This study builds on previously published RNA-seq data (Nowicka et al., 2024) from two barley (Hordeum vulgare L.) cultivars: Igri (winter type; HOR 10596; ME-responsive) and Golden Promise (spring type; HOR 16645; ME-recalcitrant). Detailed procedures for plant cultivation, ME induction, microspore isolation, and data processing are provided in Nowicka et al. (2024).

Four stages of ME-induction were analysed (Figure 1C):

Stage 0: Microspores isolated from freshly harvested tillers.

Stage I: Microspores isolated from anthers pre-treated with 0.4 M mannitol for 48 h at 21 °C.

Stages II–III: Microspores isolated from anthers cultured in KBP medium (Kumlehn et al., 2006) for 24 h (stage II) or 48 h (stage III) following the same mannitol pre-treatment.

Because Golden Promise is recalcitrant to ME induction, biochemical stimulation was provided by co-culture with immature pistils (+p) of wheat cultivar Bobwhite (Lippmann et al., 2015). Longitudinally bisected pistils (three halves per ml KBP medium) were added to isolated Golden Promise microspores, and stage II–III RNA-seq samples were collected after co-culture. Igri did not undergo co-culture. ME-induction efficiency and sample purity were assessed microscopically prior to RNA-seq (Daghma et al., 2014). Samples for RNA-seq were collected using a mannitol/maltose density-gradient method. The gradient was applied once for stage 0 and twice for stages I–III; stage III fractions were additionally sieve-filtered.

Bicellular pollen grains produced by asymmetric division were used as a gametophytic control. Pollen was isolated from freshly harvested tillers: cells were first enriched by a mannitol/maltose density gradient and then individually picked using a glass micropipette mounted on a micromanipulator under an inverted microscope, coupled to a microinjector for precise aspiration. Pollen RNA-seq data are unpublished (Kopeć et al., 2025).

In brief, RNA-seq was performed with four biological replicates per ME induction stage and cultivar (32 samples in total) and three biological replicates per cultivar for bicellular pollen (6 samples). Total RNA was extracted using the NucleoSpin RNA Plant Kit (740949.50; Macherey-Nagel, Düren, Germany). Samples with RNA integrity values (RIN) > 6.0 were used to prepare poly(A)-selected mRNA libraries (NEBNext Ultra™ RNA Library Prep Kit for Illumina). Libraries were sequenced as 150-bp paired-end reads on an Illumina NovaSeq platform (Genewiz). Raw reads were adapter-trimmed using Trim Galore (v0.4.1) and aligned to the H. vulgare cultivar Morex reference genome v2 (Monat et al., 2019) using HISAT2 (v2.1.0). Gene-level read counts were obtained with Subread/featureCounts (v1.5.2) using the corresponding genome annotation. Differential expression analysis was performed on raw count data using DESeq2 (v1.24.0) in R (v3.6.3) with the Wald test under a negative binomial model. P-values were adjusted for multiple testing using the Benjamini–Hochberg procedure, and genes with FDR < 0.05 were considered differentially expressed. Transcript abundance was reported as FPKM, and log₂ fold changes (log₂FC) were calculated by DESeq2.

The RNA-seq dataset is publicly available in the NCBI Gene Expression Omnibus (GSE233486).

To investigate redox‐associated transcriptional responses during ME induction, we have completed the RNA-seq dataset with a targeted set of antioxidant and redox-related genes. Candidate functions were curated manually from biochemical pathway knowledge and primary literature (Mittler, 2002; Foyer and Noctor, 2011; Sevilla et al., 2023). In total, 472 genes (Figure 1B) were curated and assigned to four functional categories:

i) core antioxidant genes: SUPEROXIDE DISMUTASES (SODs), CATALASES (CATs), GLUTATHIONE PEROXIDASES (GPXs);

ii) ASC–GSH cycle genes: ASCORBATE PEROXIDASES (APXs), MONODEHYDROASCORBATE REDUCTASES (MDHARs), DEHYDROASCORBATE REDUCTASES (DHARs) and GLUTATHIONE REDUCTASE (GRs);

iii) redox-regulated thiol genes: THIOREDOXINS (TRXs), THIOREDOXIN REDUCTASE (TRXRs), GLUTAREDOXINS (GRXs), PEROXIREDOXINS (PRXs);

iv) genes involved in S-glutathionylation: GLUTATHIONE S-TRANSFERASES (GSTs).

