The seeds used in this study were obtained from fruits of the open-pollinated blackberry cultivar ‘Loch Ness’ [21], collected from plants located at the experimental field of La Rábida Campus, University of Huelva (southwestern Spain), during summer 2023. The ‘Loch Ness’ cultivar was selected due to its commercial importance and its relevance for studies on blackberry germination.

Fruits were mashed and blended for 1 min using a blender with plastic-covered blades. After resting for approximately 2 min to allow seed deposition, the pulp was separated by inverting the container. The juice was discarded and the remaining solids were rinsed through a fine-mesh sieve. Seeds were subsequently cleaned by immersion in a 0.5% commercial sodium hypochlorite solution for 5 min and then dried in the shade for approximately 24 h. Seed viability was assessed using the tetrazolium test according to Peters [22]. No scarification treatment was applied before static magnetic field exposure or cold stratification. This decision was made to isolate the effects of these treatments on germination and associated biochemical responses, although it may have contributed to the low absolute germination percentages observed.

A laboratory experiment was conducted at the Plant and Crop Science Laboratory of La Rábida Campus (37°12′32.3980″ N, 6°55′09.9948″ W), University of Huelva, to evaluate the germination response of ‘Loch Ness’ blackberry seeds. Seeds were exposed or not exposed to an 80 mT static magnetic field (SMF) for 24 h, based on exposure durations previously reported as effective in seed stimulation studies [23], and subsequently subjected to cold stratification (CS) for 0, 15, 30, 45, 60, or 75 days.

Blackberry seeds were exposed to the SMF emitted by a homemade electromagnetic field generator previously described by Dueñas et al. [24], which allows a controllable magnetic intensity between 7 and 200 mT. Seeds were placed in a plastic container between the N–S poles of the electromagnet and subjected to the magnetic treatment for the required duration. The magnetic field intensity was verified using a digital teslameter.

After SMF treatment, seeds were transferred to labelled Petri dishes and subjected to cold stratification at 5–7°C for the corresponding CS periods.

Following each stratification period, seeds were incubated for germination at 20 ± 4°C under natural laboratory light conditions. Relative humidity during the experimental period was approximately 36 ± 6%. Photoperiod and light intensity were not instrumentally controlled and should therefore be considered a limitation of the study.

For biochemical analyses, seeds were placed on moistened filter paper in sterile 90 mm Petri dishes, sealed with parafilm to minimize water loss, and incubated for 15 days under the conditions described above to stimulate metabolic activation. Immediately after pre-incubation, seeds were stored at −20°C until biochemical analyses.

For germination assessment, the remaining seeds were incubated on moistened filter paper in sterile Petri dishes for 10 weeks under the same temperature and light conditions described above. Petri dishes were kept closed during incubation to minimize moisture loss, but no periodic rehydration was applied. Germination was evaluated twice per week, and seeds were considered germinated when radicle emergence reached at least 1 mm.

Due to the persistent dormancy of Rubus seeds, morphological parameters such as shoot length, root length, or chlorophyll content could not be consistently measured. Therefore, cumulative germination and internal biochemical markers were used as the main indicators of dormancy release.

Seeds of the ‘Loch Ness’ cultivar were subjected to 12 treatment combinations arranged in a 2 × 6 factorial design, consisting of two SMF levels (treated and untreated) and six cold stratification periods (0, 15, 30, 45, 60, or 75 days). The experiment followed a completely randomized design with six replications per treatment. Each replicate included 30 seeds. The entire experiment was repeated twice.

Starch content was determined following the method of Novelo-Cen and Betancur-Ancona [25], with minor modifications [12]. Seeds (five per replication) were dried at 45°C for 24 h to obtain dry weight. Samples were homogenized in 1 mL of distilled water using a polytron homogenizer (PT 2500 E, Kinematica AG, Switzerland). The homogenate was filtered through a 100-mesh filter and centrifuged at 7000× g for 15 min at 4°C. The supernatant was removed and the sediment was resuspended in 1 mL of distilled water and centrifuged again under the same conditions. This step was repeated until the supernatant became transparent. The final sediment was dried at 45°C for 24 h and weighed using a precision balance to estimate starch content (mg g⁻¹ DW, where DW indicates dry weight).

