Decrease in membrane fluidity and traction force induced by silica-coated magnetic nanoparticles

Background Nanoparticles are being increasingly used in biomedical applications owing to their unique physical and chemical properties and small size. However, their biophysical assessment and evaluation of side-effects remain challenging. We addressed this issue by investigating the effects of silica-coated magnetic nanoparticles containing rhodamine B isothiocyanate [MNPs@SiO2(RITC)] on biophysical aspects, such as membrane fluidity and traction force of human embryonic kidney 293 (HEK293) cells. We further extended our understanding on the biophysical effects of nanoparticles on cells using a combination of metabolic profiling and transcriptomic network analysis. Results Overdose (1.0 μg/µL) treatment with MNPs@SiO2(RITC) induced lipid peroxidation and decreased membrane fluidity in HEK293 cells. In addition, HEK293 cells were morphologically shrunk, and their aspect ratio was significantly decreased. We found that each traction force (measured in micropillar) was increased, thereby increasing the total traction force in MNPs@SiO2(RITC)-treated HEK293 cells. Due to the reduction in membrane fluidity and elevation of traction force, the velocity of cell movement was also significantly decreased. Moreover, intracellular level of adenosine triphosphate (ATP) was also decreased in a dose-dependent manner upon treatment with MNPs@SiO2(RITC). To understand these biophysical changes in cells, we analysed the transcriptome and metabolic profiles and generated a metabotranscriptomics network, which revealed relationships among peroxidation of lipids, focal adhesion, cell movement, and related genes and metabolites. Furthermore, in silico prediction of the network showed increment in the peroxidation of lipids and suppression of focal adhesion and cell movement. Conclusion Taken together, our results demonstrated that overdose of MNPs@SiO2(RITC) impairs cellular movement, followed by changes in the biophysical properties of cells, thus highlighting the need for biophysical assessment of nanoparticle-induced side-effects.


Background
The use of nanoparticles for diagnostic and therapeutic purposes has been rapidly increasing in medicine. However, as the small size of nanoparticles is known to enable their cellular entry and accumulation, this could potentially cause cellular dysfunction [1][2][3][4]. Moreover, compared with bulk materials, nanoparticles are known to be more reactive and owing to their higher surfaceto-volume ratio might exhibit more side-effects, such as the generation of reactive oxygen species (ROS) [5][6][7]. Nevertheless, our current knowledge regarding the effects of nanoparticles on specific physical and mechanobiological aspects of the cell remains insufficient due to limitations in the available analytical methods.
Briefly, ROS are known to oxidize membranous lipids as well as cytoskeleton proteins. In particular, ROS can oxidize polyunsaturated phospholipids, glycolipids, and cholesterol in the membrane [23], thereby reducing membrane fluidity and permeability [24,25]. In addition, internalised nanoparticles have been shown to impair cytoskeleton proteins and interfere with focal adhesion kinase-mediated signalling [26]. With lamellipodia (branched actin filaments) and filopodia (extended finger-like protrusions), cell adhesion is known to occur due to the presence of focal adhesion (FA) complexes, with cell morphology being determined through the balance between adhesion and tension [27]. The involvement of adenosine triphosphate (ATP) in the movement of myosin over filamentous actin (F-actin) and in the polymerization of actin has suggested that cellular mechanics could be altered by ATP depletion [28]. In that regard, treatment with MNPs@SiO 2 (RITC) has been reported to have caused metabolic changes in cells, including ATP depletion [21].
The submicron elastomeric pillar array is considered an excellent tool for measuring cellular force as the nanometric level of pillar deflection can be calculated by incorporating optical microscope imaging [29,30]. Furthermore, it can be used to analyse the initial contact of a cell with a substrate and has been reported to mimic continuous substrates of a specific rigidity [31]. Thus, the mechanobiological effects of nanoparticles on cells could be quantitatively studied by measuring the traction force using submicron elastomeric pillars [29,31].
Assessment of potential side-effects of nanoparticles using classical methods is limited owing to delicate changes and complexities at the nano level. Thus, instead of focusing on targeted molecules, omics approaches, including genomics, transcriptomics, proteomics, and metabolomics, have been used in nanotoxicity studies [22]. However, restrictions in assessing the intricate signalling pathways and delicate events in cells and organisms still exist. For example, although transcriptomics can reflect enormous genotypic changes, it has been shown to be insufficient for understanding the actual phenotype [32,33]. In contrast, in the case of metabolomics, which can constitute an endpoint feature of biological phenotypes [34,35], no amplification methods are available for very low-abundance metabolites, and quantitative analysis of the targeted method might likely provide only a partial representation of the overall metabolism [36]. Accordingly, a combination of transcriptomics and metabolomics, termed "metabotranscriptomics", has been utilised for a more comprehensive analysis of cells and evaluation of nanotoxicity following treatment with nanoparticles [4,21,22].
We have previously analysed the effects of MNPs@ SiO 2 (RITC) on HEK293 cells [4,21,37,38]. In addition, HEK293 cells have also been extensively used in the study of nanoparticle-induced toxicity and in renal toxicity models [39,40]. In the present study, we analysed membrane fluidity, which is an important indicator of biophysical changes, including cell freshness and cell deformability [41,42], in MNPs@SiO 2 (RITC)treated HEK293 cells using total internal reflection fluorescence microscopy (TIRFM). Moreover, we measured the changes in the traction force, which is known to directly reflect biophysical changes and has been highly correlated with changes in cell membrane in MNPs@SiO 2 (RITC)-treated HEK293 cells. Thus, metabotranscriptomics-based biophysical assessment was performed to comprehensively evaluate MNPs@ SiO 2 (RITC)-induced effects on HEK293 cells.

