balkan campion Silene sendtneri with the plant-growth-promoting bacterium Paraburkholderia phytofirmans PsJN leaves a measurable “transcriptomic memory” in the seed: thousands of genes are already re-programmed before any metal appears, and the resulting seedlings tolerate — and accumulate — cadmium that kills their untreated siblings.
What they found, in numbers
The group primed seeds with PsJN, sequenced their RNA before any stress, then grew seedlings in vitro on ½-MS agar with CdCl₂ at 0, 0.25 and 0.50 mM. Primed seeds carried 5,509 differentially expressed genes relative to non-primed seeds, with detoxification machinery — chloroplastic glutathione reductase (log₂FC ≈ 9.2), class III peroxidases, proline-biosynthetic P5CS (log₂FC ≈ 6.8), ABC and CAX2-type transporters — switched on in advance.
The physiological payoff is large. Where non-primed plants failed to develop at all at 0.50 mM Cd, primed plants still produced biomass, and at 0.25 mM they actually grew better than their own unstressed controls (shoot fresh weight 15.62 mg vs 12.59 mg). Cadmium in whole-plant dry biomass reached 854.46 mg kg⁻¹ in primed plants vs 100.41 mg kg⁻¹ in non-primed at 0.25 mM, and 1,412.50 mg kg⁻¹ at 0.50 mM — roughly fourteen times the conventional 100 mg kg⁻¹ hyperaccumulation benchmark, albeit on a whole-plant, in-vitro basis. Root peroxidase activity in primed plants hit 6,435 U mg⁻¹ protein min⁻¹ at 0.25 mM against 7.65 in non-primed controls — a peroxidase-centered defense that was already constitutively elevated before stress. Primed plants also held their micronutrients: at 0.25 mM Cd, non-primed Fe and Zn collapsed to roughly a third of control values, while primed plants kept Fe at 197.52 and Zn at 153.32 mg kg⁻¹.
The non-obvious implication: the bacterium is a ghost
The usual pitch for PGPR-assisted phytoremediation is that the bacterium colonizes the rhizosphere and keeps working in the field — a live inoculant that must establish, compete and persist in hostile, metal-laden soil. The authors report, citing their own earlier work with the same protocol, that PsJN does not persist as an active endophyte after seed priming: the wash steps strip the exopolysaccharides the bacterium needs to penetrate tissue. If the benefit lives in the seed’s rewritten transcriptional state rather than in a living symbiosis, the deployment model inverts. A nursery or seed supplier can apply the treatment under clean, controlled conditions, ship “pre-armed” seed, and no one has to keep a bacterial population alive in the target soil at all. That removes the biggest practical objection to bioinoculants in remediation — the notorious gap between greenhouse colonization success and field persistence — and it suggests the same trick could upgrade any hyperaccumulator whose seed can be primed, not just this one species.
The sober ledger
This is one seed lot, one locality — a calcareous site at Pjeskovita ravan in central Bosnia, notably not a metalliferous soil — and an in-vitro assay with no soil matrix, where the authors themselves acknowledge that bioavailability, and therefore tissue Cd, is inflated relative to the field. Shoot Cd was never measured separately; the hyperaccumulation arithmetic rests on whole-plant biomass, which the canonical criterion does not use. The “memory not colonization” interpretation is asserted from prior work, not tested here — colonization was not monitored in this study. And primed plants accumulate more Cd, which is exactly what a phytoextraction operator wants and exactly what a land manager must plan for in biomass handling.
There is also a data-hygiene problem a good reviewer should not skip. The discussion claims primed plants accumulate carotenoids “several tens of times higher” than controls; the paper’s own Table 6 supports 1.6–4.3×, and primed plants at 0.50 mM fall below the non-primed control. The differential-expression arithmetic does not sum (3,188 up + 2,320 down = 5,508, not 5,509). Two unrelated assays share identical values to two decimal places (BP 0.25 mM chlorophyll a and shoot proline both print 15.11 ± 0.02; BP 0.50 mM both print 9.04). Table 7’s header labels the treatment series in µM where every other table, and the methods, use mM. And the de novo seed transcriptome on which all gene-level claims rest is unusual: 71.95% of its 393,920 assembled transcripts are called non-coding, and the reported BUSCO coverage is 3.01% of mRNAs — a completeness signal low enough that the differential-expression lists deserve replication on a reference-grade assembly before they are treated as the mechanism. Read the physiology with those caveats, and the honest conclusion is: a real, well-documented priming effect at the whole-plant level, with a molecular mechanism that is suggestive but not yet load-bearing.
Why it still matters for the cadmium problem
Cadmium is the metal where phytoextraction most needs cheap multipliers: it is diffuse, food-chain-critical, and rarely concentrated enough to justify engineered soil washing (see the /cadmium hub for the element’s remediation context). A seed treatment that survives the supply chain — no cold chain, no live-culture shelf life — and that simultaneously protects the plant’s Fe and Zn economy while doubling uptake is precisely the kind of low-cost leverage the field keeps saying it needs. The Bosnian group’s next experiment is obvious: the same primed seed in real, Cd-contaminated soil, with shoot-specific analysis and tracked bacterial persistence. Until then, treat the transcriptome as a hypothesis and the biomass numbers as the finding.
Provenance: full open-access text read (MDPI HTML + PDF, all tables extracted); every load-bearing number independently re-retrieved and verified against the primary source by three separate reviewers with zero discrepancies; the paper’s own internal inconsistencies (carotenoid overstatement, DEG arithmetic, duplicated table values, µM/mM unit typos) were surfaced and are reported above rather than smoothed over; methods and limits at /methodology/.