Total soluble sugars were quantified using the phenol–sulfuric acid method described by DuBois et al. [26], with slight modifications [12]. Seeds were dried at 45°C for 24 h and homogenized in 1 mL of 80% ethanol. The suspension was vortexed, transferred to an Eppendorf tube and heated in a water bath at 75°C for 10 min. The extract was centrifuged at 5000× g for 10 min at 4°C. A 250 μL aliquot of the supernatant was mixed with 750 μL of concentrated sulfuric acid and 5 μL of 5% aqueous phenol. After standing for 30 min, absorbance was measured at 490 nm using a spectrophotometer (PG1800, Lan Optics; Labolan, Spain). Quantification was based on the following calibration curve: Y = 0.0596X − 0.0676 where Y represents absorbance and X the glucose concentration (mg mL⁻¹).

Five seeds per replicate were used for biochemical analyses because of the small seed size, the destructive nature of the assays, and the limited amount of material available for each treatment combination. This sampling approach was adopted for an exploratory evaluation of early metabolic changes, although it should be regarded as a limitation due to the expected biological variability among seeds.

Amylase activity was determined using the classical colorimetric method described by Bernfeld [27], adapted to independently estimate α-amylase and β-amylase activities. Seeds were dried at 45°C for 24 h, homogenized using a polytron homogenizer and centrifuged at 14,000× g for 30 min at 4°C to obtain a crude enzyme extract.

For α-amylase activity, 750 μL of 0.1% (w/v) soluble starch and 750 μL of enzyme extract were incubated at 30°C for 60 min. The reaction was stopped by adding 0.5 mL of color reagent prepared with 1 g of 3,5-dinitrosalicylic acid, 20 mL of 2 mol L⁻¹ NaOH, and 30 g of potassium sodium tartrate in 100 mL of distilled water. The mixture was heated at 50°C for 5 min and absorbance was measured at 546 nm.

Calibration curves were generated using increasing concentrations of D−(+) maltose monohydrate as standard. The calibration equation for α-amylase activity was: Y = 0.0616X + 0.0129 (R² = 0.87) where Y is absorbance and X is maltose concentration (mg mL⁻¹).

For β-amylase activity, the same procedure was followed, except that 0.1% (w/v) soluble amylopectin was used instead of soluble starch as substrate. After incubation and color development, absorbance was measured at 540 nm. The corresponding calibration curve was: Y = 0.0677X + 0.0128 (R² = 0.87) where Y is absorbance and X is maltose concentration (mg mL⁻¹).

Amylase activity was finally expressed as μmol of hydrolysed maltose equivalents g⁻¹ DW.

Cumulative germination, starch content, total soluble sugars, and α- and β-amylase activity were analysed to test the effects of SMF and CS. Germination percentages were arcsine-transformed prior to analysis to ensure normality and homoscedasticity. Experiment and replication nested within experiment were treated as random effects, whereas SMF, CS, and their interaction were treated as fixed effects. When significant effects were detected in the mixed-effects models, mean comparisons among treatments were performed using Tukey’s HSD test (p ≤ 0.05).

Pearson’s correlation coefficients were calculated to evaluate relationships among biochemical variables and germination. Linear regression models were constructed using stepwise selection procedures.

To explore potential nonlinear relationships within the present dataset, 40 machine learning regression algorithms were screened as exploratory approaches (Table S1). The dataset was randomly divided into training (70%) and test (30%) subsets. During the initial screening of algorithms, model performance was evaluated primarily using the coefficient of determination (R²) and the root mean square error (RMSE). For the final comparison of the selected models presented in the main text (Table 1), performance was summarized using R², RMSE, and mean absolute error (MAE) [28,29].