Characteristics of MNPs@SiO 2 (RITC) and silica NPs
Our generated MNPs@SiO 2 (RITC) nanoparticles consisted of an approximately 9-nm cobalt ferrite core (CoFe 2 O 3 ) chemically bonded to rhodamine isothiocyanate dye (RITC) and coated with a silica shell [13]. Moreover, silica nanoparticles, which are identical to the silica shell of MNPs@SiO 2 (RITC), have been shown to exhibit similar biological effects to those exhibited by MNPs@SiO 2 (RITC) [4,20,21]. The diameter of both MNPs@SiO 2 (RITC) and silica nanoparticles was demonstrated to be 50 nm (Additional file 1: Fig. S1), with MNPs@SiO 2 (RITC) being reported to have a zeta potential of − 40 to − 30 mV [13,20]. A previous study using inductively coupled plasma atomic emission spectrometry showed that approximately 10 5 particles of MNPs@ SiO 2 (RITC) per cell were internalised in MCF-7 breast cancer cells [13]. We determined the dosage used in the present study by treating HEK293 cells with MNPs@ SiO 2 (RITC) at concentrations ranging from 0.01 to 2.0 µg/µL for 12 h and calculating their uptake efficiencies [21]. We accordingly found that the optimal concentration of MNPs@SiO 2 (RITC) for in vitro use was 0.1 µg/ µL; this concentration had been previously used for MRI contrast without any reported toxicological effects in human cord blood-derived mesenchymal stem cells [43]. In addition, any disturbances in gene expression and metabolic profiles of treated cells at this concentration were similar to those observed in control cells [21]. We further observed that the uptake efficiency of MNPs@ SiO 2 (RITC) plateaued at 1.0 µg/µL. Therefore, we used a low dose of 0.1 µg/µL and high dose of 1.0 µg/µL in the present study.

