Background

Protoplast is a general term for a variety of substances in plant cells other than the cell wall removed, which has the totipotency as well as a part of the viability of ordinary plant cells [1], and have been extensively utilized in various experiments, such as gene promoter screening, verification of exogenous gene function, subcellular localization of proteins, protein interactions and multiplex genome editing [2,3,4,1: Table S2, and the reaction systems were provided in Additional file materials. The relative expression of each housekee** gene was calculated using the formula Q = E−ΔΔCt, where E represents the gene amplification efficiency, typically assumed to be two (100.00% efficiency). ΔCt is calculated as Ct (min)—Ct (sample), where Ct (min) is the lowest Ct value among all samples, and Ct (sample) is the Ct value of each sample [54, 55]. The data were subjected to two-sample anova and plotted using Origin.

Exogenous gene transformation

The pBI121-SaNADP-ME2 and pBI121-GFP null-loaded E. coli strains were obtained from our laboratory and contained kanamycin resistance genes and GFP reporter gene. Refer to Ren et al. [56] for the procedure of protoplast transformation and improve it. Protoplasts were used as receptors and were first resuspended in 1 mL of MMG solution, then incubated on ice for 30 min and centrifuged at 300 rpm min−1 for 2 min to remove the supernatant. 600 μL of protoplast solution was pipetted into a 5 mL centrifuge tube. 60 μL of pBI121-SaNADP-ME2-GFP plasmid vector was added to the bottom of the tube and the tube was gently flicked to mix the contents. The tube was then inverted several times to mix the 660 μL of PEG solution. The mixture was incubated in a dark environment at room temperature for 30 min, completing the transformation process. The transformation reaction was stopped by adding a twofold volume of W5 solution. After centrifugation at 300 rpm min−1 for 2 min, the supernatant was discarded. The protoplasts were then resuspended in 3 mL of WI solution and placed in cell culture chambers (Labselect, Bei**g, China). They were incubated in the dark at room temperature for 16–24 h. A suitable amount of the transformed protoplasts was used for further analysis. Protoplasts transformed with pBI121-GFP empty load were used as a control. The images were captured using a laser confocal scanning microscope (Zeiss LSM 800, Jena, Germany) and processed with Zen 2012 software. GFP and chlorophyll were excited using 488 nm and 633 nm laser lines, respectively, to observe GFP expression.

Results

Analysis of protoplasmic preparation results

Seventeen groups of experiments were conducted using the Box-Behnken design, and the results obtained were presented in Table 1. Among the 17 groups, group 12 exhibited the highest yield with a value of 1.566 × 106/100 mg. Group 5 showed the highest viability at 90.81%, while group 12 had the maximum number of viable cells, reaching 1.367 × 106/100 mg.

Table 1 Experimental results of three-factor, two-response-value Box-Behnken design test based on the preparation of protoplasts of Salsola laricifolia
Full size table

RSM results analysis

RSM regression model analysis

The test results were analyzed using RSM to establish the response values Y1 and Y2 based on the experimental factors A, B, and C. Quadratic multinomial regression equations was obtained. The model evaluation indexes were in Table 2, and the detailed results of the model were provided in Additional file 1: Tables S3, S4.

$${\text{Y}}_{{1}} = 6.119 + 1.815{\text{A}} - {4}{\text{.626B}} - {18}{\text{.608C + 0}}{\text{.436AB + 0}}{\text{.580AC}} - {0}{\text{.887BC}} - {0}{\text{.633A}}^{2} + 3.269{\text{B}}^{2} + 15.650{\text{C}}^{2}$$
$${\text{Y}}_{{2}} = 18.883 + 13.360{\text{A + 49}}{\text{.779B}} - {186}{\text{.192C}} - {9}{\text{.063AB + 7}}{\text{.010AC}} - {24}{\text{.790BC}} - {3}{\text{.289A}}^{2} - 13.194{\text{B}}^{2} + 150.865{\text{C}}^{2}$$
Table 2 Response surface model evaluation indexes
Full size table

The yield model (Y1) had an F-value of 105.270, indicating highly significant level differences (p < 0.0001). The R2 value was 0.993, and the signal-to-noise ratio was 35.072, which was greater than 10, indicating that the model was sufficiently accurate and not easily affected by external disturbances. Additionally, the data had an acceptable level of variability, as evidenced by the range of 4% to 10% for the coefficient of variation (C.V. %). The viability model (Y2) had an F-value of 21.080 (p = 0.0003), which was less than 0.010, indicating a significant model variance. The R2 value was 0.964, and the signal-to-noise ratio was 18.012, indicating a high level of accuracy and resistance to external disturbances. Additionally, the C.V.% value was less than 4.00, suggesting a low degree of data variability in the model.

