The toxicity of nanoparticles and their interaction with cells: an in vitro metabolomic perspective

Nowadays, nanomaterials (NMs) are widely present in daily life due to their significant benefits, as demonstrated by their application in many fields such as biomedicine, engineering, food, cosmetics, sensing, and energy. However, the increasing production of NMs multiplies the chances of their release into the surrounding environment, making human exposure to NMs inevitable. Currently, nanotoxicology is a crucial field, which focuses on studying the toxicity of NMs. The toxicity or effects of nanoparticles (NPs) on the environment and humans can be preliminary assessed in vitro using cell models. However, the conventional cytotoxicity assays, such as the MTT assay, have some drawbacks including the possibility of interference with the studied NPs. Therefore, it is necessary to employ more advanced techniques that provide high throughput analysis and avoid interferences. In this case, metabolomics is one of the most powerful bioanalytical strategies to assess the toxicity of different materials. By measuring the metabolic change upon the introduction of a stimulus, this technique can reveal the molecular information of the toxicity induced by NPs. This provides the opportunity to design novel and efficient nanodrugs and minimizes the risks of NPs used in industry and other fields. Initially, this review summarizes the ways that NPs and cells interact and the NP parameters that play a role in this interaction, and then the assessment of these interactions using conventional assays and the challenges encountered are discussed. Subsequently, in the main part, we introduce the recent studies employing metabolomics for the assessment of these interactions in vitro.


Introduction
Nanomaterials are dened as materials with at least one dimension smaller than 100 nm, 1 while nanotechnology is dened as the understanding and manipulation of matter at dimensions in the range of 1 to 100 nm, where unique phenomena enable novel applications. 2 Nanotechnology introduces many potential health, environmental, and industrial course of a biological process to be obtained, which is the omics technique of interest in this review. 40,41 The introduction of NPs in a cell line may cause a change in the levels of certain metabolites, which may give a clue on their effect on cells. During the past decade, many in vitro studies have used metabolomics to investigate the cytotoxicity of NPs on different cell lines.
In this review, the ways NPs and cells interact and the effects of the NP parameters on their interaction are discussed, followed by an overview of the cytotoxicity of different NPs in in vitro models, focusing on the use of metabolomics as a tool to identify the mechanisms and molecular information of their cytotoxicity.

Cellular uptake of NPs
The cytotoxic effects of NPs usually originate from their presence inside cells. 42 However, many applications of NMs in biomedicine require their entry in the cell to achieve their goal. Therefore, to further understand the cytotoxicity mechanisms of NPs on the cell and its metabolism, it is important to rst understand the cellular uptake mechanisms of NPs. This will also aid the design of environmentally safer NMs with enhanced cellular targeting and uptake properties for therapeutic purposes. 43 When immersed in a biological uid, NPs are exposed to a different medium than that employed for their synthesis. This will force the NPs to interact with the surrounding medium, which may alter their physical and chemical properties. 44 To stabilize themselves, NPs tend "to catch" the surrounding biomolecules (proteins, lipids, etc.) and form a biomolecular corona or protein corona (in case they are surrounded by proteins only), which may alter their identity. 45 NPs may be taken up by the cell in an energy-independent process, such as simple diffusion or translocation. However, most NP uptake pathways are energy dependent via endocytosis. Endocytosis is the formation of vesicles from the cell plasma membrane to take up substances such as particles, nutrients, and dead cells from the extracellular to the intracellular environment. 46 Endocytosis is described in two categories, i.e., phagocytosis and pinocytosis.

Phagocytosis
Phagocytosis is the cellular uptake of particulates (0.5-10 mm) in the plasma-membrane envelope. It is known as a host defence mechanism, engulng and internalizing cargos such as particles, dead cells, and cell debris. 43,47 This mechanism is a ligandinduced process, where NPs are engulfed by adsorbing opsonins, followed by their interaction with complement receptors on the cell surface (see Fig. 1). 48

Pinocytosis
Pinocytosis is the cellular uptake of extracellular uids and dissolved solutes. 49 It can be divided into macropinocytosis, clathrin-and caveolae-independent endocytosis, and receptormediated endocytosis. The latter is classied as clathrindependent endocytosis and caveolae-dependent endocytosis based on the proteins involved in the pathway. 50 2.2.2 Clathrin-mediated endocytosis. Clathrin-mediated endocytosis is the main process for the internalization of many NPs, which is used by all eukaryotic cells to internalize small particles and nutrients such as cholesterol. When the plasma membrane is rich in clathrin and ligand-receptor complexes start to form on the cell membrane surface, a cage of clathrin starts to form around the vesicle, resulting in vesicles with a diameter of 100-150 nm (see Fig. 2). 43,49,53 2.2.3 Caveolae-mediated endocytosis. Caveolae are bulbshaped invaginations in the plasma membrane, which are 50-80 nm in size. These vesicles are coated by caveolin and cavin and detached from the membrane by dynamin, which is a 100 kDa GTPase (see Fig. 2). 43,54,55 3. Role of physicochemical properties of NPs in cellular uptake and cytotoxicity It is important to consider the physicochemical properties (size, shape, surface functionalization, surface chemistry, chemical composition, concentration, etc.) of NPs in their design for biomedical or other applications. The interactions of NPs with the cell membrane and organelles can signicantly be altered at the bio-nano interface by these physicochemical properties, consequently changing the cellular uptake and nanotoxicity of the NPs. Therefore, before starting to assess the biological responses of NPs, thorough and proper characterization of the physicochemical properties of their core and surface should be performed. 56 In this part, we mainly focus on the effect of the size, shape, and surface chemistry of NPs on their cytotoxicity and cellular uptake (see Fig. 3). The effect of the NP core composition is not discussed here given that the surface characteristics are more important than the bulk characteristics in this context.

Size
The size of NPs plays an important role in both their cellular uptake and cytotoxicity. Thus, it is considered a key factor when designing NPs for biomedical application. Due to the fact that NPs possess a size between atoms and bulk materials, they lie on the critical transition zone between two different worlds. 57,58 It is worthy to mention that the original (primary) size of NPs differs from their hydrodynamic size in biological media. 59 This is mainly because of the formation of a biomolecular corona and the aggregation of the NPs. In this case, the aggregation of NPs can be prevented by manipulating the balance of attractive and repulsive forces. 60 For instance, Fe 3 O 4 NPs can be stabilized with citrate, preventing their aggregation due to electrostatic repulsion. 61 However, due to the formation of a biomolecular corona and the different ionic strengths of biological solutions compared to water, NPs may have new surface identity. Wei et al. 38 performed a cytotoxicity study on the different sizes of TiO 2 (5 and 200 nm) and Al 2 O 3 (10 and 50 nm) NPs and observed the formation of aggregates in solution form when the NPs were suspended in cell medium without serum, where the sizes of all the NPs became 8-388-fold larger than their original sizes due to the higher ionic strength of the medium compared to water. Upon the addition of serum, the hydrodynamic sizes of the NPs decreased to only 1.6-10 folds larger than their original sizes. This is because the formation of the protein corona around the NPs prevented them from aggregating due to steric repulsion. The authors found that the smaller NPs (in terms of primary size, rather than hydrodynamic size) for both TiO 2 and Al 2 O 3 had higher cytotoxicity and much greater decrease in cell metabolic activity.
When studying the NP-cell membrane interaction mechanism dependence on the size of NPs, it was found that it has a strong inuence. Specically, large NPs (>60 nm) may cause steric hindrance, which prevents their interaction with the cell membrane. 62 Conversely, NPs smaller than the cut-off size of receptor diffusion (<30 nm) may not recruit enough cell membrane receptors in the interaction region to overcome the elastic recoil force, preventing membrane wrapping from occurring. 63 Moreover, the membrane receptors are known to form clusters that are 10-50 nm in size. Thus, a 50 nm NP, for example, needs to interact with only one receptor cluster, while a 500 nm NP must interact with several clusters simultaneously. This makes the internalization of the 50 nm NP energetically more favourable than the 500 nm NP. 44 In general, smaller-sized NPs have been reported to have higher cellular uptake and higher cytotoxicity. For instance, Dong et al. 64 reviewed 76 carefully chosen literature reports that included in vitro studies of the size-dependent cytotoxicity of amorphous silica NPs (aSiO 2 NPs) and found that 76% of these papers showed that smaller-sized aSiO 2 NPs exhibited greater cytotoxicity. However, it is important to consider that the cell type plays a role in this process given that it depends on the predominant pathway of cellular uptake in each different cell. 65,66 For some NPs, the higher the cellular uptake of NPs, the greater their cytotoxicity. 67 Nonetheless, there are some exceptions, where the cytotoxicity of NPs is independent of their cellular uptake. In these cases, the cytotoxicity is induced by sources other than amount of toxicant, including the NP high surface area, instability, and ion release. Gliga et al. 68 found that 10 nm silver NPs (AgNPs) are more toxic to the human lung BEAS-2B cell line than other NPs with higher uptake ratios due to the release of more Ag + .