MNPs@SiO 2 (RITC)-induced lipid peroxidation decreases cell membrane fluidity
We determined the optimal concentrations of MNPs@ SiO 2 (RITC) for cell labelling, as well as the cellular uptake of nanoparticles during a 12 h-incubation to be 0.1 and 1.0 μg/µL, respectively, based on our previous studies using a fluorescence assessment method [4,21]. We found that the viability of cells treated with 1.0 μg/ µL MNPs@SiO 2 (RITC) was similar to that of cells with or without treatment with bare silica nanoparticles (NPs) of the same size (50 nm in diameter) (Additional file 1: Fig.  S2), consistent with previous reports [4,21].
We analysed the changes in lipid peroxidation and membrane fluidity in treated HEK293 cells using total internal reflection fluorescence microscopy (TIRFM). We furthermore investigated the changes in cell membrane fluidity after treatment with MNPs@SiO 2 (RITC) by measuring the values of 6-dodecanoyl-2-dimethylaminonaphthalene (laurdan) generalised polarization (GP) using TIRFM (Fig. 1a). We accordingly noted that the number of high-GP areas on the cell surface, corresponding to rigid domains, increased on treatment with MNPs@SiO 2 (RITC); in particular, abundantly distributed regions of MNPs@SiO 2 (RITC) were found to be primarily colocalised with high GP-distributed regions at a GP scale of −1.0 to 1.0 (Fig. 1b, Additional file 1: Fig. S3). We subtracted the GP frequency distribution values of treated cells from the corresponding values of untreated control cells to obtain frequency difference curves (Fig. 1c) and total mean GP values (Fig. 1d). We observed a similar trend in the relative levels of peroxidised lipids (Fig. 1e).
A previous study revealed that MNPs@SiO 2 (RITC) induced the generation of intracellular ROS via mitochondria dysfunction in HEK293 cells. In particular, generation of ROS in cells treated for 12 h was attributed to the shell of MNPs@SiO 2 (RITC) and silica nanoparticles (silica NPs) rather than the cobalt ferrite core [4,21]. Thus, to determine the MNPs@SiO 2 (RITC) treatmentinduced generation of intracellular ROS, we performed 2′,7′-dichlorodihydrofluorescein diacetate staining in HEK293 cells treated with MNPs@SiO 2 (RITC) and silica NPs. We observed that the level of intracellular ROS was increased by over 50% upon treatment with both 1.0 µg/µL MNPs@SiO 2 (RITC) and silica NPs compared with that in untreated control and 0.1 µg/µL MNPs@ SiO 2 (RITC)-and silica NPs-treated cells (Additional file 1: Fig. S4). In addition, we did not observe any significant difference in the treatment with 1.0 µg/µL MNPs@ SiO 2 (RITC) and silica NPs, in line with previous reports [4,21]. These results indicated that the rigid regions in the plasma membrane were increased through lipid peroxidation induced by ROS generated from the shell of MNPs@SiO 2 (RITC).

Treatment with MNPs@SiO 2 (RITC) decreased cell aspect ratio and spread area but increased traction force
As membrane fluidity is known to be closely related to cell morphology and focal adhesion [44,45], we investigated whether a decrease in MNPs@SiO 2 (RITC)-mediated membrane fluidity would also affect cell morphology and focal adhesion. We evaluated the effects of MNPs@ SiO 2 (RITC) on cell polarity as local cell contraction is tightly associated with these activities along with changes in focal adhesion [46]. Moreover, we measured the cell aspect ratio. We analysed images of cells and submicron pillars at 12 h after cell seeding (Fig. 2a, b). We found that the aspect ratio of cells treated with 0.1 µg/µL MNPs@ SiO 2 (RITC) did not significantly differ from that of untreated control cells. However, the ratio of cells treated with 1.0 µg/µL MNPs@SiO 2 (RITC) was shown to be significantly smaller than that of untreated control cells (Fig. 2c).
We then used pillar deflection in the magnified images to measure pillar displacement (Fig. 2d) and calculate traction force (Fig. 2e, f ). To calculate pillar traction force, the displacement of each pillar was multiplied with the pillar bending stiffness [31]. In particular, we noted that 1.0 µg/µL MNPs@SiO 2 (RITC)-treated cells showed an increase in pillar displacement (351 ± 65 nm; mean ± SD), which was significantly higher than that in untreated control cells (216 ± 46 nm). The average (8.5 ± 2 nN) and total (1112 ± 197 nN) traction forces in 1.0 µg/µL MNPs@SiO 2 (RITC)-treated cells were significantly higher than those (5.2 ± 1 and 855 ± 172 nN) in untreated control cells, indicating that cell traction force was affected by 1.0 µg/µL MNPs@SiO 2 (RITC). Taken together, the increase in cell traction force was suggested to have been caused by the MNPs@SiO 2 (RITC) treatment-induced reduction in cell polarity and spread area.

