The effect of crystallization level of silica nanoparticles on cell proliferation of MRC-5 cell line and its prediction using artificial neural network

Background: This study investigated the effect of the level of silica nanoparticles (SiO 2 NPs) crystallization on the cell proliferation of MRC-5 cells and its prediction using an artificial neural network (ANN). Methods: Variables studied included temperature (70-1000°C), calcination time (2, 12 and 24 hours), and catalyst feed rate (0.01, 0.05 and 0.1mL/min). Cell proliferation was determined by the MTT test after 24 hours of exposure, and results were analyzed using the t-test in MATLAB. Results: the synthesized particles size was less than 50nm, and the XRD peak varied from 30 to 21° during the increase in calcination temperature. The maximum level of crystallization was at 800°C (58% relative to amorphous) with the lowest cell viability. Cell proliferation decreased with increasing concentration of nanoparticles (p<0.05) and increasing feed rate. There was also a positive relationship between increased crystallization and decreased cell proliferation (R 2 =0.78), but no such association was observed for calcination time. Cell proliferation of MRC-5 was slightly correlated with the linear regression model (MSE>0.12), while ANN was well predicted by the Levenberg–Marquardt algorithm. The suggested structure in this study was 4:10:1 with R 2all =0.97, R 2 test=0.97, RMSE=0.25 and MSE=0.003. The correlation between laboratory results and ANN prediction was 0.94, and the minimum and maximum OD level in the laboratory data and predicted ANN were attributed to 20 and 13 runs. Conclusion: changes in the degree of crystallization of SiO 2 NPs, an increase in concentration, and the rate of catalyst feed during crystallization of SiO 2 NPs were practical factors in increasing cytotoxicity.


Introduction
SiO 2 nanoparticles (NPs) are one of the most common and widely used nanomaterials in industrial and medical fields such as drug delivery, biosensors, cancer treatment, environmental sensors and dispersion of solvent including toluene in industries resulting in increased environmental and occupational exposure (1). SiO 2 NPs are usually available in either amorphous or crystalline forms, the former being classified as industrial environmental products and the latter as natural environmental pollutants. The SiO 2 NPs are initially colloidal, which become amorphous during synthesis and then crystallized by calcination (2). During this process, the amorphous SiO 2 NPs have a long-range order in their crystalline structure and form tetra and hexagonal crystals. Although during the silica crystallization process due to the increase in effective nucleation function of silica, the crystallization of SiO 2 NPs increases (3), but several factors are involved in the SiO 2 NPs crystallization, such as temperature, time, pressure, size of nanoparticles, water content, percentage of elements such as NaCl, addition of calcifying agents, silicatein filaments (4)(5)(6)(7). But it can be said that temperature is one of the most effective and practical factors in the crystallization of many nanoparticles, including silica, whose role in hydrothermal synthesis has been shown to increase their crystallization (8). In many glass and ceramic industries, the crystallization of silica compounds is also carried out by temperature (9). Because of its nature, silica is toxic to human cells, but if it is nano-sized, its toxicity is exacerbated. Previous studies have shown the toxicity of deposited, colloidal, pyrogenic, and crystalline silica on human and mouse cells (10)(11)(12)(13). Although the mechanisms of toxicity of amorphous and crystalline nanoparticles are similar (14), changes in the structure of the material occur during the crystallization process of silica. This could affects many of its properties such as physicochemical ones, biological behavior, and cytotoxicity. SiO 2 NPs penetrate in the plasma membrane at a very tiny size (less than 10 nm) because of their smaller size than the cellular organs, and deposits in the mitochondria or even on the nucleus, and induce the cell death (1). A 24-hr exposure of lung fibroblast cells to SiO 2 NPs is able to induce genotoxicity and cell death (10,12,15,16). The cytotoxicity of SiO 2 NPs depends on their crystallinity. Thus, in occupational studies, exposure to crystalline silica has been shown to cause silicosis and lung disease (17). However, insufficient attention has been paid to the degree of crystallization and the effect of the crystallization process on the cytotoxicity of SiO 2 NPs. In addition, prediction of these effects on lung cells is another aspect of the studies that can be investigated by non-linear, and in some cases, linear modelling.
Among the models, in the case of a multi-agent intervention, the artificial neural network has great power in predicting effects. The objectives of this study was to investigate the effect of SiO 2 NP crystallization rate on cell proliferation in human pulmonary fibroblast (MRC-5) and its prediction using artificial neural networks.