Candidate genes were initially retrieved from the H. vulgare Morex v2 annotation (Monat et al., 2019) using keyword- and function-based searches. When gene families were incompletely annotated or ambiguous, orthologous Arabidopsis thaliana protein sequences were used as queries for manual homology searches in EnsemblPlants to identify the corresponding barley genes.

Stage-specific markers were genes that, at a given induction stage, met all of the following criteria: (i) FPKM > 10 in both cultivars; (ii) no significant cultivar effect at that stage (FDR-adjusted P > 0.05); (iii) log₂FC > 2 relative to the subsequent stages; and (iv) absent or very low expression in the gametophytic control (bicellular pollen, P). Cultivar-specific markers were genes that, at a given stage, met: (i) FPKM > 5; (ii) a significant difference between cultivars (FDR-adjusted P < 0.05); (iii) log₂FC > 2 relative to the second cultivar at the same stage; and (iv) absent or very low expression in P. Genes that did not meet these thresholds but showed directionally consistent patterns are reported as trend-level candidates for stage or cultivar specificity.

For cross-study validation, thiol-related and GST genes reported for the ME-responsive cultivar Gobernadora (Bélanger et al., 2018), originally annotated against the Morex v1 (Mascher et al., 2017) reference genome, were mapped to Morex v2 identifiers using reciprocal BLASTP searches using EnsemblPlants database.

Gene filtering, selection, and summarization were performed in R (v4.2.2) using custom scripts. Heatmaps were generated with Heatmapper (http://heatmapper.ca/expression/). Matrix bubble charts and additional visualizations were produced using ggplot2.

Using the published dataset of Nowicka et al. (2024), we assessed oxidative-stress-related transcriptomic responses across four ME-induction stages in two barley cultivars, Igri (responsive) and Golden Promise (recalcitrant): stage 0 (isolation from freshly harvested tillers), stage I (48 h in 0.4 M mannitol at 21 °C), and stages II–III (24 h and 48 h after transfer to KBP, respectively). Each stage comprised mixed cell populations with cultivar-specific composition (Figure 1C; Supplementary Figure S1). At stage 0, Igri was enriched for uninucleate microspores (73%) with fewer bicellular structures after symmetric division (17%). Golden Promise contained fewer uninucleates (14%) and more bicellulars (21%) whereas a high proportion of microspores remained unidentified. Stage I showed comparable reprogramming (≈30% symmetric divisions), but uninucleates remained more frequent in Igri (68%) than in Golden Promise (22%). Across stages II–III, Golden Promise maintained ≈30% bicellular structures, while multicellular structures—arising from continued symmetric divisions—increased from 3% at stage II to 18% at stage III. In Igri, bicellulars remained ≈40% across stages II–III, with multicellular structures at 12% in stage II and 6% in stage III. These differences were more pronounced in cultures lacking the stimulatory effect of co-cultured pistils from the highly embryogenic wheat cultivar Bobwhite (−p; Figure 1C; Supplementary Figure S1). This staged cytology provides the framework for the antioxidant/redox transcriptomic analyses that follow.

We first profiled genes encoding the core antioxidant enzymes that constitute the primary defense against ROS — SODs, CATs and GPXs (Figure 2, Supplementary Figure S2). Of 21 annotated SOD genes, 15 were transcriptionally active (FPKM > 0) in at least one induction stage, together with five of nine CATs and all five GPXs (Supplementary Figure S2A). SODs spanned a wide expression range: in Golden Promise, most transcripts were moderate (10–100 FPKM) with few highly expressed (>100 FPKM), whereas in Igri expression was more evenly split between low (1–10 FPKM) and moderate levels. CATs were generally weakly expressed in both cultivars, while GPXs were consistently stronger, predominantly at moderate–high levels (Supplementary Figure S2B). Most expressed SODs belonged to Cu/Zn-SOD orthogroups (Arabidopsis CSD1/2/3), with fewer chloroplastic Fe-SODs and a single mitochondrial MnSOD. Two CATs aligned with CAT2 and several GPXs with GPX1/6 (Supplementary Dataset 1).