Treatment with MNPs@SiO 2 (RITC)-impaired cell movement
We previously analysed the effect of treatment with MNPs@SiO 2 (RITC) on the migratory activity of human bone marrow-derived mesenchymal stem cells (hBM-MSCs). Using conventional assays, such as scratch and invasion assays, we observed an impairment in the migratory activity of hBM-MSCs [42]. To evaluate the biophysical changes related to biological functions of HEK293, we analysed the effect of 0.1 or 1.0 µg/ µL MNPs@SiO 2 (RITC) on the movement of HEK293 cells using conventional assays. Accordingly, using the scratch assay, we did not observe any difference in the migratory activity between MNPs@SiO 2 (RITC)-treated HEK293 cells and untreated control cells (Fig. 3a). Similarly, no difference was observed between the invasion ability of MNPs@SiO 2 (RITC)-treated HEK293 cells and untreated control cells analysed using the transwell invasion assay (Fig. 3b).
However, it should be mentioned that the aforementioned results of the scratch and invasion assays did not exclude the cell growth effect and treatment with growth-arrest agents, such as mitomycin C, as these assays are known to be highly toxic to HEK293 cells [47]. Thus, we analysed the movement of individual cells by tracking cells on pillars for 24 h after treatment with MNPs@SiO 2 (RITC) for 12 h on a dish (Additional files 2, 3, 4: Movie S1, Movie S2, and Movie S3). We remarked that the distances travelled by cells were decreased to a significantly greater extent in MNPs@ SiO 2 (RITC)-treated HEK293 cells compared with untreated control cells at both 6 and 24 h (Fig. 3c). In addition, we obtained similar results for movement speeds at 6 h and 24 h (Fig. 3d).

Treatment with MNPs@SiO 2 (RITC) decreased intracellular level of ATP
To evaluate the changes in the intracellular level of ATP in MNPs@SiO 2 (RITC)-treated cells, HEK293 cells were treated with MNPs@SiO 2 (RITC) at concentrations ranging from 0 to 2.0 µg/µL for 6, 12, and 24 h (Fig. 4). We found that intracellular levels of ATP were decreased in a dose-dependent manner in MNPs@ SiO 2 (RITC)-treated cells, starting from the 0.13 µg/µL dose. Moreover, we observed that the decrement pattern was similar for 6, 12, and 24 h treatments.