Shape
The shape of NPs can be controlled by manipulating the experimental conditions during their synthesis, such as supersaturation, reducing agents, temperature, surfactants, and secondary nucleation. 69 There are many different shapes and geometries of NPs, such as spherical, rod, ower, star, disc, cubic, prismatic, and needle-like structures. The aspect ratio (AR), which is the proportion between width and height of NPs, is used to compare different shapes of NPs. For example, spherical AuNPs have an AR of 1, while Au nanorods (AuNRs) have a higher AR.
It was proven that the cellular uptake and cytotoxicity of NPs are affected by the AR of NPs. Given that AuNPs are common in many biomedical applications, many studies investigated their shape-dependent cellular uptake and cytotoxicity. For instance, Woźniak et al. 70 compared the in vitro cytotoxicity proles of different shapes and sizes of bare (non-coated) AuNPs in cancer (HeLa) and normal (HEK293T) cell lines. They found that Au nanospheres (AuNS) and AuNRs had higher cytotoxicity than star-, owerand prism-shaped AuNPs. However, the sizes of these different AuNPs shapes also differed. Specically, the AuNSs and AuNRs had smaller sizes (10 nm and 38 × 16 nm, respectively), while the ower-, prism-, and star-shaped AuNPs had larger sizes (∼370 nm, ∼160 nm, and ∼240 nm, respectively). Thus, their sizes may also play a crucial role in this cytotoxicity tendency, given that smaller NPs are known to have higher cellular uptake and aggregation rate inside the cell, which explains the observed cytotoxicity.

Surface charge
NPs can have negative, positive, or neutral surface charge depending on their surface functional groups. 71 The surface charge can affect the NP-cell membrane interactions, protein corona, and consequently the cellular uptake of NPs. 72 Therefore, it is one of the most important physicochemical properties to control when designing NPs for biomedical applications. Generally, reports have shown that charged NPs have higher cellular uptake than neutral NPs. 63 The cell membrane is negatively charged due to the anionic head group of phospholipids and the existence of some carbohydrates, such as sialic acid. 73 Considering this, cationic NPs, in most nonphagocytic cells, are taken up by the cells to a greater extent than anionic NPs. However, in some cases, anionic NPs have greater cellular uptake in phagocytic cells. 74,75 The surface charge of NPs can also tune their cellular uptake pathway. For instance, Untener et al. 76 reported that positively charged AuNRs had a higher extent of internalization compared to their negatively charged counterparts. It was found that cationic AuNRs were taken up through macropinocytosis and clathrin-mediated endocytosis, while anionic AuNRs were internalized through macropinocytosis and caveolae-related mechanisms.
The cytotoxicity of NPs is also, as expected, affected by their surface charge. Similar to the dependence of the cellular uptake of NPs on their surface charge, in nonphagocytic cells, charged NPs were found to be more cytotoxic than their neutral counterparts, with the positively charged NPs, in most cases, being more cytotoxic than negatively charged NPs. 74 Moreover, the surface charge of NPs does not only affect their cytotoxicity level but also their mechanisms. A study by Schaeublin et al. 77 showed that although both charged and neutral AuNPs were taken up in similar amounts and caused cell morphology disruption and decreased cell viability through ROS generation in a human keratinocyte cell line (HaCaT) model, only charged NPs caused signicant mitochondrial stress. This suggested that the surface charge of AuNPs can affect the mechanism of cell death. Further investigations on mitochondrial-mediated toxicity revealed that neutral AuNPs did not affect the mitochondrial outer membrane potential, which has a slight negative charge, and thus apoptosis was not initiated, and the authors suggested that necrosis may be the cell death mechanism in this case. However, charged AuNPs affected this membrane in different ways. On the one hand, cationic AuNPs accumulated on the mitochondrial outer membrane due to its slight negative charge, which eventually damaged the membrane and caused the release of apoptotic proteins such as caspase-3 inducing mitochondrial-mediated apoptosis. On the other hand, anionic AuNPs increased this slight negative charge on the outer membrane, which forced the mitochondria, trying to adjust this potential disruption, to release the positively charged calcium ions into the cytosol, inducing calcium-evoked apoptosis (see Fig. 4).

Hydrophobicity
It has been shown that the hydrophobicity of NPs can affect the protein binding, cellular uptake, and cytotoxicity of NPs. [78][79][80][81][82] The hydrophobicity and hydrophilicity of NPs can originate from the core or the functionalities of the NPs. In a recent systematic simulation study, Li et al. 78 showed that changing the spikes of virus-like NPs (VLP) signicantly altered the cellular uptake efficiency, while the effect of the core hydrophobicity of VLP was secondary. This study reported that VLP with hydrophobic or amphiphilic spikes were internalized more efficiently than that with hydrophilic spikes.
Generally, when keeping the other properties of NPs such as surface charge constant, their hydrophobicity has a positive trend with their cytotoxicity. 74 Muthukumarasamyvel et al. 81 controlled the hydrophobicity of dicationic amphiphilestabilized AuNPs by conjugating the dicationic functionality with different numbers and locations of H and OH groups. The authors observed increasing anticancer or cytotoxicity properties with an increase in the surface hydrophobicity of the NPs against A549 lung cancer cells.  77 The neutrally charged NP (*) does not disrupt the mitochondrial membrane potential, and therefore apoptosis is not activated. The positively charged NP (+) disrupts the slight negative charge on the cytosolic side of the outer membrane, leading to a disruption in the mitochondrial membrane potential. The disruption damages the membrane and proteins, such as caspase activators, leak into the cytosol. The negatively charged NP (−) increases the negative charge on the outer membrane, which leads to a disruption in the mitochondrial membrane potential. The mitochondria compensate by releasing calcium ions that were stored in the matrix of the mitochondria. The spike in calcium induced apoptosis. 77 Table 1 Recent studies highlighting the influence of the physicochemical properties of NPs on their cellular uptake and cytotoxicity

Surface functionalization
Changing the ligands on the surface of NPs will mostly tune the previous parameters (surface charge and hydrophobicity), which affects the protein corona, cellular uptake, and cytotoxicity of the NPs. 77,83,84 However, the specic functionalities on the surface of NPs can be useful for targeting purposes. Here, overexpressed or unique receptors on the cell membrane are targeted by functionalizing the NPs with a complementary aptamer, protein, or antibody, which can specically bind to the cell receptors. Tao et al. 85 targeted cervical cancer cells through folic acid (FA)-poly(ethylene glycol)-b-poly(lactide-co-glycolide) blended NPs, which enhanced the efficacy of cancer chemotherapy through the targeted-delivery of anticancer drugs. Lund et al. 86 showed that AuNPs functionalized with 50% PEG-NH 2 /50% glucose had an eighteen-fold higher internalization rate than NPs functionalized with either PEG-NH 2 or glucose alone due to their different organization patterns. Alternatively, Yeh et al. 87 studied the role of ligand coordination of two quantum dots (QDs) on their cytotoxicity. The authors found that monothiol-functionalized QDs had greater levels of cytotoxicity compared to dithiol-functionalized QDs in HeLa cell lines. However, the monothiol-functionalized QDs had a higher charge density, and thus it is difficult to tell if this tendency is solely related to the ligand coordination or charge density.
Studying the dependency of cellular uptake and cytotoxicity on a certain physicochemical property of NPs can be very complex. For instance, changing their surface charge may lead to a change in hydrophobicity, hydrodynamic size, and protein corona. Furthermore, this may be done by changing the functionalities and coating of the NPs. [73][74][75][76][77] Table 1 summarizes some recent studies exemplifying the effect of the physicochemical properties of NPs on their cellular uptake and cytotoxicity.