Family-level averages revealed distinct trends (Figure 2A, Supplementary Figure S2C). Mean SOD expression ranged from ~40 FPKM (Igri, stage II) to ~60 FPKM (Golden Promise, stage I), with maxima at stage 0 in Golden Promise and stage III in Igri; bicellular pollen (P; gametophytic control) showed SOD levels comparable to induction stages. CATs averaged 12–28 FPKM and peaked at stage III in both cultivars. GPXs exceeded SODs and CATs overall, tended to be higher in Golden Promise, and peaked at stage III (~180 FPKM). In contrast to induction stages, CAT and GPX transcripts were low in mature pollen.

Gene-level profiles highlighted discrete regulatory modes (Figures 2A, B). Among SODs, some transcripts were induced early (stages 0–I; e.g. MnSOD HORVU.MOREX.r2.2HG0173140.1; Cu/Zn-SOD HORVU.MOREX.r2.2HG0154740.1), whereas others rose later (stages I–III; e.g. Cu/Zn-SODs HORVU.MOREX.r2.4HG0329510.1 and HORVU.MOREX.r2.6HG0513460.1—the latter strongly upregulated in both cultivars at stages II–III with FPKM > 10 but not detected in mature pollen, supporting its candidacy as an ME-stage marker). Additional cultivar specificity was evident: Cu/Zn-SOD HORVU.MOREX.r2.7HG0573050.1 was moderate at stages 0–II in both cultivars but became Igri-specific at stage III, whereas HORVU.MOREX.r2.3HG0190910.1 was confined to Igri.

Three CATs showed clear stage specificity: HORVU.MOREX.r2.1HG0067700.1 at stage 0 (uninucleate microspores), HORVU.MOREX.r2.2HG0109480.1 at stage I (osmotic treatment), and HORVU.MOREX.r2.4HG0341470.1 at stage III, with higher expression in Golden Promise. Another CAT (HORVU.MOREX.r2.7HG0623480.1) increased progressively across Stages II–III in both cultivars and served as an II–III marker. All CATs were very low in mature pollen (Figures 2A, B).

GPXs also exhibited stage- and cultivar-dependent regulation: HORVU.MOREX.r2.4HG0304790.1 peaked at stages I and III, and HORVU.MOREX.r2.2HG0156400.1 marked Golden Promise at stage III. As with CATs, GPX transcripts were low in mature pollen (Figures 2A, B).

Collectively, core antioxidant genes show dynamic regulation during ME, with stage- and cultivar-specific patterns that distinguish the responsive Igri from the recalcitrant Golden Promise.

We next analyzed APX, MDHAR, and GR families (Figure 3, Supplementary Figure S3). A distinct DHAR gene family is not annotated in the barley Morex v2 genome; homology-based searches indicated that DHAR-like functions are represented by GST-annotated genes (Supplementary Dataset 1). Most genes were expressed (APX: 4/5; MDHAR: 5/7; GR: 2/2; 14 total; Supplementary Figure S3A), mapping to orthogroups containing the corresponding Arabidopsis genes (Supplementary Dataset 1).

At the family level, APX genes exhibited the highest mean expression among all antioxidant gene families, maintaining consistently high transcript abundance across induction stages and in pollen in both cultivars (≈300–500 mean FPKM). Notably, a single APX transcript (HORVU.MOREX.r2.2HG0088330.1) displayed very high and stable expression across all ME stages and in pollen, reaching ≈1000 FPKM (Figures 3A, B, Supplementary Figure S3B). In contrast, MDHARs increased progressively from stage 0 to stage III, particularly in Igri (up to ~77 FPKM), with pollen lower than induction stages. GRs were expressed at moderate levels (~60 FPKM) across ME stages and in pollen, except for a marked rise in Igri at stage III (~115 FPKM).

Gene-specific patterns reinforced these trends (Figures 3A, B). Within APX, HORVU.MOREX.r2.4HG0320930.1 was high in Igri at stage 0 and re-emerged at stage III, showing stage-III specificity in both cultivars. Several MDHARs (HORVU.MOREX.r2.6HG0503910.1, HORVU.MOREX.r2.7HG0581330.1, HORVU.MOREX.r2.7HG0571990.1) showed progressive induction marking stages II–III, whereas HORVU.MOREX.r2.6HG0503900.1 peaked at stage I in both cultivars. Among GRs, HORVU.MOREX.r2.6HG0521730.1 emerged as a stage III–specific marker of Igri. In mature pollen, APXs expression was high in one or both cultivars, whereas MDHARs and GRs were low.