Synthesis of SiO 2 Nanoparticles
Due to the high homogeneity and purity of the sol-gel method, this method has been used for the synthesis of nanoparticles (18) based on the Stober method. According to this method, under ultrasonic conditions, (40 kHz) 5 ml of TEOS was dissolved in 30 ml of ethanol for 10 min. Then, to facilitate ionization, 2 ml of distilled water was added to the solution at a feed rate = 0.2 ml/min under ultrasonic conditions. After 1.5 h, ammonia was added as a catalyst at feed rate = 0.01 ml/min and sonicated for 3 h to obtain a gelatin solution.
Subsequently, for aging, the sample was kept under ambient conditions (1h) and finally washed with ethanol and distilled water, 3 × 7000 rpm (g-force = 5478.2). The product was dried for 24 h at 70 °C and then crystallized at different temperatures (350, 600, 800, and 1000 °C). The degree of crystallization of the nanoparticles was evaluated by X-ray dispersion (XRD). Then, Field emission scanning electron microscopy (FESEM) coupled to energy dispersive spectroscopy (EDS) was used to determine the size and purity of NPs. Finally, the distribution of nanoparticles in solution was determined by Dynamic light scattering (DLS).

Cell culture
Human fibroblastic lung cells (MRC-5) were purchased from Iran Genetics Resource Center. The cell line was cultured under standard conditions (37 °C, 5% of CO 2 , and 90% of humidity). The used medium culture was DMEM enriched with 15% of FBS and 1% pen-strep. Experiments were performed at the passage of 3 after opening from the nitrogen tank.

MTT assay
Cell proliferation was evaluated by studying the mitochondrial dehydrogenase activity in MRC-5 cells.
A number of 15,000 cells/well were cultured in a 96-well plate that cultured for 48 h before exposure.
The cells were exposed to 0.06, 0.1, 0.6, 6 or 60 mg of SiO2 NPs/ml during 24h. Unexposed cells were used as negative controls (100% viability). The MTT assay was performed according to the manufacturer's instructions. Briefly, 10µl of MTT was added to each well. After 4 h, the light absorption was measured at 570 nm wavelengths.

Statistical analysis
Descriptive statistics were used to determine cellular viability. The cellular viability at each exposure concentration and each nanoparticles were calculated using the one-way ANOVA test. Correlation between cellular viability and variables was analyzed by Spearman, and the significance level was set at 0.05. Then linear and nonlinear regression tests were applied to improve its mathematical model to determine the effect of calcination temperature on the nanoparticles. Statistical analyses were performed using Matlab 2018.

ANN modelling
ANN is one of the modeling methods for determining the non-linear relationship between the variables, which was carried out in this study by feed-forward backpropagation and Matlab 2018 (19).
In the neural network, there are several inputs, the hidden layer, and the output, in which the number of neurons in each hidden layer has a very important response. In addition, the relationship between the layers is determined by the assigned weight. Initially, the network is trained with derived laboratory data from cell viability, of which 70, 15, and 15% of the data are used for training, validation, and testing, respectively. Subsequently, the training was used to determine the weight and bias, thus controlling the validation error rate. Then, the training was stopped to avoid an over-fitting increase when the validation error was increased by special iterations (20). Given the significance and accuracy of the mean square error (MSE) and correlation coefficient (R), the best ANN structure for predicting cell proliferation in MRC-5 exposed to SiO 2 NPs, based on MSE and R from the number of neurons, in the hidden layer recommended. The number of neurons in the hidden layer is determined by two equations 1 and 2:

Characterization of SiO 2 NPs
In this study, due to the effect of calcination temperature on the crystallization of nanoparticles (21), the crystallization process of SiO 2 NPs during calcination temperature increase from 70-1000 °C was investigated. As the calcination temperature increased, the maximum angle in the XRD diagram varied from 23° in sample A (dried at 70 °C) to nearly 30° at 800 °C (Fig. S1). Because drying occurred at low temperature (70-150 °C), at which temperature the silica was amorphous (22). In previous studies, the XRD peak for amorphous and crystalline SiO 2 NPs was also 14-21 and 26-27°, respectively (7,23,24). Also, the degree of crystallization, XRD peak and mean diameter of nanoparticles in aqueous media by DLS, and weight percent of elements in SiO 2 NPs by EDS during calcination temperature increase is shown in Table 1. of rapidly agglomeration due to their high surface area and reactivity, resulting in differences in primary and secondary size (27,28). So that particle size reached above 1000 nm, whereas for B-E samples, the mean particle size was <50 nm. Although the results of Table 1 show the process of crystallization of nanoparticles, the increase in temperature did not have a significant effect on the crystallization of silica. Thus, the increase in calcination time was used as a variable to increase the crystallization of the silica, the results of showed increasing of degree for the XRD peak from 2 h to 24 h of calcination time. Thus, the crystallization of nanoparticles increased as calcination time increased. That, the crystallization of SiO 2 NPs at 1000 °C during 24 h was longer because there is sufficient opportunity for the development of regular SiO 2 crystals with increasing time.