Cytotoxicity assessment
In vitro cytotoxicity of NPs is assessed using cell models. Although this assessment does not replace the in vivo evaluation of their cytotoxicity, it represents a screening bridge between the investigation of the quality and in vivo application of materials. 56,93 Herein, we focus on the in vitro assessment of nanotoxicity. In the case of in vivo assessment, readers are encouraged to read the wholistic review by Kumar et al. 94 Many in vitro assays are used to investigate or measure the cytotoxicity of NPs. These assays can be categorized to ve main categories including cell viability and proliferation, ROS generation, cell stress, cell morphology phenotyping, and cell-NP uptake assays. 56 Fig. 5 demonstrates some pathways of the effect of NPs on cells.
Many types of interference between NPs and cell viability assays have been reported. One way is the adsorption of the mitochondrial activity-related proteins on the NP surfaces. This may lead to the enzyme denaturation, giving false results of the cell viability proles. 97 For instance, Stueker et al. 98 used molecular dynamics simulation to investigate the effect of LDH enzyme binding on functionalized AuNPs. The authors observed that the dynamics of the side chains of the enzyme were largely constrained in all four active sites. Another way of interference is that the light absorbance spectra of the NPs can interfere with the absorption window of the assay, leading to false colorimetric measurements. 99 For example, Díaz et al. 100 reported that ve NPs (magnetic iron/graphite, magnetite/silica, bare and poly(ethylene glycol)(PEG)-ylated silica, and magnetite/FAU zeolite) in culture medium aer 72 h (in the absence of cells) showed absorbance at the same wavelength (525 nm) used in the MTT assay. This absorbance increases with the NP concentration, depending on their type. The third way of interference is that NPs may interact with the assay reagents. For instance, Hoshino et al. 101 reported that cysteamine-coated quantum dots catalytically reduced MTT to formazan without cellular metabolism taking place (see Fig. 6).

ROS generation and oxidative stress
Reactive oxygen species (ROS) are a type of unstable molecule (free radicals) that contain oxygen and can easily react with the other molecules in cells. The ROS include the superoxide anion (O 2 c − ), hydrogen peroxide (H 2 O 2 ), and hydroxyl radical (HOc). ROS are normally produced by cells at certain levels to maintain regular metabolism and homeostasis, which are considered as critical signalling molecules in cell proliferation and survival. 99,102 However, they may be produced through interactions with exogenous sources such as NPs. If this event produces excessive ROS that the cellular antioxidant defense system (enzymatic antioxidants such as glutathione (GSH) peroxidases) cannot handle, oxidative stress is triggered. 102,103 This may lead to the destruction of organelles and biomolecules, including triggering membrane damage, lipid peroxidation, DNA damage, protein damage, apoptosis, necrosis, and inammatory response, leading to many diseases such as cancer, diabetes, neurodegenerative, and cornea diseases. 103,104 NPs can generate ROS by acting as a catalyst in ROS generation reactions. For instance, Higashi et al. 105 reported the catalytic generation of ROS by AuNPs and showed that this reaction can be controlled by changing conditions such as the type, concentration, and pH of the NP solution.
ROS detection can be performed by the direct measurement of ROS levels or the measurement of their oxidative damage or other outcomes. 106 Some direct methods for the detection of ROS are uorescein-compound-based tests and electron paramagnetic resonance (EPR). The reactive uorescein probes 2 ′ ,7 ′ -diuorescein-diacetate (DCFH-DA) and dichlorodihydro-uorescein diacetate (H 2 DCFDA) are non-uorescent; however, when they are exposed to the cell cytosol enzymes, they get hydrolysed. Then, the cellular ROS oxidize them into a highly uorescent compound, dichlorouorescein (DCF), yielding an optical ROS concentration-dependent response, which can be measured using uorescence microscopy or ow cytometry. 107 Alternatively, indirect approaches for the detection of ROS include many assays that depend on the stimulated oxidative effect of the ROS. One approach is by measuring the enzymatic or non-enzymatic antioxidants levels. 106 Oxidative stress can also be assessed by measuring the oxidative damage of the cell biomolecules. These damaged biomolecules include proteins, lipids, and DNA and can be detected by measuring the protein carbonyl content, 108,109 malondialdehyde levels, 110,111 and 8-oxo-2 ′ -deoxyguanosine (8-OdG) lesion, [112][113][114] respectively. Other genotoxicity assays include the comet, Ames, micronucleus, and chromosome aberration assays. 115 During the course of measuring NP-induced ROS generation and oxidative stress, NP-assay interferences may occur. 116,117 In colorimetric-and uorimetric-dependent assays, NPs may interact with the nal form of the dyes in a way that alters, by enhancing or reducing, the absorbance or uorescence of the dye. For example, Aranda et al. 116 observed the quenching effect of several NPs on the dye uorescence emission in the DCFH-DA assay, which was correlated with the cellular uptake of the NPs. The authors suggested a threshold concentration of NPs at which their oxidative effect can be detected, and they proposed that changing the experimental conditions can reduce this interference. Conversely, Pfaller et al. 117 reported the dye uorescence enhancement of the DCFH-DA assay in the presence of Au or Fe 2 O 3 NPs. This conrms that both scenarios (quenching and enhancement) may occur due to NP-probe interactions during colorimetric-and uorimetric assays.

Inammatory response
The inammatory response induced by NPs in a cell line can be measured by detecting the produced inammatory biomarkers. Macrophages and other cells release many cytokines, which play a crucial role in cell communication in the immune system by, for instance, promoting inammation. Interleukins (ILs), such as IL-1b, IL-6, IL-8, and IL-10, in addition to other cytokines, such as tumor necrosis factor TNF-a and granulocytemacrophage colony-stimulating factor (GM-CSF), play a central role in inammation regulation. The expression of these biomarkers can be assayed to determine the inammatory response caused by NPs. ELISA (enzyme-linked immunosorbent assay) or western blotting, and electrophoretic mobility shi assays (EMSAs) or real-time polymerase chain reaction (RT-PCR) systems are used for the measurement of cytokines and the related genetic expressions, respectively. [118][119][120][121][122][123] NPs were reported to induce an inammatory response in different cell lines. Many studies used conventional assays to measure this response. [119][120][121] However, these assays can also interfere with NPs during the measurement of inammatory response in cell lines. Some inammatory cytokines were reported to be adsorbed on the NP surface, causing interference ( Fig. 7a). Guadagnini et al. 122 investigated the interferences of different NPs with some in vitro cytotoxicity assays. The authors reported that Fe 2 O 3 , TiO 2 , and SiO 2 NPs signicantly adsorbed IL-6, IL-8, and GM-CSF cytokines on their surfaces at different levels at a NP concentration of 75 mg cm −2 . Fig. 7b shows that all the studied NPs adsorbed the IL-8 cytokine except PLGA-PEO NPs, which surprisingly increased the apparent level of cytokines, probably due to the stabilization of the peptides and their protection from proteolysis. In the case of other NPs, the level of adsorption depends on the NPs and the cytokine studied. OC-Fe 3 O 4 NPs are the most cytokine-adsorbing NPs tested given that cytokines could not be detected in the supernatants. Furthermore, Piret et al. 123 observed a high inter-laboratory variability for the ELISA assay for IL1-b and TNF-a measurements and they suggested that testing of NP-cytotoxicity assay interferences should be always performed. Readers should kindly refer to ref. 122 for more information about the interference between different assay and NPs and some solutions to this problem.