In summary, APX transcripts are abundant but largely stage-stable, MDHARs rise with ME progression, and GR showed an Igri-biased stage-III induction.

We examined also the transcriptional activity of thiol–redox regulatory families (Figure 4, Supplementary Figure S4). Of 143 annotated genes, most were expressed in at least one stage: 57/63 TRXs, 37/56 GRXs, all 6 TRXRs and all 8 PRXs (Supplementary Figure S4A). TRX and GRX transcripts generally accumulated at low–to–moderate levels, with few highly expressed members; TRXR and PRX showed similar distributions (Supplementary Figure S4B).

At the family level, mean expression was relatively stable across stages 0–III and in pollen, with only minor cultivar differences (Figure 4A, Supplementary Figure S4C). TRXR was notably invariant: all six members showed no clear induction with ME or genotype effects (Supplementary Dataset 1).

In contrast, several TRXs and GRXs were selectively mobilized at later stages (II–III), particularly in Igri. Representative trajectories illustrated these contrasts (Figures 4A, B). For example, TRXs HORVU.MOREX.r2.1HG0023970.1, HORVU.MOREX.r2.4HG0315890.1, and HORVU.MOREX.r2.2HG0091330.1 exhibited biphasic dynamics: high at stage 0, decreased at stages I–II, then re-induced at stage III—Igri-specific for HORVU.MOREX.r2.1HG0023970.1, shared by both cultivars for the other two. Strong cultivar biases were evident: HORVU.MOREX.r2.7HG0534000.1 was consistently higher in Golden Promise, whereas HORVU.MOREX.r2.2HG0174890.1 was higher in Igri across stages. Marker-like behavior included Igri-specific induction at stages II–III (HORVU.MOREX.r2.2HG0093810.1) and stage-III-restricted increases (HORVU.MOREX.r2.5HG0432930.1, HORVU.MOREX.r2.7HG0560380.1). Among GRXs, HORVU.MOREX.r2.2HG0125570.1 acted as a stage-III marker in both cultivars, HORVU.MOREX.r2.7HG0552610.1 was enriched in Golden Promise at stage III, and HORVU.MOREX.r2.3HG0230900.1 was Igri-stage III marker. Several transcripts (HORVU.MOREX.r2.6HG0496700.1, HORVU.MOREX.r2.2HG0095970.1, HORVU.MOREX.r2.1HG0072330.1) accumulated progressively across induction. Within PRXs, HORVU.MOREX.r2.6HG0476250.1 was stage-I specific in both cultivars, whereas HORVU.MOREX.r2.3HG0231680.1 showed clear Igri specificity at stage III.

To validate these patterns, a cross-study comparison was performed with the ME-responsive cultivar Gobernadora, in which three TRX and two GRX genes were reported as upregulated at day 5 of culture (Bélanger et al., 2018). The expression profiles of these candidates were therefore examined in the Igri and Golden Promise datasets. Two TRXs showed Igri stage III-specific accumulation, whereas the third was induced in both cultivars across stages I–III; similarly, both GRXs displayed increased expression in both cultivars across stages I–III (Figure 4A, Supplementary Dataset 1).

Notably, a subset of TRX, GRX and PRX genes was not ME-responsive but was strongly expressed in mature pollen, indicating developmental rather than reprogramming regulation (Supplementary Figures S4D, E).

Hence, although family-level expression appears stable, distinct TRX/GRX/PRX members are selectively deployed in a stage- and cultivar-dependent manner, with late (II–III) induction particularly prominent in Igri.

To assess the role of GST during ME induction, we profiled the expression of 280 annotated GST genes (Figure 5, Supplementary Figure S5). Exactly half of these genes (140/280) were transcriptionally active (FPKM > 0) in at least one developmental stage in either cultivar (Supplementary Figure S5A). The expressed GST genes are distributed across various classes, with a particularly strong representation of the theta class (GSTT3, n = 41, Supplementary Dataset 1). Most transcripts accumulated at very low to low levels (0–10 FPKM), with relatively few reaching moderate (10–100 FPKM) or high (≥100 FPKM) abundance. Notably, the responsive cultivar Igri contained more strongly expressed GSTs at stage III than Golden Promise (Supplementary Figure S5B).