The effect of the feed rate of catalyst on the crystallinity of SiO 2 NPs
In this study, used silica nanoparticles were smaller than 50 nm. Because smaller sizes of nanoparticles are more likely to penetrate the cell. Due to the effect of the feed rate of catalyst on nanoparticle size (29), crystallization level, purity percentage, size in aqueous media, and XRD peak degree at feed rates of 0.01, 0.05 and 0.1 ml/min are shown in Table 2, respectively. According to Table 2, the rate of crystallization of nanoparticles decreased with increasing feed rate.
So that the maximum crystallization was at feed rate = 0.01 ml/min. Because higher feed rates do not provide sufficient time for the catalytic reaction, and because of the lower surface-area-to-volume ratio of catalyst droplets, many TEOS molecules may remain unchanged. FESEM and DLS also showed particle size <50nm under these conditions. Because these nanoparticles are calcinated at high temperatures, which made them crystallize, and as a result, their agglomeration has decreased.

Toxicity of SiO 2 NPs based on calcination temperature
In this study, cellular viability due to exposure of MRC-5 to synthesized SiO 2 NPs at various temperatures, times of calcination, and different feed rates, were determined by the MTT test, as shown in Fig. 1, 2 and 3, respectively.
As shown in Fig. 1 viability also decreased with an increasing concentration of 250-1500 μg/ml (32). Also, the lethality at 600 μg/ml concentration was higher than 60μg/ml (p = 0.01). In a study by Petrache Voicu et al., 2015, cell viability also decreased dramatically with increasing nanoparticle concentrations, from 12 to 62 μg/ml. Cytoplasmic vacuolization was also observed at concentrations above 60 µg/ml. Because at higher concentrations, the cell is more likely to be exposed to nanoparticles (16). According to Fig.   1, cell survival was increased during the calcination temperature up to 600 °C (at concentrations of 60-600 μg/ml) but then decreased. So that the maximum and minimum relative viability were observed at 600 °C and amorphous. However, the difference between relative viability in A and E was statistically significant only at 60μg / ml (p = 0.0256). Because A NPs are amorphous, they have a greater reactivity with MRC-5 due to the high surface area to volume ratio. However, at higher temperatures, due to the regular structure and reduced surface area of nanoparticles, their reaction capacity and catalytic activity have decreased which is effective in reducing their toxicity (14,33).
However, the results indicated that the increase in calcination temperature did not significantly change the degree of crystallization, but, due to the formation of regular crystals at 1000 °C, the maximum toxicity of MRC-5 at concentrations of 600 and 6000 µg/ml was due to exposure to D nanoparticles. In addition, the difference between OD in D and E nanoparticles was significant only at

Toxicity of SiO 2 NPs based on calcination time at 1000°C
The degree of cytotoxicity induced by increasing the calcination time of E nanoparticles at 1000°C also shows that there is no linear pattern for OD changes in nanoparticles with different crystallization times. So that with increasing time, OD increased at 6000 μg/ml concentration. However, the maximum toxicity at concentrations of 60 and 6000 μg/ml in the calcinated nanoparticles was at 12 h.
Also, at a concentration of 100 μg/ml, there was no difference between cell proliferations (p> 0.05).
As the crystallization of silica nanoparticles increases, the relationship between concentration variables and the degree of crystallization has changed with the degree of cell proliferations, which needs further evaluation.

Toxicity of SiO 2 NPs based on catalyst feed rate at 1000°C
Based on the used method in the synthesis of SiO 2 NPs, details of the cell proliferations of MRC-5 at three concentrations of 42, 420, and 4200 μg/ml are shown in Fig. 3.
As shown in Fig. 3, cell proliferations was different in various feed rates. So the minimum cell proliferations was at feed rate = 0.05 ml/min (p <0.05). Pearson correlation also showed that at 42μg/ml concentration, there was a negative correlation between cell proliferation and feed rate increase (R 2 = -0.97). Also, at 0.01 ml/min, the maximum toxicity was at 4200 μg/ml (p> 0.05), while at B and C nanoparticles, the minimum cell proliferations was at 42 μg/ml. However, only in C nanoparticles, there was a significant difference between the cell proliferations of different concentrations (p <0.05). Thus, increasing the catalyst feed rate during the nanoparticle synthesis steps is an effective factor in decreasing cell proliferations, and it is suggested that the feed rate decrease. Because, according to SEM results, the change in feed rate did not affect the physicochemical properties of the nanoparticles, such as size, but was effective in inducing nanoparticle toxicity in lung cells.