Apoptosis and necrosis
Apoptosis is a programmed cell death pattern, 124 while necrosis is an unprogrammed cell death. 125 Both patterns of cell death can be an outcome of NP treatment. 126 129 and Ag 130 NPs induced apoptosis in A549, BEAS-2B, ECV304, C6, and HepG2 cell lines, respectively. Alternatively, Reus et al. 131 reported dosedependent cell necrosis induced by SiO 2 NPs in BALB/c 3T3 cell line. Apoptotic cell death is mostly non-inammatory, while necrotic cell death can be inammatory. 132 Both pathways are extremes, and many cases are a complex combination of both. For instance, Kumar et al. 133 observed that AgNPs caused cell death in L-929 broblast cell lines in association with both necrosis and apoptosis. The cell death pathway is controlled by many parameters such as the surface charge, concentration, and exposure time of NPs. Schaeublin et al. 77 reported that charged AuNPs caused cell death through apoptosis, while neutral AuNPs caused it through necrosis (see Fig. 4).
Many assays are used to detect apoptosis and necrosis. Phosphatidylserine (PS) migration to the extracellular side of the cell membrane and caspase activation into initiator and effector enzymes are two events that accompany apoptosis and can be used as markers to detect it. Externalized PS on the surface of the cell can be detected using uorescein isothiocyanate (FITC)-labelled Annexin-V. Annexin-V specically binds to the exposed PS on the cell surface in the early apoptotic cells, and then can be measured via ow cytometry or uorescent microscopy. Alternatively, the membrane-impermeable propidium iodide (PI) dye exclusion assay is used for the iden-tication of cellular necrosis. PI binds to DNA in the nucleus and stains it only when the cell membrane integrity is lost (which is an event that accompanies necrosis). Thus, a combination of the above-mentioned assays can determine the pattern of cell death. 134 For instance, Vafaei et al. 135 used the Annexin V-FITC/PI staining kit to study the apoptotic efficacy of zinc-phosphate NPs (ZnPNPs) against the MCF-7 breast cancer cell line. The untreated cells with NPs showed a live cell (Annexin V-FITC−/PI−) percentage of 98.6%. Conversely, aer exposure to ZnPNPs, the apoptotic cell (Annexin V-FITC+/PI−) ratio increased from 0.190% to 44.8% and the necrotic cell (Annexin V-FITC+/PI+) percentage increased to 1.34% (see Fig. 8).
Flow cytometry-based assays have negligible NP interferences. 133 Bancos et al. 136 reported that SiO 2 NPs have low or no interference with ow cytometry assays. However, other colorimetric and uorimetric-based assays face the same problems mentioned in the previous sections.

Metabolomics for the cytotoxicity assessment of NPs
In general, most studies on the cytotoxicity of NPs use the conventional (phenotypic) assays. However, many of these assays, as mentioned before, have been reported to interfere with the NPs because of their color, uorescence, chemical activity, light scattering, etc. Thus, to precisely reveal the cytotoxicity of NPs, it is necessary to use a combination of more than two assays. This involves testing for NP interferences and eliminating them by changing experimental conditions or comparing the results of two similar tests, which is a complex and time-consuming process. However, many reports only used one or two cytotoxicity assays and ignored any potential interference with NPs. 137 In addition, even though the conventional cytotoxicity assays can reveal that a certain cytotoxicity outcome happened, these assays are limited in terms of detecting the molecular information that caused this event.
The current toxicological assays need to be updated and new tools should be incorporated progressively in this eld. 138 A more advanced and emerging approach to study the toxicity of particles is the "omics" technique, which is based on the change in epigenome, transcriptome, proteome, genome, lipidome, and metabolome proles introduced by internal or external stimuli. In increasing number of studies are using this approach to investigate the in vitro and in vivo toxicity induced by NPs. The determination of new targets and biomarkers for NP toxicity is one of the strengths of the omics technique. Moreover, the omics technique has high sensitivity, which is useful because of the low levels of environmental exposure to NPs that sometimes cannot be detected using the conventional assays. 39 Another strength is that unlike the conventional assays, the omics technique has low or no interferences with NPs. 39,122 In the eld of toxicology, the most related omics discipline is metabolomics. 139 Metabolomics, one of the newest in the omics era, is an emerging eld, which is broadly dened as the comprehensive measurement of all metabolites and lowmolecular-weight molecules in a biological specimen (tissues, cells, uids, or organisms), 40 and is one of the most powerful bioanalytical strategies that allow a picture of the changes of metabolites levels of an organism to be obtained during the course of a biological process either as a footprint (analysis of extracellular metabolites) or ngerprint (analysis of intracellular metabolites). 41 The detailed analysis of low molecular weight compounds provided by nuclear magnetic resonance (NMR) spectroscopy or mass spectrometry (MS), besides the analysis performed by the powerful chemometric soware (MetaboAnalyst), 140 provides an accurate and quick detection and comparison of many types of chemical entities including carbohydrates, amino acids, nucleotides, lipids, steroids, fatty acids, and their derivatives, which are produced by cell metabolism. 141 Currently, metabolomics is applied in many elds such as disease ngerprinting, biotechnology, environmental and plant research, toxicology and safety research, clinical medicine, and pharmacology. 139,142,143 Our group has been investigating the metabolic changes in serum, urine, and feces induced by different diseases such as lung cancer and diabetes, or other stimuli such as kidney transplant. [144][145][146] Due to the non-invasive sampling in the metabolomic approach, the relatively low number of metabolites (compared to transcripts and proteins), and good level of knowledge about the role of most metabolites, metabolomics provides a well-grounded and precise methodology to investigate the biochemical effects and toxicity of NPs, 139,147 and it can present insight into the genotype and phenotype changes with a biological response. 148 Moreover, single-cell metabolomics is achievable today, making it possible to determine phenotypic heterogeneity among individual cells. 149 Many cellular activities such as intercellular signal transduction, energy transfer, cell proliferation, and differentiation occur at the metabolite molecular level and are regulated by the presence and level of specic metabolites. Furthermore, metabolites are the end result of the expression of functional genome, transcriptome, and proteome (see Fig. 9). 150,151 This indicates that metabolomics can detect many NP cytotoxicity outcomes and reveal the molecular information behind these events even at low levels of NP exposure and with no interferences. Therefore, it is a great tool in nanotoxicology, which is being applied to reveal the effect and toxicity of NPs in many elds including environmental and agricultural elds 152-154 and cancer research. 155 Metabolomics can help in better understanding of the transition from in vitro to in vivo systems of NP Fig. 9 Overview of the connection of the main omics-sciences: genomics, transcriptomics, proteomics, and metabolomics. Metabolomics represents the final output of cellular processes. toxicity and its effect given that it is applied in both types of experiment. 156,157 Furthermore, metabolomics can be combined with other omics techniques to provide a more comprehensive understanding of the effects of NPs on cells. [158][159][160] When comparing NMR-and MS-based metabolomics, generally, NMR has lower sensitivity than MS, and thus it is considered more suitable to analyze extracellular metabolites (exometabolome), which is done by the analysis of the cell culture media. Alternatively, the more sensitive MS techniques are more suitable for the analysis of relatively low levels of intracellular metabolites (endometabolome), especially when isolated from a limited number of cells. However, both analysis techniques are complementary and should be used simultaneously to maximize the metabolic window.
This emerging technique has not yet been widely applied for the investigation of NP cytotoxicity in in vitro systems and more research needs to be done on different NPs and cell lines. In this section, we focus on the metabolic changes induced by different NPs in different cells in vitro. The workow of a metabolomics experiment is demonstrated in Fig. 10. This review does not go into detail on the workow of metabolomics. In this case, for a detailed demonstration of how metabolomic workows generate data, the reader is directed to read the following reviews and book chapters. 39,160-166