Mean GST expression increased progressively from stage 0 to stage III in both cultivars, peaking in Igri at stage III (~40 FPKM). Pollen (P) showed low GST expression, comparable to stage 0 (Figure 5A, Supplementary Figure S5C).

Closer inspection revealed diverse expression trajectories (Figures 5A, B). Several transcripts were preferentially induced in Golden Promise at stage I (e.g. HORVU.MOREX.r2.7HG0529160.1, HORVU.MOREX.r2.3HG0267220.1) or at stages II–III (e.g. HORVU.MOREX.r2.6HG0520240.1). HORVU.MOREX.r2.1HG0013750.1 was the most highly expressed GST overall, with strong accumulation at stages II and III in both cultivars. Additional common stage markers included genes specifically induced at stage III (e.g. HORVU.MOREX.r2.7HG0619090.1, HORVU.MOREX.r2.5HG0383770.1). A substantial subset was enriched in Igri at stages II–III, with maxima at stage III: some maintained high expression across both stages (e.g. HORVU.MOREX.r2.7HG0551820.1, HORVU.MOREX.r2.3HG0260240.1), whereas others showed a stepwise rise from moderate (stage II) to strong (stage III) levels (e.g. HORVU.MOREX.r2.1HG0040480.1, HORVU.MOREX.r2.1HG0016790.1, HORVU.MOREX.r2.1HG0040620.1, HORVU.MOREX.r2.1HG0016800.1). A final group acted as Igri-specific stage-III markers (e.g. HORVU.MOREX.r2.2HG0178470.1).

Similarly, cross-study comparison with the ME-responsive cultivar Gobernadora supported the stage specificity observed here (Bélanger et al., 2018). Seven GSTs reported as upregulated at days 0–2 of culture were examined in the present dataset and were found to be associated primarily with early induction: three were upregulated at stage I in both cultivars but accumulated more strongly in Golden Promise, whereas four showed elevated expression across stages 0–II; in all cases, expression declined at stage III. In contrast, of 19 GSTs reported as upregulated at day 5, thirteen displayed clear Igri stage III-enrichment, including seven genes with strict marker-like behavior. This concordance was taken as independent validation of the dataset and supported the association of late GST activation with responsive genotypes (Figure 5A, Supplementary Figure S5D, Supplementary Dataset 1).

Together, these results indicate that GST activity intensifies during the transition to embryogenesis, with a pronounced stage-III induction in the responsive cultivar Igri.

Our survey revealed a broad antioxidant repertoire in barley microspores: 35 core antioxidant genes, 14 ASC–GSH cycle genes, 143 TRX/GRX/PRX genes, and a large GST family (280 members). A similarly large set of 330 TaGSTs was reported in wheat (Wang et al., 2019). Approximately 80% of these genes were transcriptionally active during ME induction. Significant expression changes were detected for nearly 80% of core antioxidant and TRX/GRX genes, ~90% of ASC–GSH genes, and ~70% of GSTs—evidence of extensive redox rewiring during induction.

Among core enzymes, SODs emerged as prominent candidates. Most annotated SODs corresponded to Arabidopsis Cu/Zn isoforms localized to the cytosol, chloroplasts, or peroxisome. These isoforms were also the most abundant in anthers of triticale subjected to ME-inducing cold pre-treatment (Żur et al., 2014a). They are known to function in development and respond strongly to temperature, drought, and salinity (Gill and Tuteja, 2010; Wang et al., 2016; Li et al., 2017). Genotype-biased differences were clearest at stage III, where several SOD transcripts (e.g. HORVU.MOREX.r2.7HG0573050.1) accumulated more strongly in the responsive cultivar Igri. Similarly, increased accumulation of SOD transcripts was observed in Brassica napus microspores redirected towards ME (Rueda-Varela et al., 2025).