Metabotranscriptomic network of MNPs@ SiO 2 (RITC)-treated HEK293 cells
To analyse the dispersion phenomena of MNPs@ SiO 2 (RITC)-treated cells, we constructed a coexpression network of genes and metabolites using a transcriptome generated from microarray analysis and the metabolome derived from amino acid and organic acid profiling through the Ingenuity Pathway Analysis (IPA, http://www.ingen uity.com) [21]. Microarray expression analysis showed that the levels of 21 and 31 genes associated with lipid peroxidation and focal adhesion formation, respectively, were altered in MNPs@ SiO 2 (RITC)-treated cells (Fig. 5a, b). Using a threefold cut-off only to determine changes in the transcriptomic network, we detected more pronounced changes in 1.0 µg/µL MNPs@SiO 2 (RITC)-treated cells compared with 0.1 µg/µL MNPs@SiO 2 (RITC)-treated and control cells; these changes were related to cell movement (Fig. 5c, Additional file 1: Fig. S5). In silico prediction of the network revealed activation of lipid peroxidation, as well as suppression of focal adhesion and cell movement in 1.0 µg/µL MNPs@SiO 2 (RITC)-treated HEK293 cells (Fig. 5d).
Although transcriptomics provided comprehensive information regarding MNPs@SiO 2 (RITC)-treated cells, the data obtained were only qualitative. Thus, for a network-based evaluation, we used a combination of transcriptome and metabolome networks, termed the metabotranscriptomic network, through substitution of amino acid and organic acid profiles (cut-off ± 20% change), as described in our previous report [21]. In the group treated with 1.0 µg/µL MNPs@SiO 2 (RITC), we found that the amino acid, organic acid, and free fatty acid profiles showed increased levels of tyrosine, pyruvic acid, glutamic acid, lysine, and lignoceric acid; in contrast, we observed decreased levels of cysteine, asparagine, 3-hydroxybutyric acid, glutamine, serine, aspartic acid, oxaloacetic acid, glycine, and acetoacetic acid [37]. This combined network provided more reliable information than the transcriptomic network alone, allowing also the detection of more pronounced changes in 1.0 µg/µL relative to 0.1 µg/µL MNPs@SiO 2 (RITC)-treated and control cells (Fig. 6a, Additional file 1: Fig. S6, Table S1). In silico prediction of the integrated network also revealed a similar trend to that observed in the transcriptome network (Additional file 1: Fig. S7). We subsequently quantified the expression level of genes associated with the network using semi-quantitative reverse transcription PCR (Fig. 6b) and quantitative (q) PCR (Fig. 6c). In particular, we found that expression levels of superoxide dismutase 2 (SOD2) and LIM zinc finger domain-containing 1 (LIMS1) were increased, whereas those of NCK adaptor protein 1 (NCK1) and complement C3a receptor 1 (C3AR1) were decreased in the 1.0 µg/µL MNPs@SiO 2 (RITC)-treated cells relative to those in untreated control cells.