AuNPs
Gold nanoparticles (AuNPs) are very common in the biomedical eld. AuNPs have many unique properties such as ease of synthesis, tunable size, ease of surface modication, surface plasmon resonance (SPR), and X-ray attenuation. 167 This makes them the center of attention in many applications, including the growing eld of nanomedicine, biosensors, targeted drug delivery, radiation therapy, photothermal therapy, biomedical imaging, and cancer diagnostics and therapeutics. 155 Metabolomics is used in several studies to assess the cytotoxicity of AuNPs and reveal their molecular information. Au nanorods (AuNRs) are one example of AuNPs that have strong absorption in the near-infrared spectral region and can be used in tumor thermal therapy (hyperthermia), and also in targeted tumor therapy. Wang et al. 168 observed, using conventional assays, that AuNRs have a unique inuence on cell viability by causing the death of cancer cells (A549 cell line), while having negligible effect on normal cells (16HBE and MSC cell lines). The authors showed that AuNRs were released from the lysosome of cancer cells, and then translocated into the mitochondria, causing oxidative stress by the production of ROS. Alternatively, the normal cells had more intact lysosomes, and thus the AuNRs were not released in the cell cytoplasm. However, the molecular information during this cellular translocation was unclear. Later, the same group, 169 used a metabonomic approach, a subset of metabolomics, 170 by applying 1 H NMR and multivariate data analysis, to study the metabolic change with time during the exposure of A595 and 16HBE cell lines to AuNRs. The authors found that both cell lines had intracellular disruption by the reduction of lactate levels and by causing oxidative stress. However, the normal cells resisted this oxidative stress by de novo GSH synthesis, unlike the cancer cells, which did not trigger this pathway, causing severe damage of their mitochondria (see Fig. 11). The metabonomic study further indicated the downregulation of nucleosides and nucleotides in the cancer cells, indicating cell death. Alternatively, the amino acid levels were upregulated in the normal cells, indicating cell stress. This study shows the usefulness of metabolomics in revealing the molecular information of the effect of NPs on cells, aer conventional assays played the role of a general scanner for these effects.
Metabolomics can help in identifying biomarkers for NP cytotoxicity. For example, Xu et al. 171 investigated the potential harmful effects of AuNRs on male reproduction by studying the metabolic change in spermatocyte-derived cells (GC-2) and Sertoli (TM-4) cell line aer exposure to 10 nM of AuNRs. Employing metabolomics, the authors observed a strong downregulation in glycine levels in TM-4 cells, while there was no signicant change in GC-2 cells. To identify what may accompany this reduction of glycine (potential biomarker), high content screening (HCS) and JC staining were used, and it was found that AuNRs decreased the membrane permeability and mitochondrial membrane potential of TM-4 cells. Moreover, the authors observed a disruption in the mRNA and protein levels of blood-testis barrier (BTB) factors using RT-PCR and western blot. Then, to conrm that glycine is a biomarker for these events, the authors repeated the experiments aer adding glycine to the medium and noticed that the cells recovered from the previous harmful effects. This experiment reveals that glycine can be recognized as a biomarker to the changes in membrane permeability, mitochondrial membrane potential, and blood-testis barrier (BTB) factors in further similar experiments.
Huang et al. 172 observed that spherical AuNPs (20 nm) were not cytotoxic against the human dermal broblast (HDF) cell line. The authors combined bioinformatics with metabolomics to determine the molecular information of this toxicity resistance. Firstly, they detected that 29, 30 and 27 metabolites were differentially expressed in HDFs aer 4, 8, and 24 h treatment with AuNPs, respectively. Among them, only six metabolites were determined to be key metabolites using bioinformatics techniques including expression pattern analysis and metabolic pathway analysis using MetaboAnalyst online tool. The key  metabolic pathway was identied to be the GSH pathway with GSH as the key metabolite. Subsequently, these results were veried and it was found that the increase in GSH levels aer AuNP treatment may be the reason behind the toxicity resistance behaviour of the cells, given that GSH can trigger an oxidative stress protection mechanism that helps in avoiding cytotoxicity. 169 This reveals that GSH can be considered as a biomarker for oxidative stress resistance. Lindeque et al. 173 used MS metabolomics to study the effect of citrate-, poly-(sodium styrene sulfonate)-, and polyvinylpyrrolidone (PVP)-capped AuNPs on the intracellular metabolites of HepG2 cells. Surprisingly, aer 3 h of treatment, a holistic depletion of intracellular metabolites was observed for all the capped AuNPs. Usually, metabolic changes result in the upregulation of the metabolite levels because of secondary pathways, clearance issues, and reduced enzyme functionality. 174 Firstly, the authors suggested that a loss of cell membrane integrity happened, but the exometabolomic data, measured using the NMR technique, was not consistent with this reasoning. Subsequently, they hypothesized that the AuNPs bind to the intracellular metabolites with or without replacing the surface coatings.
Gioria et al. 175 combined proteomics and metabolomics to gain a further understanding of the effects of two sizes, i.e., 5 and 30 nm, of AuNPs on the human colon adenocarcinoma Caco-2 cell line. The proteome and metabolome are directly interconnected and inuence each other given that the protein levels can change the metabolic prole of a cell system and vice versa. Genomics and transcriptomics were excluded from this study due to their restricted value given that they provide limited information about phenotyping. The authors used liquid chromatography high-resolution tandem mass spectrometry (LC-HRMS/MS) and two-dimensional gel electrophoresis (2DE) to obtain qualitative and quantitative data of deregulated metabolites and proteins, respectively. Subsequently, the data was combined and interpreted using systems biology analysis. Aer 72 h of exposure to AuNPs, 61 proteins and 35 metabolites in the cell extract were identied to be up-/ down-regulated. The internalization mechanism was found to be endocytosis due to the downregulation of the SH3GL1 and EAA1 proteins, which are involved in the endocytic pathway. The smaller-sized AuNPs caused a greater number of deregulated proteins and metabolites due to their higher internalization in the cells. Concerning metabolomics, the metabolite propionylcarnitine (C-3 carnitine) and glycine levels increased upon exposure to AuNPs, which indicates apoptosis. This study further reported the accumulation of GSH in both 5 and 30 nm AuNP-treated cells, which indicates that an antioxidative mechanism occurred as a self-defense system against oxidative stress. These results were conrmed using uorescence microscopy analysis, where the over-expression of Annexin-V and nuclear fragmentation induced by AuNPs were evident, emphasizing that apoptosis occurred.
Omics technology together with complementary methods not only offer a promising tool in nanotoxicology to understand the molecular mechanisms of NP toxicity, but they also enhance the development and design of nano-drugs. For instance, Ali et al. 176 combined MS-based metabolomics and proteomics results through network analysis to better understand the molecular mechanism of AuNR photo-thermal therapy in the human oral squamous cell carcinoma (HSC-3) cell line. The results showed an upregulation in phenylalanine, which is considered an outcome of apoptosis pathways, indicating the good photo-thermal therapy efficiency of the AuNRs. Table 2 summarizes the studies that used the metabolomics technique to assess the effect of AuNPs in vitro on different cell lines.