Mixed-effects models were fitted in R using the lme4 package, and inference for fixed effects was obtained through Type III tests with Satterthwaite’s approximation of denominator degrees of freedom. Post hoc comparisons among treatment means were performed using estimated marginal means with Tukey adjustment.

Mixed-effects models, correlation analyses, and linear regressions were conducted in R using the packages lme4, PerformanceAnalytics, and MASS [30,31,32]. Machine learning analyses were performed in Python using the LazyPredict library.

Fresh seed viability exceeded 95% according to the tetrazolium test. Despite this high viability, seeds exhibited a low mean germination percentage of 4.4 ± 1.8% after 10 weeks under laboratory conditions (20 ± 4°C, natural laboratory light) (Fig. 1 and Fig. 2a).

Cold stratification (CS) significantly affected cumulative germination (Table 2). Mean germination was 6.0 ± 1.4% without CS, decreased to 0.0% after 15 days of stratification, and slightly increased to 1.5 ± 0.3% after 30 days. Germination increased again with longer stratification periods, reaching 6.0 ± 1.4%, 6.8 ± 1.3%, and 6.0 ± 1.5% after 45, 60, and 75 days of CS, respectively (Fig. 2a). The transient reduction in germination observed after 15 days of stratification may reflect incomplete and heterogeneous dormancy release during the early chilling phase. Under these conditions, short-term chilling may have initiated partial physiological changes without being sufficient to support consistent radicle emergence.

Type III tests also revealed significant effects of SMF and the SMF × CS interaction (Table 2). Except for the 15- and 30-day CS treatments, SMF-treated seeds consistently showed higher germination values than untreated seeds, with differences approaching twofold in several stratification treatments (p ≤ 0.05), irrespective of stratification duration (Fig. 2a).

Starch content was significantly influenced by CS and the CS × SMF interaction (Table 2).

Seeds subjected to stratification periods associated with higher germination showed a marked reduction in starch reserves. In SMF-treated seeds, the CS treatments with the highest germination values (0-, 45-, 60-, and 75-day CS) showed approximately twofold lower starch content compared with treatments with the lowest germination (15- and 30-day CS) (Fig. 2b).

In contrast, when seeds were not exposed to SMF and germination remained minimal, starch levels did not differ significantly among CS treatments.

Cold stratification significantly affected total soluble sugar levels independently of SMF treatment (Table 2).

Soluble sugars increased markedly from 0 to 15 days of CS (mean increase of 49.8%) and from 15 to 30 days (32.4%). After 30 days of stratification, sugar levels remained relatively stable until 75 days of CS, with no significant differences detected among later stratification periods (Fig. 3a).

Enzymatic activity of both α- and β-amylase was significantly influenced by SMF treatment (Table 2).

On average, SMF-treated seeds exhibited significantly higher α-amylase and β-amylase activities than untreated seeds, with increases of 15.2% and 42.6%, respectively (Fig. 3b,c).

These results suggest that SMF exposure was associated with higher enzymatic activity related to starch hydrolysis during germination.

Correlation analysis revealed moderate relationships between cumulative germination and several biochemical variables (Fig. 4).

Germination showed a significant negative correlation with starch content and significant positive correlations with α- and β-amylase activities (|r| = 0.18–0.31; p ≤ 0.05), whereas the relationship with total soluble sugars was weak and not significant (r = 0.065; p > 0.05).

Linear regression analysis indicated that α-amylase activity alone explained a small proportion of germination variability (R² = 0.09). A multivariate regression model incorporating starch content and β-amylase activity slightly improved predictive performance (R² = 0.12) (Table 1).

Among the exploratory nonlinear approaches tested (Table S1), regression tree analysis provided the best, albeit still limited, predictive performance within the present dataset. The model achieved R² = 0.40 for the training dataset, whereas the test dataset yielded R² = 0.32 (Table 1). Under these conditions, starch content emerged as the first splitting variable in the model, with a threshold value of 46 mg g⁻¹ DW (Fig. 5).

