Callus induction: The leaf and root explants were cultured on ½ MS medium supplemented with different concentrations of Kin either alone or in combination with 2,4-D and or NAA. After seven days, callus induction was observed in the margin of root and leaf explants treated with or without different concentrations of PGRs combinations. Callus induction was observed at a high frequency (100%) in all treatments. In addition, the morphology of the leaf-derived callus was friable and light green, but root-derived callus was friable and cream (Fig. 2).
Total Phenol Content (TPC): Data from statistical analyses showed that the interaction between Kin×explant, 2,4-D×explant, Kin×2,4-D, and Kin×2,4-D×explant had a significant effect (p ≤ 0.01) on the TPC of leaf and root-derived callus (Table 1A). The interaction slicing method was applied to analyze the effects of different concentrations of Kin in combination with 2,4-D and or NAA on the TPC. The obtained data revealed that 0.5, 1, and 2 mg/L Kin in combination with different concentrations of 2,4-D had a significant effect on TPC in both leaf and root-derived callus (Table 1B).
The results demonstrated that the effect of explant, Kin, and NAA× explant was not significant on TPC. However, the effects of other treatments were significant (p≤0.01) on TPC (Table 2A). Interestingly, root-derived callus showed a significant increase (p≤ 0.01) in TPC when grown in the medium supplemented with 0.5, 1, and 2 mg/L Kin in combination with different concentrations of NAA (Table 2B).
The comparison of means revealed that the highest TPC (32.23 mg/g DW) was observed in leaf-derived callus grown on ½ MS medium supplemented with the combination of 0.5, 1 mg/L Kin, and 2,4-D, respectively. Meanwhile, the highest TPC (35.5 mg/g DW) was observed in the root-derived callus, grown on ½ MS medium containing 2, 2 mg/L Kin, and NAA (Table 3).
Up to now, several reports have been published to show the effect of PGRs on the phenolic acids production in different medicinal plants in tissue culture conditions. However, there is no report for the callus culture of L. undulata. Our previous research showed that TPC in different parts of L. undulata in the reproductive phase is as follows: stem>root>leave. But during the vegetative phase, there is no significant difference in the TPC between different organs of this species (Ramezannezhad et al. 2019b). Secondary metabolites biosynthesis depends on various factors such as plant species and developmental stages. Because of the role of Secondary metabolites in the plant life cycle and growth stages, the concentration and accumulation of these compounds vary in different plant organs (Cheynier et al. 2013; Jiang et al. 2013). It was reported that TPC reached the maximum level in many herbaceous plants during the flowering time (Ammar et al. 2012; Fernando et al. 2013; Feduraev et al. 2019). On one hand, similar results have been reported by Lannucci et al. (2013), who reported a correlation between concentrations of phenolic compounds and plant growth. On the other hand, the obtained results indicated that the type of PGRs has an effective role on TPC in callus derived from different explants. Aremu et al. (2015) demonstrated that cytokinin can affect the phenylpropanoids pathway, especially at molecular levels. Another report confirmed that NAA induces phenolic compounds metabolism pathways via binding with TIR1 (Auxin receptor) (Buer et al. 2006).
Phenolic acids contents: Cichoric acid, chlorogenic acid, and caffeic acid contents were measured in leaf and root-derived callus, which contained the highest TPC. These metabolites were identified in both root and leaf explant and their corresponding callus. The obtained results revealed that cichoric acid and chlorogenic acid contents in root explant were more than leaf explant. However, the leaf explant contained a higher amount of caffeic acid than the root explant (Fig. 2A-D). The amount of polyphenols (such as cichoric acid) depends on various parameters, including harvested organ, harvest time, plant species, plant age, climatic factors, growth conditions, and others (Lee and Scagel 2013). For example, Qu et al. (2005) reported that the amount of cichoric acid in roots and shoots of Echinacea purpurea ranged from 2.03 to 38.55 mg/g DW. On the contrary, Zolgharnein et al. (2010) reported that the cichoric acid content is about 1.50 mg/g DW in the same plant (Zolgharnein et al. 2010).
Leaf-derived callus showed the highest amount of cichoric acid (3.95 mg/g DW) when grown on ½ MS medium supplemented with 0.5, 1 mg/L Kin, and 2,4-D, respectively (Figure 2A). In contrast, cichoric acid amount increased to 4.59 mg/g DW in root-derived callus when grown on medium contained 2, 2 mg/L Kin, and 2,4-D (Figure 2B).
On one hand, the obtained results showed that the amount of chlorogenic acid in root explant (1.03 mg/g DW) was almost two times more than in leaf explant (0.58 mg/g DW) (Figure 2C-D). On the other hand, the highest amount of chlorogenic acid (2.50 mg/g DW) was obtained by the treatment with 1, 0.1 mg/ l Kin, and 2,4-D, respectively (Figure 2 C). However, chlorogenic acid accumulation reached to 3.17 mg/g DW in root-derived callus when grown on the medium containing 2, 2 mg/ l Kin, and 2,4-D (Figure 2 D).
In addition, 0.5, 1 mg/l Kin, and 2,4-D were the most effective PGRs concentration for caffeic acid production (24.23 mg/g DW) in leaf-derived callus (Figure 2 E). Interestingly, the highest amount of caffeic acid (13.23 mg/g DW) was recorded in root-derived callus when grown on the medium containing: 2, 2 mg/l Kin, and NAA (Figure 2F).