AgNPs
Silver nanoparticles (AgNPs) have various interesting biological properties and are known for their well-reported antibacterial activity. 183 They have a wide range of applications including cosmetics, textiles, and biomedical products. Also, their therapeutic application as antiviral and anticancer drugs is expected to be further expanded. 184,185 Regarding the use of AgNPs as potential drug carriers for cancer therapy from proteogenomic and metabolomic perspectives, the reader is directed to the review by Raja et al. 186 AgNPs have been shown to inuence different cells causing apoptosis, lipid peroxidation, and DNA damage. [187][188][189][190] One of the advantages of metabolomics is that it is capable of detecting early biochemical events and metabolic changes even during the absence of a signicant cytotoxic response by conventional assays. Carrola et al. 191 studied the effect of citratestabilized 30 nm AgNPs on the human epidermis keratinocyte (HaCaT) cell line aer 48 h of exposure at two concentrations, i.e., 40 mg mL −1 (close to IC 50 = 38.7 ± 2.5 mg mL −1 ) and 10 mg mL −1 (no signicant cell viability loss). Using NMR-based metabolomics, the authors observed that most metabolic changes happened at the lower concentration, which allowed the detection of early biochemical events, including upregulated GSH-based antioxidant protection, downregulated tricarboxylic acid (TCA) cycle activity, energy depletion, and cell membrane modication. In a similar study, 192 NMR metabolomics was used to assess the metabolic effects of two types of coated AgNPs towards the human hepatoma (HepG2) cell line and signicant metabolome changes were observed at a subtoxic concentration of AgNPs. These changes include energy production, antioxidant defence system, protein degradation, and lipid metabolism pathways, suggesting that the cells have metabolism-mediated protective mechanisms against AgNPs. In the third study by this group, 193 they investigated the effect of size and surface chemistry of AgNPs on the metabolic change caused in the HaCaT cell line. The authors used citrate-coated AgNPs with a diameter of 10, 30, and 60 nm, and 30 nm AgNPs coated with citrate, polyethylene glycol (PEG), or bovine serum albumin (BSA). It was found that the largest NPs and the PEG-coated NPs had the least impact on cell metabolism and viability, which is the expected tendency, as mentioned before in Section 3. Furthermore, Carrola et al. 194 used NMR metabolomics to characterize the responses of RAW 264.7 macrophages to subtoxic concentrations of AgNPs (30 nm) and ionic silver (Ag + ). They observed that the exposure to AgNPs caused a downregulation in intracellular glucose utilization, possibly due to the reprogramming of the TCA cycle towards anaplerotic fuelling and production of anti-inammatory metabolites. Also, an upregulation in the synthesis of GSH was observed, enabling the cells to control the ROS levels. In contrast, macrophages exposed to Ag + at equivalent subtoxic concentrations showed reduced metabolic activity, lower ability to counterbalance ROS generation, and alterations in membrane lipids. This indicates that the ionic form of silver has a greater effect on the cells and is one of the sources of AgNP cytotoxicity.
Huang et al. 172 compared the effect of AgNPs and AuNPs, and showed that while AuNPs had no cytotoxicity, AgNPs induced grade 1 cytotoxicity aer HDF cells were exposed to them for 72 h. Using metabolomics, the citrate cycle pathway was determined to be the key metabolic pathways in the AgNPtreated cells with malic acid as the key metabolite. Thus, the mechanism of AgNP cytotoxicity is by the upregulation of citric acid content, which indicated the inhibition of malic acid synthesis, inuencing the production of ATP (mitochondrial dysfunction) and inhibiting cell proliferation, leading to cytotoxicity (see Fig. 12). Conversely, AuNPs were not cytotoxic due to the triggering of the antioxidant defence system by the upregulation of GSH. Kim et al. 195 used high-resolution magic angle spinning (HR-MAS) NMR-based metabolomics to study the cytotoxicity of AgNPs against human Chang liver cells. The authors observed the depletion of GSH, lactate, taurine, and glycine levels, while most amino acids, choline analogues, and pyruvate were upregulated by the AgNPs. It is probable that the downregulation of GSH induced the conversion of lactate and taurine to pyruvate.
The effect of AgNPs was also studied on non-mammalian cells such as yeast and unicellular alga. Babele et al. 196 studied the effect of 1.0 mg L −1 of 50-100 nm-sized AgNPs, prepared using aqueous gooseberry extract, on yeast Saccharomyces cerevisiae cells. Untargeted 1 H NMR-based metabolomics revealed a several-fold increase or decrease in the levels of 55 different metabolites, including the ones involved in amino acid metabolism, glycolysis, and tricarboxylic acid (TCA) cycle, organic acids, nucleotide metabolism, urea cycle, and lipids metabolism. The authors noticed a reduced level of GSH, which indicates that oxidative stress occurred, leading to the strong cytotoxicity of AgNPs to the yeast cells. Qu et al. 197 investigated the effect of AgNPs on the performance of Chlorella vulgaris F1068 unicellular green alga in phosphorus assimilation (phosphorus removal by algae-based biotechnology). Using MS-based metabolomics, the authors observed the inhibition of algal assimilation. AgNPs disturbed the metabolic responses related to the phosphorus assimilation by reducing the levels of guanine, glutamine, alanine, and aspartic acid and increasing the levels of succinic acid. The NPs also inhibited phospholipid metabolism by the Fig. 12 Comparison of the metabolic changes induced due to the interactions between AuNPs or AgNPs with HDFs cells. While AgNPs (Right) induced cytotoxicity in the HDF cells, the effect of AuNPs (Left) was suppressed by an antioxidant mechanism. 172 Table 3 Summary  Review Nanoscale Advances downregulation of glycerol-3-phosphate and myo-inositol and upregulation of serine. Furthermore, GSH metabolism was affected by the NPs, which induced oxidative stress in the alga cells (upregulation of glycine). Cao et al. 198 showed that the effect of AgNPs on Chlorella pyrenoidosa can be altered by the number of repeated exposures. In this study, NP single exposure had a greater impact on the C. pyrenoidosa metabolome than repeated exposure. Table 3 summarizes the studies that used the metabolomics technique to assess the effect of AgNPs in vitro on different cell lines.