Functionally, SODs catalyze the dismutation of O₂^•– to H₂O₂, thereby limiting the formation of highly reactive ^•OH via the Haber–Weiss reaction (Mittler, 2002). Being less reactive, H₂O₂ acts as a signaling molecule that modulates gene expression and enzyme activity through cysteine oxidation-based redox regulation (Souza, 2025). Our previous work supports a role for H₂O₂ as a trigger for ME as excessive H₂O₂ elimination improved microspore survival but reduced reprogramming efficiency (Żur et al., 2014a; Żur et al., 2021a). It suggests that maintaining a signaling-competent H₂O₂ pool is essential for successful ME induction. However, subsequent studies in triticale and barley demonstrated that both O₂^•– and H₂O₂ accumulate in the microspore cytoplasm after ME-inducing treatment (Żur et al., 2021b). Notably, enhanced generation of O₂^•– was detected in proximity to the nuclei, pointing to its potential role in early signaling events. This interpretation is consistent with recent insights summarized by Karpinska and Foyer (2024), who highlight a previously underappreciated function of superoxide in compartment-specific signaling within the plant cell nucleus.

Consistently, CATs showed stage-specific expression—distinct isoforms predominating at different phases—whereas many GPXs (with higher affinity for H₂O₂ and frequently implicated in signaling) were induced by mannitol and further upregulated at 48 h of in vitro culture. Higher CAT and GPX transcript abundance in Golden Promise (e.g. CAT HORVU.MOREX.r2.4HG0341470.1; GPX HORVU.MOREX.r2.2HG0156400.1) may reflect over-scavenging that dampens H₂O₂ signaling required for the developmental switch, consistent with the negative non-linear relationship between CAT activity and ME efficiency reported in triticale (Żur et al., 2014a).

Low-molecular-weight antioxidants ASC and GSH are abundant and widely distributed in plant cells. They directly detoxify ROS, protecting cell components from oxidative damages and help maintain the cellular redox environment required for effective metabolism and development. Their oxidized forms (MDHA, DHA, GSSG) are efficiently recycled in the ASC–GSH cycle due to concerted action of reducing enzymes (MDHAR, DHAR and GR; Foyer and Noctor, 2011). Two observations stood out. First, several MDHAR transcripts increased progressively during induction—either similarly in both cultivars or with a stronger response in Igri (e.g. HORVU.MOREX.r2.6HG0503910.1 and HORVU.MOREX.r2.7HG0581330.1)—consistent with sustained ASC recycling during reprogramming. Second, one GR (HORVU.MOREX.r2.6HG0521730.1) showed a cultivar-biased induction at stage III, consistent with prior evidence that elevated GR activity is associated with microspore competence for ME (Żur et al., 2021b). In contrast, APX transcripts were highly abundant but largely stage stable in both cultivars, suggesting a constitutive role in maintaining basal H₂O₂ detoxification during both gametophytic and embryogenic development rather than driving stage-specific transitions.

Together, these patterns support a framework in which ASC-dependent redox buffering primarily supports microspore and pollen survival in both cultivars, whereas efficient GSH recycling becomes especially important in rapidly dividing, embryogenic cells of Igri. Supporting this view, a recent report showed that a local GSH burst in wounded Arabidopsis roots shortens G1 to accelerate division and regeneration (Lee et al., 2025), offering a plausible mechanistic link. Consistently, exogenous GSH in our earlier works maintained microspore viability and stimulated ELS development (Żur et al., 2019; Żur et al., 2021b).

Thiol oxidoreductases further integrate redox control with development. TRXs and GRXs—abundant and broadly localized—regulate protein redox state, reduce disulphide and mixed disulphide bonds using ferredoxin/NADPH (TRX systems) or GSH (GRX), and can donate electrons to other antioxidant enzymes (Meyer et al., 2009; Meyer et al., 2012; Sevilla et al., 2023). They also participate in retrograde communication that coordinates gene expression during stress acclimation (Sevilla et al., 2023). We observed selective late (stages II–III) induction of specific TRX and GRX members, particularly in Igri, consistent with roles in supporting rapid divisions and stabilizing redox-sensitive enzymes during multicellular structures formation. Notably, this cultivar-biased late activation was independently supported by comparison with the ME-responsive cultivar Gobernadora, in which several of the same thiol–redox genes were previously reported as upregulated during late induction (Bélanger et al., 2018), providing cross-study validation of the observed patterns. Among PRXs, only one gene (HORVU.MOREX.r2.3HG0231680.1; a B-type/1-Cys peroxiredoxin orthologue of Arabidopsis) showed a clear association with efficient ME, being strongly induced at stage III in Igri and reaching high abundance (>200 FPKM). Plant 1-Cys PRXs, though less characterized, interact with TRX/GRX systems to mitigate ROS, transduce stress signals, and modulate metabolism, under severe stress their peroxidase activity can switch to a chaperone function (Dietz et al., 2006; Kim et al., 2011; Liebthal et al., 2018). Its strong induction in the responsive cultivar points to a role in proteostasis during rapid proliferative remodeling.