Prediction of Cellular Viability using ANN
In this study, cell proliferation in MRC-5 exposed to different concentrations of SiO 2 NPs was first studied by the linear regression model. In Table 3, error rates and correlation coefficients for different concentrations are shown by using the linear regression model. According to Table 3, the cell proliferation was only more consistent with the linear regression model at the concentration of 100 μg/ml (MSE = 0.08, RMSE = 0.26 and R 2 = 0.78). In this way, the artificial neural network method, which is a nonlinear method, is used for predicting and modeling. The Levenberg-Marquardt algorithm for weight and bias was used to determine the best network structure. In this study, based on the number of input and output variables, the number of neurons in the hidden layer = 3 -10 was used to predict cell proliferations by using ANN, which the best of it was related to 10 neurons, according to Table 5. Because it had the highest correlation coefficient and the lowest MSE for all datasets (0.97) and test datasets (0.98). Figure 4 shows the structure of the best predictive model and the relationship between input and output by the hidden layer. The minimum MSE = 0.003 (Fig. 4a) was also at Epoch = 1, which after this point, overfit data and validation set error began to increase. When the cell proliferations error is increased for seven iterations, the training is stopped based on the stop algorithm. This prevents network overfits. Also, the regression coefficient in Fig. 4b shows that the correlation in test dataset = 0.98 and all test = 0.97. The prediction rate by using ANN compared with the experimental value is shown in Table 4. The maximum and minimum cell proliferation in real samples was 0.1 and 0.001 M, respectively, which was at the calcination temperature of 1000°C with feed rate = 0.01 ml/min and HRT = 2 h at runs 20 and 13, respectively. The results of the ANN prediction were similar, too. In Fig. 5

Conclusion
In this study, the effects of temperature and time of calcination and feed rate of catalyst on the crystallization rate of SiO 2 NPs were investigated, and then their cell proliferations on MRC-5 during 24 h exposure was studied.
The results of the study are shown as follows: The size of the synthesized nanoparticles was less than 50 nm in all conditions. Whereas DLS results showed that the primary and secondary size of amorphous nanoparticles changed in aqueous media.
Thus, unlike other investigated nanoparticles in this study, amorphous SiO 2 NPs rapidly agglomerated, which can affect its properties (primary diameter <50 nm and secondary diameter> 1000 nm).
During the calcination temperature up to 1000 ° C, the peak of the XRD diagram changed from 21-30 °, indicating a change in their crystallization. The maximum crystallization rate was observed at 1000 °C, which was 58% higher than amorphous. The EDS results also confirm that with increasing calcinated temperature, the purity of the nanoparticles increased (only Si and O were observed in the crystals).
Thus, increasing calcination temperature has been an effective factor in the crystallization of SiO 2 Nanoparticle crystallization increased with increasing calcination time and reduced feed rate but did not have a significant effect on MRC-5 cell proliferations.
By increasing the concentration of nanoparticles and increasing the crystalinity of amorphous cells, cell proliferation decreased. Thus, the crystallization of SiO 2 NPs is an effective factor in MRC-5 cell death.
An increase in calcination temperature was highly correlated (R 2 = 0.78) with cell proliferation at 600 μg/ml concentration, whereas no correlation was observed at other concentrations. Thus, besides to the crystallization of the nanoparticles, the concentration is also an effective factor in MRC-5 cell death.
Cell proliferation was positively correlated at concentrations of 60 and 600 μg/ml (R 2 = 0.5) while negatively correlated with concentrations of 6000 μg/ml (R 2 = -0.54 and R 2 = -0.79). Consequently, with logarithmic increasing of SiO 2 NPs concentration, no logarithmic decrease in cell proliferation has been observed, which requires the use of other prediction models. Cell proliferation in MRC-5 was slightly correlated with linear regression model (MSE> 0.08 and RMSE = 0.26), but modeling results by using artificial neural network showed that the best structure for this study was 4:10:1 with R 2 all = 0.97,

Funding
Shiraz University of medical science.

Competing interests
There are no competing interest.

Availability of supporting data
The supporting data are available from the corresponding author on reasonable request.

Ethics approval and consent to participate
All ethical aspects of this study were approved by Shiraz University of medical science' Ethics Committee.

Consent for publication
Not applicable.    The structure of best model for prediction of cellular proliferation related to calcination temperature Figure 5 Cell proliferation of MRC-5 exposed to SiO2 NPs for the actual laboratory results and the predicted by ANN model.