Model validation using the test dataset yielded R² = 0.32, with RMSE = 4.77 and MAE = 3.51, indicating only moderate predictive ability under the present experimental conditions. Therefore, these results should be interpreted cautiously and viewed as exploratory evidence of possible associations rather than as a robust predictive framework.

Seed germination remained low after 10 weeks, which is consistent with previous reports describing slow and irregular germination in Rubus species [4,5,33]. This behaviour reflects the presence of deep dormancy, typically involving both physical and physiological components. In particular, dormancy in Rubus seeds has been associated with seed coat impermeability and internal physiological constraints linked to embryo growth [3,6,7].

The absence of germination after 15 days of cold stratification suggests that early chilling exposure was insufficient to produce a stable transition from dormancy to germination. Rather than indicating a simple linear response to stratification, this result may reflect heterogeneous dormancy status within the seed lot, partial metabolic activation, or an intermediate physiological state in which some dormancy-related processes had already started but radicle protrusion was still prevented. This interpretation is consistent with the strong and complex dormancy previously reported in Rubus seeds.

The persistence of low germination even after stratification supports the idea that physiological dormancy remains only partially released under the tested conditions. In this context, the limited mobilization of seed reserves observed in low-germinating treatments may represent a key constraint, as insufficient metabolic activation can delay or prevent radicle emergence. These findings reinforce the complexity of dormancy regulation in Rubus and highlight the need to consider both structural and biochemical limitations simultaneously.

Although germination increased in some treatment combinations, the maximum values observed in this study remained low. This is consistent with the strong dormancy commonly reported in Rubus seeds, but it also suggests that the conditions tested were not sufficient to fully overcome dormancy in the ‘Loch Ness’ material. Several factors may have contributed to this outcome, including the absence of scarification, the possibility that the stratification periods were still insufficient, and the fact that germination conditions were not fully optimized. Therefore, the present study should be interpreted as an exploratory physiological evaluation of germination-associated processes rather than as an optimized propagation protocol.

Cold stratification significantly influenced germination, confirming its well-established role as a dormancy-breaking factor in temperate species [6]. However, the effect of CS alone was limited, suggesting that stratification duration was not sufficient to fully overcome dormancy in all cases.

In contrast, SMF-treated seeds generally showed higher germination than untreated seeds in several stratification treatments. However, this pattern was not observed uniformly across all CS durations, and the absolute germination values remained low. Similar effects of magnetic fields on seed germination have been reported in several crops, including maize, sunflower, and soybean [15,16,17]. These effects have been associated with increased membrane permeability, improved ion transport, and enhanced enzymatic activity, thereby contributing to faster metabolic activation during germination [10,20].

Importantly, SMF alone did not replace the role of stratification, but was associated with higher germination in several stratification treatments, suggesting a complementary interaction between both factors. This pattern is consistent with previous findings in tree species such as Tilia miqueliana, where magnetic treatments accelerated dormancy release when combined with stratification [19].

The interaction between SMF and cold stratification may reflect the dependence of magnetic stimulation on the physiological state of the seed, as suggested by recent work indicating that magnetic responses depend on treatment intensity, exposure duration, and seed physiological status [10,19]. Under strong dormancy, SMF alone may not be sufficient to promote consistent germination, whereas after partial dormancy alleviation through chilling, its effect may become more evident through enhanced metabolic responsiveness. Similar context-dependent effects of magnetic treatments on seed physiology have been reported in other species, where responses varied according to treatment intensity, exposure duration, and the pre-existing dormancy or activation status of the seeds. A comparable response was reported in Calotropis procera, where static magnetic field exposure promoted faster germination and increased germination rate, together with biochemical changes consistent with stimulated cellular metabolism [18].

Overall, these results suggest that SMF may have contributed to metabolic activation associated with germination once dormancy had been partially alleviated, rather than acting as a direct dormancy-breaking factor.

Treatments showing higher germination generally exhibited lower residual starch content, suggesting that reserve mobilization is associated with germination progression.