RSM model three-dimensional result graph analysis

From Fig. 1a, b and c, it can be observed that the interaction effects of cellulase R-10 content (A) and macerozyme R-10 content (B) were the most significant. The influence of A on protoplast yield (Y1) was greater than that of B, which was manifested by the fact that the surface intersecting AB was the steepest, and the contours intersecting with the A-axis were more densely packed than those of the B-axis (Fig. 1a), the interaction between A and mannitol concentration (C) had more significant effect on Y1. A had a greater effect on Y1 than C (Fig. 1b), where the surface was smoother and the contours intersecting the A-axis were denser than those of the C-axis. The interaction between B and mannitol concentration (C) had a significant effect, and B had a greater effect on Y1 than C (Fig. 1c), where the surface was smoothest and the contours intersecting the B-axis were denser than those of the C-axis. Therefore, it can be concluded that the effects of the experimental factors A, B, and C on Y1 following the order of A > B > C. Additionally, the effect of the interaction between these factors on Y1 followed the order of AB > AC > BC.

Fig. 1
figure 1

Three-dimensional response surface plots based on the RSM model resolving the interaction of various factors of the experiment on the results of protoplast preparation of Salsola laricifolia. The plots labeled a, b, and c represent the protoplast yield (Y1) and show how it is influenced by the interaction between two experimental factors, Similarly, the plots labeled d, e, and f represent protoplast viability (Y2) and how it is influenced by the interaction between two experimental factors. X-axis, Y-axis are the experimental factors including cellulase R-10 content, macerozyme R-10 content, and mannitol concentration, denoted by A, B, and C, respectively

Full size image

Similarly, as shown in Figs. 1d, 1e, and f, the experimental factors on the protoplast preparation viability (Y2) resulted in the order of the influence of B > A > C, the interaction between the factors on the degree of influence of Y2 in the order of AB > BC > AC.

ANN result analysis

The mean square values, error values, and simulation effects of the ANN model were shown in Fig. 2. The yield model was optimized using the Levenberg–Marquardt algorithm with 18 hidden layers. The vitality model was optimized using the scaled conjugate gradient method with 17 hidden layers. The viable cell model was optimized using the Levenberg–Marquardt algorithm with 14 hidden layers.

Fig. 2
figure 2

Artificial neural network (ANN) model simulation effect. It displays the training, verification, test and all data fitting for each model. The fitting effect is represented by correlation coefficient (R). a represents the yield model, b represents the viability model, and c represents the viable cell model

Full size image

The correlation coefficient of the yield model obtained after training the ANN model was 0.96736. The end-of-training condition was met after 5 iterations, with a mean square error (MSE) of 0.045 (Figs. 2, 3a). The correlation coefficient of the ANN-viability model was 0.977, and the MSE of the ANN-viability model was 8.5388 e−05 after 10 iterations, satisfying the end-of-training condition (Figs. 2, 3b). The correlation coefficient of the ANN-vital cell model was 0.995, and its mean square error was 0.003 after four iterations, also satisfying the end-of-training condition (Figs. 2, 3c).

Fig. 3
figure 3

Artificial neural network (ANN) model mean square error (MSE) effect plot. The top shows the number of ANN model epochs when the minimum MSE is obtained for validation data. a represents the yield model, b represents the vigor model, and c represents the viable cell model

Full size image

Comparison between RSM and ANN model fitting and prediction performance

Fitting effect of RSM, ANN

According to Fig. 4, the RSM yield and viability model had a better fit compared to the ANN yield and viability model. The R and R2 values of the RSM model were 0.996, 0.992, and 0.982, 0.964 respectively, which were closer to 1 than the values of the ANN model. Additionally, the RSM model had smaller RMSE and MAPE values compared to the ANN model. However, the ANN model showed improved accuracy in predicting the number of live cells, with its R and R2 values being second only to the best-fitting RSM yield model (Fig. 4, Table 3).

Fig. 4
figure 4

Comparison of experimental values with predicted values of response surface methodology (RSM), artificial neural network (ANN) models. a for protoplast yield, b for protoplast viability

Full size image
Table 3 Evaluation of response surface methodology (RSM), artificial neural network (ANN) model based on statistical indexes
Full size table

When analyzing only yield and viability, protoplast yield fluctuated greatly by changes in experimental conditions, while viability fluctuated less. The coefficients of variation (C.V.%) of the actual and predicted values of protoplast yield were 42.203, 41.943 and 48.438, which were much larger than those of the actual and predicted values of viability, which were 3.428, 3.366, and 3.354 (Fig. 4). Consequently, the RMSE and MAPE of the predicted values for yield were higher than those for viability predictions in Table 3, indicating a greater bias in yield prediction.