Discussion
The present study demonstrated the effects of nanoparticles on the biophysical properties of cells. In addition, functional analysis using the metabotranscriptomics approach allowed us to deduce inter-connective relationships between genes and metabolites associated with lipid peroxidation, focal adhesion, and cell movement in MNPs@SiO 2 (RITC)-treated cells.
We observed that the 1.0 µg/µL MNPs@SiO 2 (RITC)treated cells exhibited markedly increased levels of lipid peroxidation compared with the 0.1 µg/µL MNPs@ SiO 2 (RITC)-treated cells and untreated control. These results indicated that exposure of cells to 1.0 µg/µL MNPs@SiO 2 (RITC) might impair certain biological functions due to the production of ROS and lipid peroxidation. This was consistent with the finding that oxidative stress induces lipid peroxidation, which is often observed in disease states, such as aging, sickle-cell disease, malaria, and diabetes [48][49][50]. It was also consistent with the decrease in the deformability of hBM-MSCs over time owing to the generation of intracellular ROS and lipid peroxidation, which was shown to lead to decreased membrane fluidity and deteriorated cell quality [41]. Moreover, we previously showed that ROS-distributed regions distinctly co-localised with MNPs@SiO 2 (RITC)distributed regions in cells [21]; hence, we assumed that regions of abundantly distributed MNPs@SiO 2 (RITC) might be primarily co-localised with high GP-distributed regions along with lipid peroxidation, thus distinguishing damaged membrane regions from intracellular regions. Thus, our results suggested that exposure of cells to 1.0 µg/µL MNPs@SiO 2 (RITC) could potentially impair the biological functions of exposed cells by increasing their membrane rigidity. Nanoparticle-induced cell membrane damage is reportedly caused through direct interactions between membrane lipids and nanoparticles [51]. Further, silica and gold nanoparticles are known to penetrate cell membranes [52,53]. In particular, 50-nm NPs were demonstrated to be likely internalised through endocytic pathways as nanoparticles > 10 nm were found to be coated directly with the plasma membrane and be internalized via clathrin-and caveolae-mediated endocytic pathways [54,55]. In this study, the altered gene expression of the clathrin-mediated endocytic pathway was reported to significantly alter the internalization of MNPs@SiO 2 (RITC)s in HEK293 cells and SH-SY5Y neuroblastoma cells [4]. Thus, we proposed that MNPs@ SiO 2 (RITC) enter cells through endocytosis; this might constitute one reason for the observed decreased membrane fluidity because membrane tension regulates membrane deformation or changes in cell shape (e.g., endocytosis, exocytosis, cytokinesis, and cell motility) [56].
As the membrane and cytoskeleton are known to be tightly associated with phosphoinositides, the status of cell membranes has been highly linked to the regulation of the cytoskeleton [57,58]. Nanoparticle-induced lipid peroxidation has been reported to alter cell morphology and membrane roughness in human lymphocytes [59]. We accordingly found that the cell aspect ratio and cell spread area were decreased in MNPs@ SiO 2 (RITC)-treated cells, which exhibited a round morphology, suggesting that focal adhesion was impaired owing to the imbalance between adhesion and tension caused by treatment with MNPs@SiO 2 (RITC) treatment. Nanoparticles have also been shown to increase traction force but inhibit cellular migration [60]. Traction force is known to be primarily regulated by phosphorylation of myosin II through the assembly (dephosphorylated myosin) and disassembly (phosphorylated myosin) of the myosin light chain (also termed the myosin head) and actin filaments, with these processes obtaining energy from the hydrolysis of ATP [61,62]. Our previous study showed that the intracellular level of ATP in 1.0 µg/ µL MNPs@SiO 2 (RITC)-treated cells was decreased to approximately 50% due to mitochondrial damage [21]. In the present study, we observed that treatment with MNPs@SiO 2 (RITC) decreased the intracellular levels of ATP in dose-and time-dependent manners. This might have been one of the factors responsible for the observed increment in the total traction force and reduction in cell movement activity.