TiO 2 NPs
Micro-titania (titanium oxide, TiO 2 ) particles are known as biologically inert in humans, enabling their use in many products such as cosmetics and pharmaceuticals. 204,205 Nano-titania (TiO 2 NPs) are also used as additives in many products such as sunscreen products, paints, printing ink, rubber, paper, sugar, cement, toothpaste, lm, biomedical ceramics, implanted biomaterials, antimicrobial plastic packaging, and self-cleaning sanitary ceramicss. 206 However, TiO 2 NPs can enter the body via inhalation, ingestion, and dermal contact and they have been shown to exert signicant toxic effects, such as cell metabolic change, 206 chronic pulmonary inammation, 207 and pro-inammatory effects in cells. 208 Raja et al. 209 reviewed the microenvironmental inuence of TiO 2 NP-induced mechanical stimuli on tumor cells and showed using the omics analysis that the exposure of cancer cells to TiO 2 NPs caused gene mutations, protein alterations, and metabolite changes. Chen et al. 210 observed mitochondrial dysfunction caused by TiO 2 NPs in a macrophage (RAW) cell line and primary mouse bone marrow-derived macrophages (BMDM) using a combination of metabolomics, lipidomics, and proteomics. The targeted UPLC-MS-based metabolomic analysis revealed a signicant upregulation in the production of COX-2 metabolites including PGD2, PGE2, and 15dPGJ2, indicating an inammatory response in macrophages. The authors also used GC-MS-based metabolic ux analysis, which is a technique that uses MS to track the fate of stable isotope tracers (e.g., 13 C-glucose and 15 Nglutamine), allowing the investigation of the contribution of specic metabolic pathways to the prevailing levels of specic metabolites, 211 to measure the metabolic ux in the tricarboxylic acid (TCA) cycle using 13 C-labelled glutamine. They observed a downregulation in TCA cycle metabolism and ATP production caused by signicant mitochondrial dysfunction aer the exposure of macrophages to TiO 2 NPs. In a similar study, Tucci et al. 206 studied the response of the human keratinocyte HaCaT cell line aer exposure to 10-100 nm TiO 2 NPs and found that the NPs were only present in the phagosomes of the cells without their internalization in any other cytoplasmic organelle. Specically, "268" metabolites were detected using GC/LC-MSbased metabolomics, of which 85 metabolites were found to be signicantly altered at 100 mg mL −1 dose of NPs. As stated in other studies, TiO 2 NPs have shown signicant and rapid effects on mitochondrial function by altering energy metabolism and anabolic pathways. However, they did not affect the cell cycle phase distribution or cell death. Jin et al. 212 used GC/TOFMS-based metabolomics to study the metabolic changes in L929 cells and their corresponding culture media induced by 5 nm-TiO 2 NPs. At concentrations higher than 100 mg mL −1 , the NPs caused a depletion in the cellular carbohydrate metabolism (the major biochemical metabolism pathway) aer causing energy metabolism disruption, pentose phosphate pathway inhibition, nicotinamide metabolism block, mitochondria damage, and oxidative stress activation. Bo, Jin, Liu et al. 213 again used GC/TOFMS-based metabolomics to study the change in amino acid levels in L929 cells aer they were exposed to TiO 2 NPs. The study revealed that seven metabolic pathways among the regulated pathways were signicantly altered including 12 amino acids, i.e., L-a-alanine, b-alanine, glycine, L-aspartate, L-methionine, Lcysteine, glutamate, L-pyroglutamate, L-asparagine, L-glutamine, S-adenosyl methionine, and L-lysine.
In dental science, the use of TiO 2 NPs as an additive to glass ionomer cements is known to improve their mechanical and antibacterial properties. However, the study by Garcia-Contreras et al. 214 showed that these NPs may induce pro-inammation in human gingival broblast (HGF) cells. Nevertheless, the molecular mechanism of the pro-inammatory action of TiO 2 NPs on these cells was still unclear. MS metabolomics was used to reveal the mechanism of this pro-inammatory action by the treatment of HGF cells with IL-1b alone or in combination with TiO 2 NPs. 215 A total of 109 metabolites was successfully identied and quantied by CE/ TOFMS. Most amino acids levels were downregulated at high concentrations of TiO 2 NPs, while ophthalmate, a-aminoadipate, kynurenine, and b-alanine were upregulated. The activation of the urea cycle, polyamine, S-adenosylmethionine, and GSH synthetic pathways was stronger than that of the other pathways. The intracellular levels of urea cycle metabolites were downregulated signicantly in the presence of both IL-1b/TiO 2 NPs. In conclusion, ornithine was downregulated, which led to an immediate decline in putrescine. That latter is used to synthesize spermidine, which has anti-inammatory properties. Thus, the reduction of this polyamine level accelerated the inammation in HGF cells upon exposure to a combination of IL-1b/TiO 2 NPs.
Kitchin et al. 216 studied the effect of four different TiO 2 NPs (in addition to two CeO 2 NPs) on human liver HepG2 cells. Using LC/GC MS-based metabolomics, ve out of the six NPs were found to cause a signicant downregulation in GSH concentration. The authors observed a decrease in the GSH system in GSH precursors (glutamate and cysteine), GSH itself, and GSH metabolites (the gamma-glutamyl condensation products, glutamine, alanine, valine, 5-oxoproline, and cysteine-GSH). Among the 265 metabolites detected, the reduction in GSH was the largest deregulation. This indicates that the NPs are acting via an oxidative stress mode, which is a consistent biochemical effect of NPs.
Metabolomics can help to better understand the transition from in vitro to in vivo systems of NPs toxicity given that it can be applied in both types of experiment. For example, Cui et al. 217 employed LC-MS-based metabolomics to investigate the effect of four metal oxide NPs, including TiO 2 NPs, in vitro on human bronchial epithelial (BEAS-2B) cell line, and in vivo on mouse model aer lung exposure. Their study showed that in vitro metabolomic ndings can effectively reveal the biochemical effects in vivo, and that LC-MS-based metabolomics is sensitive enough to detect the tiny metabolomic changes when conventional cytotoxicity assays cannot detect any signicant effect. Fig. 13a shows the workow of this study. BEAS-2B cells were exposed to the four studied NPs, and then the metabolomics experiment was performed in vitro. This was followed by validation in vitro by enzymatic assays, in vivo using a mouse model aer lung exposure to respective NPs, and nally by cellular   function assays. The TiO 2 NPs signicantly altered the metabolic pathways of sphingosine-1-phosphate, fatty acid oxidation, folate cycle, inammation/redox, and lipid metabolism, inducing inammation. In addition, this effect was dosedependent for some metabolites. Fig. 13b shows the altered metabolites and effect of the four studied metal oxide (MO x ) NPs and their numbers, respectively. Metabolomics is also applied in many in vivo nanotoxicity studies. 218,219 For instance, Chen et al. performed three recent studies of TiO 2 NP toxicity in vivo using MS-based metabolomics, once in rats by feces metabolite analysis, 220 and then screened for urine 221 and serum 222 biomarkers in human workers exposed to these NPs in factories. This group also performed another metabolomics study using rat serum aer subchronic oral exposure of TiO 2 NPs. 223 Han et al. 224 used MSbased metabolomics to study the inuence of TiO 2 NPs on the fecal metabolome in rats aer oral administration for 90 days. Åslund et al. 225 used NMR-based metabolomics to assess the effects of 5 nm-TiO 2 NPs on Eisenia fetida earthworms and observed metabolic changes related to oxidative stress. Eight years later, Zhu et al. 226 used transcriptomics besides metabolomics to investigate the same effect of TiO 2 NPs on the same earthworm and noticed that the antioxidant system and metabolic proles of the earthworms were signicantly affected. Ratnasekhar et al. 227 used MS-based metabolomics to investigate the effects of TiO 2 NPs on the soil nematode Caenorhabditis elegans. The results indicated the disruption of the tricarboxylic acid (TCA) cycle, arachidonic acid metabolism, and glyoxylate dicarboxylate metabolism pathways. For more about the in vivo metabolic effects of NPs including Ag, TiO 2 , and carbon-based NPs on organisms (plants, aquatic, and terrestrial invertebrates), the reader is kindly referred to the chapter by Farré and Jha. 165 Metabolomics reveals the global responses that cannot be observed by conventional toxicity endpoints, leading to an effective assessment of the effects of NPs in the environment, in vivo, and in vitro. Metabolomics has also been used to reveal the metabolite corona that is surrounding TiO 2 NPs. 228,229 Table 4 summarizes the studies that used the metabolomics technique to assess the effect of TiO 2 NPs in vitro on different cells.

SiO 2 NPs
The annual global production of SiO 2 NPs is reported to exceed 1.5 million tons, making SiO 2 NPs one of the most widely used NPs in the industrial manufacturing, drug delivery, cancer therapy, and biotechnological elds. 39 This widespread is due to their biocompatibility, stability, and other unique properties compared with their bulk. 237 Although SiO 2 NPs have been shown to have different cytotoxic effects on cells, the molecular mechanism of this cytotoxicity still needs to be explored using novel analytical techniques, such as metabolomics. Huang et al. 238 used MSbased metabolomics to reveal the molecular information of the effect of SiO 2 NPs on the human fetal lung broblast MRC-5 cell line. The authors observed NP dose-dependent changes in the metabolic proles of the cells. As the dose increased, there  240 used NMR-based untargeted-metabolomics to study the effect of amorphous SiO 2 NPs on the human hepatoma HepG2 cell line and mice liver (Fig. 14). Firstly, this study determined the altered metabolites in the cells and mice liver using OPLS-DA analysis ( Fig. 14a and d, respectively). Subsequently, the selected signicantly altered metabolites were determined ( Fig. 14b and e), followed by pathway analysis using the MetaboAnalyst 3.0 soware ( Fig. 14c and f). In both in vitro and in vivo systems, the perturbation of GSH metabolism and the depletion of the GSH pool were detected aer aSiO 2 NP treatment. Moreover, the in vitro results were further supported by the in vivo data, specically for metabolite proling and pathway analysis, were there were 8 common altered metabolic pathways in the two systems. This study revealed that the major causes of aSiO 2 NP-mediated hepatotoxicity were the suppression of GSH metabolism and oxidative stress. In a similar study, Bannuscher et al. 230 studied the responses of rat lung epithelial cells (RLE-6TN) and alveolar macrophages (NR8383) (in vitro) to four well-selected SiO 2 NPs, differing in structure, size, and surface charge, and compared the results to in vivo responses in rat lung tissues. The authors observed a cell-specic time-and concentration-dependent changes in vitro and identied several biomarker candidates such as Asp, Asn, Ser, Pro, spermidine, putrescine, and LysoPCaC16:1 in vitro, and then veried them in vivo.
It was proven that SiO 2 NP exposure inevitably induces damage to the respiratory system, however, knowledge of its mode of action and metabolic interactions with the cells is limited. Zhao et al. 241 performed a study to reveal the molecular information of the metabolic responses of the lung bronchial epithelial BEAS2B cell line aer SiO 2 NP exposure, using MSbased metabolomics. They revealed that even with low cytotoxicity, SiO 2 NPs still caused global metabolism disruption. Specically, ve metabolic pathways were signicantly perturbed; in particular, oxidative stress, as conrmed by GSH depletion, mitochondrial dysfunction-related GSH metabolism, and pantothenate and coenzyme A (CoA) biosynthesis. The identied key metabolites were GSH, glycine, beta-alanine, cysteine, cysteinyl-glycine, and pantothenic acid. Oxidative DNA damage and cell membrane disintegration were detected by observing elevated 8-oxo-2 ′ -deoxyguanosine (8-OdG) and decreased phospholipids levels.
Several studies compared the effect of SiO 2 NPs on cells to other NPs using metabolomics and other omics techniques. For example, Karkossa et al. 231 used targeted metabolomics and global proteomics to compare the effect of SiO 2 NPs with different particle sizes, surface charges, and hydrophobicity to the effect of TiO 2 , graphene oxide (GO), phthalocyanine blue, phthalocyanine green, and Mn 2 O 3 NPs on RLE-6TN alveolar epithelial cells. Alternatively, Cui et al. 217 used MS-based metabolomics to reveal the signicantly altered metabolites and metabolic pathways in human bronchial epithelial cells and a mouse model exposed to four different types of metal oxide NPs (SiO 2 , ZnO, TiO 2 , and CeO 2 ) at both high (25 mg mL −1 ) and low (12.5 mg mL −1 ) doses (see Fig. 13). Table 5 summarizes the studies that used metabolomics technique to assess the effect of SiO 2 NPs in vitro on different cells.