RSM, ANN model optimization and verification

The RSM model was used to solve for the optimal values under the conditions of simultaneous consideration of the yield and viability. The results revealed that under the conditions of a cellulase R-10 content of 1.86%, macerozyme R-10 content of 1.00% and mannitol concentration of 0.50 mol L−1, the theoretical values for protoplast yield were 1.464 × 106/100 mg and viability was 90.81% (Table 4). GA was utilized for optimization of ANN-yield model, ANN-viability model and ANN-viable cell number model respectively. ANN-yield model predicted a theoretical maximum protoplast yield of 1.625 × 106/100 mg. This prediction was based on the following experimental conditions: 1.98% cellulase R-10 content, 1.00% macerozyme R-10 content, and 0.50 mol L−1 mannitol concentration. Similarly, the viability model predicted a theoretical maximum viability of 96.29 based on experimental conditions including 1.65% cellulase R-10 content, 0.98% macerozyme R-10 content, and 0.53 mol L−1 mannitol concentration. Additionally, the ANN viable cell number model predicted a theoretical maximum viable cell number of 1.342 × 106/100 mg, which was based on 2.00% cellulase R-10 content, 1.00% macerozyme R-10 content, and 0.70 mol L−1 mannitol concentration (Table 4). To validate these predictions, the predicted values, experimentally validated values, and relative errors of the results were calculated under the respective optimal conditions of the RSM and ANN models, and Table 4 was obtained.

Table 4 The response surface methodology (RSM) yield-vigor model, the artificial neural network(ANN) yield model, the vigor model, and the viable cell number model predictions were validated, and the results were considered credible with a relative error of less than 5%
Full size table

From Table 4, the relative errors between the prediction results and the actual results using RSM and ANN models were less than 5.00%, indicating a high degree of confidence in the prediction results of both models [57]. Considering the protoplast yield, viability and viable cell number, the ANN yield model predicted the optimal results, and the validation resulted in a protoplast yield of 1.550 × 106/100 mg, the viability of 90.65% and a viable cell number of 1.405 × 106/100 mg, which corresponded to the preparation conditions of 1.98% cellulase R-10 content, 1.00% macerozyme R-10 content and 0.50 mol L−1 mannitol concentration. When compared to the RSM model prediction validation results (1.470 × 106/100 mg, 89.73%, 1.319 × 106/100 mg), the indicators showed improvements of 5.44%, 1.03%, and 6.52% respectively. Figure 5 demonstrates the results of the protoplast preparation under these conditions.

Fig. 5
figure 5

Results of the preparation of protoplasts of Salsola laricifolia under the optimal preparation conditions. To enhance visualization of protoplasts, a and b were captured using blood counting plates (25 × 16) as a background under a microscope (10 × 40). Additionally, the protoplast concentrations in b were diluted by a factor of two

Full size image

Reference gene stability analysis

The A260/A280 and A260/A230 ratios of RNA in both protoplasts and untreated leaves of S. laricifolia were approximately 2.00, which satisfied the requirements for subsequent experiments. The relative expression of the three housekee** genes in both protoplasts and leaves was shown in Fig. 6. There was no significant difference in the expression of β-actin and EF1-α pre-and post-treatments. However, there was a significant difference in the expression of 18SrRNA (p < 0.05), indicating that it could not be used as an reference gene in the protoplast experiment.

Fig. 6
figure 6

Relative expression of the three housekee** genes (a, b, c represent 18SRNA, EF1-α, and β-actin, respectively) in the protoplasts and leaves of Salsola laricifolia. * represents significant differences (p < 0.05)

Full size image

Results of protoplast transformation

The results of the transformation of S. laricifolia protoplasts using the pBI121-SaNADP-ME2 plasmid vector were shown in Fig. 7, where fluorescence signals of GFP and chlorophyll were detected respectively at 500–530 nm and 650–750 nm after excitation at 488 nm and 633 nm. The protoplasts transformed with pBI121-SaNADP-ME2 plasmid vector could observe green fluorescence of GFP protein in chloroplasts (Figs. 7a, b, c, d), while protoplasts imported with empty plasmid only emitted green fluorescence in cytoplasm, with no obvious fluorescence in chloroplasts (Figs. 7e, f, g, h). The protoplast cells that were successfully transformed appeared transparent and had a regular shape without any breakage.