In spite of no published reports concerning L. undulata tissue culture, some reports demonstrated that the amount of phenolic acids in explant is higher than corresponding callus in different plant species (Butiuc-Keul et al. 2012). It seems that plant species, the age of explant or callus, and concentrations of applied PGRs play a vital role in phenolic acids production under tissue culture conditions. According to the current data, the effect of PGRs on the phenolic acids content in leaf and root-derived callus was different, and it can be due to the type and concentration of PGRs. Caffeic acid is produced via Phenylpropanoids biosynthesis pathways, and PGRs regulate the biosynthesis of phenolic acids (Buer et al. 2006). The authors also found that NAA is linked to TIR1 (Auxin receptor) and induces phenolic acids metabolism pathway. Our data revealed that the root explant of L.undulata contained higher cichoric acid and chlorogenic acid than leaf explant. Whereas it contained lower caffeic acid, compared to the leaf explant. It can be because converting caffeic acid to its derivative (cichoric acid and chlorogenic acid) in root explant is faster than leaf explant. Besides, caffeic acid conversion to cichoric acid and chlorogenic acid is done by two separate pathways (Buer et al. 2006). It is possible that in both leaf and root explant, caffeic acid is converted to cichoric acid rather than chlorogenic acid. Overall, the current results indicated that both leaf and root-derived callus accumulated a higher amount of cichoric acid, chlorogenic acid, and caffeic acid, compared to the corresponding explants.
ANN modeling
In this study, the ANN model has been applied to predict the effects of different concentrations of PGRs (Kin, 2,4-D and NAA) on the production of phenolic acids in leaf and root-derived callus from L.undulata. Based on the data obtained from the interaction slicing method, several concentrations of Kin, 2,4-D, and NAA were selected, which had the most stimulatory effect on the TPC in root and leaf-derived callus (Table 4).
Figure 3 reveals the value of experimental outputs (cichoric acid, chlorogenic acid, and caffeic acid production) versus its predicted values obtained from leaf and root-derived callus in the most effective models of ANN.
Some of the performance factors of the used models were summarized in table 5. The results showed that the tangent- sigmoid activation function-based neural network with 30 hidden layer neurons had the most accurate prediction for the cichoric acid content in leaf-derived callus. Meanwhile, the correlation factor (R2) between the experimental outputs and the predicted output by ANN was equal to 0.8733 (Figure. 3A). By contrast, tangent-tangent activation function-based neural network with 20 hidden layer neurons indicated the best prediction for the content of cichoric acid in root-derived callus with R2 value equal to 0.8035 (Figure 3B). The sum of the square error (SSE) and the relative error (RE) were very low in both structures. Therefore, the current models can be reliable for the dataset (Table 3).
The structure 3-15-1 showed the best prediction for chlorogenic acid content in leaf-derived callus when tangent-tangent activation function was used in the hidden and output layer, respectively (Table 5). When, the experimental output variables were plotted against ANN-predicted output variables (Figure 3C), the value of R2 between those values were 0.6838. In contrast, the structure of 3-25-1 with tangent-tangent activation functions was determined as the best model for predicting the amount of chlorogenic acid in root-derived callus (Table 5). The correlation between the experimental value of the variables and the ANN-predicted values of the output variables is shown in Figure 4D. According to this plotted curve, R2 is 0.975 and it demonstrates a strong correlation between the MLP-model outputs and experimental data. SSE and RE value of model was very low (Table 5). Therefore, the proposed model can predict the amount of chlorogenic acid in root-derived callus.
The optimized structure to predict caffeic acid values in leaf-derived callus is 3-25-1 with a tangent-tangent activation function (Table 5). The obtained R2 value was 0.72 (Fig. 3E). The best structure for the production of this metabolite in root-derived callus was sigmoid-sigmoid activation function-based neural network with 20 hidden layer neurons. In this case, the obtained R2 value was 0.93 (Figure 3F). Value of SSE and RE were also very low (Table 5) and R2 datasets revealed a strong correlation between experimental and predicted data (Figure. 3F).
To date, there were no published reports for ANN modeling to predict phenolic acids production in L.undulata. However, few studies have predicted the production of phenolic acids using ANN in several plant species. Yu et al. (2019) used ANN to predict the optimized chlorogenic acid value in Lonicera japonica. Their results revealed that ANN modeling was successful in predicting the chlorogenic acid value. The constructed ANN indicated high R2 values (0.9898). Also, ANN modeling was used for total phenol production in two species of bananas (cv. Musa nana and Musa cavendishii) (Guiné and Costa 2016). In this case, the sigmoid function was selected as the activation function. The results indicated that ANN could predict the experimental results. The highest error percentage between experimental and predicted data was less than 2.7, and R2 was equal to 0.95. Xi et al. (2013) used ANN to optimize the extraction of polyphenols from tea. They used a feed-forward neural network with three input neurons, one hidden layer with eight neurons, and one output layer with a single neuron. The trained network showed the lowest MRE value (0.03) and maximum R2 value (0.9571), which indicates a good agreement between the predicted value and the experimental value.