There were limitations in interpreting the actual phenotype obtained in the transcriptome network associated with lipid peroxidation, focal adhesion, and cell movement. Thus, we combined datasets providing information on the interactions between differentially expressed genes and altered metabolites, and generated a metabotranscriptomic network. In particular, the network showed that the expression levels of SOD2, which is tightly associated with oxidative stress [63], and LIMS1, which is an important molecule in the linkage between actin and integrin for the formation of focal adhesion [64], were increased, whereas the levels of NCK1, an upstream regulator in the formation of actin-rich protrusions in the plasma membrane [65], and C3AR1, which is reduced in oxidative conditions [66,67], were decreased in 1.0 µg/µL MNPs@SiO 2 (RITC)-treated cells relative to those in untreated controls. Thus, the convergence of the metabolic profiles and transcriptome disturbances allowed interpretation of the changes in the movement of MNPs@SiO 2 (RITC)-treated cells.
Dose justification is determined based on uptake (labelling) efficiency, and using a fluorescence assessment method, we previously reported that the plateau of uptake (labelling) occurred at 1.0 μg/μL MNPs@ SiO 2 (RITC)-treated HEK293 cells [21]. Moreover, with respect to toxicity, reduction in the cell viability of MNPs@SiO 2 (RITC) treated HEK293 cells was not identified. However, changes in ATP levels were observed with respect to NP concentration, that is, an approximately 50% decrement in the level of ATP in 1.0 μg/μL MNPs@ SiO 2 (RITC)-treated HEK293 cells exhibiting a plateau uptake, whereas an approximately 5% decrement in the level of ATP in 0.1 μg/μL MNPs@SiO 2 (RITC)-treated HEK293 cells with an adequate uptake.
The effects of MNPs@SiO 2 (RITC) and silica NPs on HEK293 cells were similar with respect to cell viability and ROS generation. In addition, cellular localization of MNPs@SiO 2 (RITC) and silica NPs indicated that both types of NPs were accumulated in lysosomes (Additional file 1: Fig. S8). Consistent with our previous reports, we postulated that the effects of MNPs@SiO 2 (RITC) were exerted by silica, which is a peripheral part of the MNPs@SiO 2 (RITC) [4,21,42].
The decrease in moving speed was recovered in control and MNPs@SiO 2 (RITC)-treated HEK293 cells in a timedependent manner. This could be explained by a couple of possibilities. One possibility is that cells adapted to the altered conditions. Another possibility is that the MNPs@SiO 2 (RITC)-induced decrement in moving speed  (Fig. 4), the decreased levels of ATP were recovered, but not fully. These possibilities might have been reflected in the moving speed of MNPs@SiO 2 (RITC)treated HEK293 cells. The possibility of the dye and cobalt ferrite leaching out of MNPs@SiO 2 (RITC) and exerting toxic effects in cells exposed to these nanoparticles for a long time has been a main concern. For instance, the dye signal of MNPs@ SiO 2 (RITC) was reduced by approximately 90 and 40% in human cord blood-derived mesenchymal stem cells during their third and seventh passages, respectively. These observed reductions might have been due to the dilution effect generated by cell proliferation [19]. As there have been reports that treatment with MNPs@ SiO 2 (RITC) did not cause any significant loss and measurable toxicological effects in HEK293 cells for 7 d and in a mouse model for 4 wk, respectively, it was assumed that the highly toxic cobalt ferrite substance did not leach out of the core of MNPs@SiO 2 (RITC) [4,18,21]. Based on these reports, we assumed that our MNPs@SiO 2 (RITC) nanoparticles were also stable enough to allow the discernment of subtle toxicity in cells exposed to MNPs@ SiO 2 (RITC) for a long time. RT-PCR and qPCR were performed using gene-specific primer pairs for SOD2, LIMS1, NCK1, and C3AR1. GAPDH was used as the internal control. PCR products were normalised relative to the levels of internal control. Data represent the mean ± SD of 3 independent experiments. *p < 0.05 vs untreated control. Transcriptome and metabolome data are reproduced from our previous study, Copyright © 2019 Springer Nature [37] should be used at the lowest possible dosage in therapeutic or diagnostic applications to prevent potential nanotoxicity. Our comprehensive approach of toxicological evaluation will aid the future assessment of nanoparticle sensitivity and associated potential toxicity. Moreover, our findings on nanotoxicity will be beneficial in terms of the development of safe nanoparticles for biomedical applications.