In addition to ME-associated induction, a substantial subset of TRX/GRX/PRX genes was strongly expressed in mature pollen but showed little or no responsiveness during ME induction, indicating that thiol–redox regulation also supports late gametophytic development independently of embryogenic reprogramming. This separation of expression patterns suggests functional specialization within thiol–redox families, with distinct members contributing either to pollen maturation and stress tolerance or to the redox remodeling required for embryogenic transition.

GSTs likely contribute to ME through multiple mechanisms. They catalyzing S-glutathionylation—a reversible modification that shields protein thiols from irreversible oxidation while modulating protein function (Sevilla et al., 2023). GSTs detoxify stress-derived metabolites and participate in biosynthetic pathways (Piślewska-Bednarek et al., 2018). Although some plant GSTs show limited glutathionylating activity (Micic et al., 2024), the family’s size underscores functional breadth. In the present study, GST transcription was progressively activated during ME induction, with a pronounced enrichment at stage III in the responsive cultivar Igri. This pattern closely parallels observations made in the independently characterised ME-responsive cultivar Gobernadora (Bélanger et al., 2018), in which late-stage GSTs induction was likewise associated with successful embryogenic development. The concordant late GST activation in both Igri and Gobernadora therefore confirms their shared responsive phenotype and provides cross-study validation that stage-III GST upregulation is a hallmark of effective embryogenic reprogramming.

Early studies identified GSTs among the first ME-responsive genes in barley (Vrinten et al., 1999), and subsequent studies documented dynamic GST expression across induction and its association with plant regeneration capacity (Maraschin et al., 2006; Muñoz-Amatriaín et al., 2006, Muñoz-Amatriaín et al., 2009; Joosen et al., 2007; Malik et al., 2007; Tsuwamoto et al., 2007; Jacquard et al., 2009; Sánchez-Díaz et al., 2013; Żur et al., 2014b; Bélanger et al., 2020). Our findings extend these observations by resolving stage- and cultivar-specific programs and implicating elevated GST activity as a hallmark of the responsive trajectory.

Integrating transcript abundance patterns across antioxidant and redox gene families with their functional relationships (Figure 6), we propose a staged model of redox control during ME. An early phase, common to both cultivars, is characterized by activation of core antioxidant defenses that buffer the oxidative burst associated with microspore isolation and osmotic/starvation stress, thereby enabling cell survival and entry into developmental reprogramming. This phase is marked by stable or moderately elevated expression of SODs, APXs, and selected CAT and GPX isoforms, consistent with maintenance of basal ROS detoxification and redox homeostasis.

In contrast, the embryogenically responsive trajectory represented by Igri shows a second, coordinated redox reinforcement at later stages (II–III), coincident with the initiation of the embryogenic developmental program. This late phase involves enhanced glutathione recycling via GR, pronounced activation of GSTs, and selective deployment of thiol–redox regulators (TRXs, GRXs, and PRXs), rather than a global increase in antioxidant capacity (highlighted in Figure 6). Such targeted engagement of thiol-based redox hubs is well suited to fine-tune protein redox status, preserve proteostasis, and sustain redox-sensitive signaling pathways during rapid cell proliferation and multicellular structure formation. We propose that this program maintains a signaling-competent O₂^•– and H₂O₂ pools while preventing oxidative damage, thereby favoring cell-cycle progression and ELS development. Intensive utilization of GSH as an electron donor by GST, GRX, and GPX enzymes may shift the cellular redox balance toward a more oxidized state conducive to continued embryogenic development (Stasolla, 2010). By contrast, the recalcitrant cultivar Golden Promise may over-scavenge or mistime key steps, attenuating ROS signaling required for the fate switch.

The presented model appears to be supported by multiple experimental data (as referenced above); however, further validation could be achieved via (i) targeted perturbation of candidate nodes (e.g. CAT/GPX inhibition or GSH supplementation), (ii) real-time compartment-specific redox imaging (e.g. roGFP-based reporters), and (iii) genetic manipulation of discriminative TRX/GRX/PRX and GST candidates.