Starch is the main energy reserve in seeds and is hydrolyzed into soluble sugars that support respiration and embryo growth during germination [8,9]. Therefore, the reduction in starch content observed in this study likely reflects the metabolic transition from dormancy to active growth.

Similar patterns have been reported in other species, where the initiation of germination is closely linked to the degradation of stored reserves [11]. In Rubus, where dormancy is particularly strong, limited starch mobilization may be one of the factors contributing to the low germination percentages observed in untreated or poorly responsive treatments.

Although total soluble sugars increased during the early stratification periods, their weak relationship with final germination suggests that sugar accumulation alone is not a reliable indicator of successful dormancy release. Under the present conditions, this increase may reflect partial reserve mobilization, transient accumulation of soluble carbohydrates, or incomplete metabolic activation during the early stages of chilling. It is also possible that sugars accumulated before being effectively used to support embryo growth, particularly in seeds that remained dormant or only partially activated.

The increase in α- and β-amylase activity observed in SMF-treated seeds indicates that magnetic exposure enhanced enzymatic starch hydrolysis. Amylases play a central role in germination by converting starch into soluble sugars that can be used for metabolic processes [9].

Previous studies have shown that β-amylase activity is particularly important in facilitating reserve mobilization and is often associated with successful germination [11]. In this study, the higher β-amylase activity in SMF-treated seeds is compatible with the hypothesis that enhanced enzymatic activity may contribute to the modest germination differences observed among treatments.

Furthermore, the association between low enzymatic activity and poor germination reinforces the role of metabolic activation as a limiting factor in Rubus seeds. In this context, the higher amylase activity observed in SMF-treated seeds is compatible with a possible association between SMF exposure and enhanced carbohydrate metabolism during germination. However, because the present study did not directly evaluate the underlying regulatory pathways, a direct mechanistic effect of SMF on enzymatic activation cannot be confirmed. Therefore, these findings should be interpreted cautiously as exploratory physiological evidence, in line with previous studies on magneto-priming and seed physiological quality [10,20].

Reserve mobilization during germination is closely linked to hormonal regulation, particularly the antagonistic balance between abscisic acid (ABA) and gibberellins (GA). In general, ABA promotes dormancy maintenance and restricts embryo growth, whereas GA promotes hydrolytic enzyme synthesis and the mobilization of storage reserves. In this context, the higher amylase activity observed in SMF-treated seeds may be physiologically compatible with a shift toward a more germination-prone metabolic state. However, because ABA, GA, or related signalling pathways were not measured in this study, this interpretation should be regarded as hypothetical and requires direct confirmation in future studies.

The statistical analyses suggested that starch-related traits were among the biochemical variables most closely associated with germination performance in the present dataset. Both linear and nonlinear approaches pointed in this direction, although overall predictive accuracy remained limited.

Linear regression models showed low explanatory power, which is not surprising given the complexity of germination responses and the expected nonlinearity among biochemical variables. The regression tree approach slightly improved model performance, probably because it was able to capture threshold-like relationships and interactions among variables. Similar advantages of nonlinear models have been reported in seed science and crop prediction studies [34,35]. However, this improvement remained modest.

These analyses should not be interpreted as robust predictive models, but rather as exploratory tools to identify candidate biochemical variables potentially associated with germination under the present conditions. The relatively low R² values indicate that germination in blackberry seeds is influenced by additional factors not captured in this study, including seed-to-seed variability and potentially relevant physiological processes such as hormonal regulation.

The results of this study suggest that SMF deserves further evaluation as a complementary pre-sowing treatment in species with strong dormancy, such as Rubus. However, its effectiveness appears to depend on the physiological state of the seed and the degree of dormancy release achieved through stratification.

Future research should focus on evaluating SMF effects across different cultivars and species, integrating hormonal and molecular analyses to better understand underlying mechanisms, and testing the scalability of magnetic treatments under practical conditions. Additionally, combining biochemical markers with advanced modelling approaches may improve the prediction of germination behaviour in complex systems.