Microfabrication of submicron elastomeric pillar array
Photolithography was used to fabricate a silicon wafer mould with a series of holes [72]. To make the pillars, polydimethylsiloxane (PDMS) was mixed in a 10:1 ratio with its curing agent (Sylgard 184; Dow Corning, Midland, MI, USA) and vacuumed for 15 min; the mould was then spin-coated with PDMS at 1,000 RPM for 1 min and again vacuumed for 30 min to remove bubbles. The mould with PDMS was cured at 80 °C for 3 h.
The pillar array was manufactured using this method. Each pillar was 900 nm in diameter (D), 1 µm in height (L), and 1.8 µm in centre-to-centre distance. To calculate the bending stiffness (k) of the pillar, the Euler-Bernoulli beam theory was applied [31]: where E is the Young's modulus (2 MPa) of the cured PDMS. Using this equation, the k of the pillar was calculated to be 24.2 nN/µm.

Measurement of cell aspect ratio
The cell aspect ratio was measured by representing the area of the entire cell with a comparable elliptical shape using ImageJ (NIH); the division from the major axis of the ellipse to the minor axis was then taken as the cell aspect ratio.

Measurement of traction force
Pillar images were acquired using a fluorescence microscope (Deltavision, GE Healthcare, Chicago, IL, USA) and a camera (CoolSNAP HQ 2 , Photometrics) at 37 °C and 5% humidity with a live cell chamber at 1 Hz. The location of each pillar in each frame was determined using the pillar tracking plugin (PillarTracker 1.1.3 version) for ImageJ. In the pillar reconstruction algorithm, the PillarTracker works to establish the exact grid of the pillar array, thereby allowing users to automatically detect and track the pillar locations. Throughout this study, pillars that had no contact with cells were used as reference pillars. To account for stage drift, the average displacement of reference pillars was deducted from the displacement data of pillars deflected by cells. To avoid the unwanted displacement of pillars by MNPs@ SiO 2 (RITC), MNPs@SiO 2 (RITC)-treated cells were washed five times using PBS (Sigma-Aldrich, St. Louis, MO, USA) before seeding onto the pillar array. The displacement of each pillar was multiplied with its bending stiffness to calculate the traction force (F).

Scratch assay
HEK293 cells were seeded and cultured to 100% confluence in 6-well plates. Cell monolayers were scratched using 200-μL micropipette tips. Media were replaced with serum-free media, and HEK293 cells were treated with MNPs@SiO 2 (RITC) for 16 h. Images of scratched areas were captured both before treatment (0 h) and after treatment (16 h) using an Axio Vert 200 M fluorescence microscope (Zeiss, Jena, Germany). Migration activity was quantified by counting the number of migrated cells from the initial scratch.

Transwell invasion assay
Cell invasion activity was analysed using an 8-µm pore size Transwell polycarbonate membrane (Corning, CA, USA), as in a previous study [73]. The upper side of the insert was coated with Matrigel (1:10 dilution in 0.01 M Tris pH 8.0, 0.7% NaCl) for 2 h at 37 °C. Subsequently, 1 × 10 3 HEK293 cells were treated with MNPs@ SiO 2 (RITC) in serum-free medium for 12 h and seeded onto the insert. Then, 10% FBS-containing medium was added to the lower chamber as a chemoattractant. Cells were incubated for 12 h at 37 °C to allow invasion. The upper side of the membrane was washed with a cotton swab, and invaded cells on the lower side of the membrane were fixed in Cytofix buffer (BD, San Jose, CA, USA) and stained with 10 μg/μL Hoechst 33342 for 15 min at 25 °C. Images were acquired using an Axio Vert 200 M fluorescence microscope (Zeiss, Jena, Germany). The number of invaded cells was counted using the ImageJ software (NIH).

Cell tracking
PDMS pillars were coated with fibronectin for 1 h before seeding of cells. Cells were seeded at a density of 3000 cells per sample and tracked every 5 min for 6 h on a JuLi ™ Br live cell movie analyser (NanoEntek, Inc., Seoul, Korea). Finally, cell movement speed was calculated by tracking cells using the ImageJ software (NIH).

Metabotranscriptomic data analysis
Differences in the gene expression of cells were examined using the Affymetrix system (ISTECH, South Korea) in conjunction with the Human U133 Plus 2.0 50 K microarray, which contains 54,675 probes. Differences in the data distribution were analysed using the GenPlex 3.0 software [21], with the probe signals being quantilenormalised. Amino acid and organic acid profiles were imported from previously reported data [21]. Biological pathways and functions were identified using the Ingenuity pathway analysis web-based bioinformatics software (IPA; Ingenuity Systems, Redwood City, CA, USA). Three-fold change in gene expression and 20% change in metabolites were used as cut-offs to generate datasets of significantly altered genes and metabolites.

Statistical analysis and error correction
Results were analysed by one-way analysis of variance (ANOVA) with Bonferroni's multiple-comparison test using the IBM-SPSS software (IBM Corp., Armonk, NY, USA). p < 0.05 was considered statistically significant. In the experiments using micropillars, errors in the pillar deflections were corrected by reducing the average pillar deflection of pillars outside the cell.