Fig. 7
figure 7

Transformation results of protoplasts under laser confocal scanning microscope, green and red fluorescence indicate the localization of GFP protein and chlorophyll, respectively. a, b, c, d: sequentially, pBI121-SaNADP-ME2 plasmid transformed protoplasts in GFP fluorescence, chloroplast autofluorescence, white light and three images superimposed; e, f, g, h: sequentially, pBI121-GFP empty plasmid transformed protoplasts in GFP fluorescence, chloroplast autofluorescence, white light and three images superimposed

Full size image

Discussion

In this study, we utilized RSM and ANN models to predict the best preparation conditions for S. laricifolia protoplasts. Our findings revealed that the Box-Behnken design test provided the necessary data for our analysis. The RSM model demonstrated a better fit compared to the ANN model (Table 3). However, the final protoplast preparation conditions were determined using the ANN-yield model (Table 4). Because the amount of protoplast yield was fluctuated more when changing experimental conditions compared to protoplast viability, leading to the fact that protoplast yield tends to be more determinative of the number of viable cells (Table 3, Fig. 4). Although the ANN-yield model produced a viability value of 90.65%, slightly lower than the predicted value of the ANN-viability model (90.81%), the final number of viable cells obtained was 1.405 × 106/100 mg, which was the highest value (Table 4). As a result, the preparation conditions predicted by the ANN-viability model were chosen as the final conditions. Compared to the previous experimental results of S. laricifolia protoplasts [41], the protoplast yield increased by 28.10%, the protoplast viability increased by 6.65%, and the number of viable cells increased by 36.61%. Furthermore, when compared to the best results before the ANN treatment (Table 1), there was an additional increase of 3.87% in the protoplast viability and 2.78% in the number of viable cells.

After modeling the ANN model using the number of viable cells, the fitting effect of the ANN model was significantly improved, with R, and R2 were next only to the best-fitting RSM-yield model, and the relative error between the experimental validation results and the predicted results was also less than 5.00% (Tables 3, 4). Therefore, it is recommended to consider the number of living cells when using the ANN model to optimize the results of protoplast preparation. This consideration, along with the analysis of protoplast yield and viability, can help obtain predicted results that align more closely with experimental expectations. With the further application of deep learning in cell research, studies have been conducted to identify dead/surviving cells directly using convolutional neural networks [58], which makes it possible to count the number of living cells more easily. Additionally, the fitting ability of RSM and ANN models varies depending on the predicted objects [33, 52, 59,60,61]. Therefore, when different materials are used and steps are taken to prepare protoplasts, they need to be evaluated in detail with specific evaluation indexes and experimental data.

In this study, the expression differences of three housekee** genes (18sRNA, EF1-α, and β-actin) before and after the preparation of pine leaf pigweed protoplasts. were shown in Fig. 6. We optimized the protoplast preparation system and identified suitable internal reference genes to avoid any potential alteration of their expression in the pigweed protoplasts, which could affect the results of subsequent experiments. Housekee** genes are essential for maintaining minimum cellular functions and are generally considered to be stably expressed [62,63,64,65] However, their expression may be altered to varying degrees under different adverse conditions. Unlike β-actin and EF1-α, the expression of 18sRNA selected in this experiment were found to be unstable and unsuitable for use as endogenous genes after the preparation process of S. laricifolia protoplasts (Fig. 5), this aligns with the findings of a previous study that investigated the suitability of these genes as endogenous genes in S. laricifolia leaves under drought stress [53].

Conclusions

In this experiment, Box-Behnken design method with RSM-ANN model was used to optimize the preparation conditions of protoplasts of S. laricifolia, these methods are not commonly used in previously similar studies. To validate the methodology, a series of experiments were conducted, including transient gene conversion and internal reference gene analysis. After considering the protoplast yield, viability and number of viable cells, the ANN yield model predicted the best results, and the experimental validation protoplast yield of 1.550 × 106/100 mg, viability of 90.65%, and a number of viable cells of 1.405 × 106/100 mg. Corresponding preparation conditions were 1.98% cellulase R-10, 1.00% macerozyme R-10, 0.50 mol L−1 mannitol. Furthermore, β-actin and EF1-α were identified as internal reference genes in protoplast experiment. The experiments also demonstrated that, in addition to protoplast yield and viability, the number of viable cells (yield × viability) can serve as an evaluation index for predicting protoplast experiments using RSM and ANN. This index not only considers the viability and yield of protoplasts simultaneously but also aligns better with the requirements of subsequent genetic experiments.