ZnO NPs
Zinc oxide NPs are gaining increasing attention due to their unique properties, especially their optical and electronic properties. Also, they can be prepared using a variety of methods and in a range of different morphologies. 246 This makes them the third highest global production volume among metalcontaining NMs 247 and excellent for a broad range of applications, including optoelectronic devices (light-emitting diodes (LEDs), laser diodes, solar cells, and photodetectors), electronic devices (transistors), 246 and active compounds in sunscreens, drug delivery, biomedical engineering, food additives, and cosmetics. 248 It was shown that human exposure to these engineered NPs can cause health problems for both consumers and industry workers, making it important to further investigate in their toxicity and improve their safety when they are used and produced. 217,249 The respiratory tract is the primary route of exposure to airborne NPs such as ZnO. Thus, it is common to use the bronchial epithelial BEAS-2B cell line as an in vitro model to study the toxicity of these NPs. For instance, Lim et al. 247 used this cell line to perform an MS-based metabolomics study to reveal the effect of ZnO NPs on the respiratory system. The authors revealed ROS-mediated cell death associated with mitochondrial dysfunction and interference in regulating energy metabolism. This was concluded aer observing a signicant decrease in the levels of amino acids (valine, tryptophan, lysine, proline, threonine, glycine, serine, glutamic acid, and aspartic acid) and TCA intermediate metabolites (citrate) (Fig. 15a). These results indicate that ZnO NPs can be seriously harmful to human health if they were inhaled.
Although Zn is a key micronutrient for plants, a high dose of this metal is toxic to plants either in the nano or other forms.    254 elucidated toxicodynamic differences at the molecular scale between ZnO NPs and ZnCl 2 in Enchytraeus crypticus, a model species in soil ecotoxicology, using non-targeted metabolomics. They found that the number of altered metabolites aer Zn 2+ exposure was larger than the number of altered metabolites aer ZnO NP exposure, indicating the higher toxicity of the Zn ionic form (Fig. 15b and c). For more information about nanotechnology in agriculture and the effect of metallic-, metal oxide-, and carbonbased-NPs on plants, the reader is advised to read the review by Paramo et al. 255 and review by Majumdar et al. 256 The toxic effects of a NP may be reduced by applying coexposure with another NP. For instance, Wu et al. 257 studied the combined effects of graphene oxide (GO) and ZnO NPs on human A549 cells using NMR-based metabolomics. PLS-DA analysis showed that the control and GO-alone exposure groups overlapped, indicating a low effect of 10 mg L −1 GO on the metabolome proles. In contrast, ZnO NP-alone exposure signicantly altered the metabolome proles in A549 cells. A total of 14 altered metabolites was shared in the ZnO NP-alone and the co-exposure with GO groups. However, the levels of fold changes of the 14 shared metabolites were lower in the coexposure group than that in the ZnO NP-alone group. This tendency indicates that GO alleviated the toxicity induced by ZnO NPs in the cellular metabolism by reducing or blocking their internalization in the cells (Fig. 15d-f). Table 6 summarizes the studies that used metabolomics technique to assess the effect of ZnO NPs in vitro on different cells.

5.6
Other metal-and metal oxide-NPs 5.6.1 Cobalt ferrite (CoFe 2 O 4 ) NPs. Cobalt ferrite (CoFe 2 O 4 ) NPs have interesting properties, such as mechanical hardness, excellent chemical stability, high anisotropy, superparamagnetism, and coercivity. 259 Oliveira et al. 260 studied the cytotoxic effect, cellular uptake, and metabolomic effect of CoFe 2 O 4 NPs on the HeLa and HaCaT cell lines. This study revealed, using NMR-based metabolomics, that although the uptake of NPs at 2 mg mL −1 caused low cytotoxicity, it signicantly impacted the cell metabolism. Both cell lines shared stress-related metabolic changes such as upregulation in alanine and creatine. A downregulation in fumarate level was present in HeLa cells treated with the NPs. Given that this metabolite is associated with cell proliferation and tumor growth, it was concluded that CoFe 2 O 4 NPs can inhibit tumorigenesis.
5.6.2 Copper oxide (CuO) NPs. Copper oxide (CuO) NPs have been used in heat transfer uids, semiconductors, and intrauterine contraceptive devices. 261 Human exposure to CuO NPs is rapidly increasing, and thus reliable toxicity test systems are urgently needed. It was shown that CuO NPs are more toxic than their microparticles (MPs). To reveal the mechanism of this toxicity, Murgia et al. 262 used MS-based metabolomics to the study the effect of CuO micro-and nano-particles against  human bone marrow mesenchymal stem cells (hBMMSCs). It was found that the MPs increased the levels of serine, glyceric acid, and succinic acid, while glutamine was the only discriminant metabolite for the class of samples treated with NPs. This proves that ROS formation is the active mode of action in NP treatment, providing the rst step toward the understanding of the mechanism of toxicity of CuO NP-treated cells. Wang et al. 263 compared the effect of CuO NPs, MPs, and Cu ions on microalga Chlorella vulgaris aer 5 days exposure using global metabolomics. A total of 75 differentiated metabolites was identied. Most metabolic pathways perturbed aer CuO NP exposure were shared by that aer CuO MP and Cu ion exposure. Only one difference between metabolic responses to particles and that to ions was observed, which is the accumulation of fatty acid oxidation products, i.e., particles caused higher fold changes at 1 mg L −1 and lower fold changes at 10 mg L −1 compared with ions. This indicates the signicant role of dissolved Cu ions on the toxicity of CuO NPs and MPs. Kruszka et al. 264 compared the effect of Cu and CuO NPs on the secondary metabolism of Hypericum perforatum L. cell suspension cultures and found that metal NPs induce higher metabolic changes than their counterpart metal oxide NPs. Table 7 summarizes the studies that used the metabolomics technique to assess the effect of other metal/metal oxide NPs in vitro on different cells.

5.7
Carbon-based NPs 5.7.1 Graphene. Graphene has attracted signicant attention due to its unique and novel properties, which has promising applications in different elds, including biomedical engineering, tissue engineering, and biosensors. However, graphene-based drug delivery systems and other biomedical applications are associated with challenges related to the safety of carbon NMs for clinical use. Many groups have investigated the cytotoxicity of graphene. In this case, although the conventional in vitro toxicity assays of graphene yielded contradictory results, Jiao et al. 286 used the metabolomics approach to investigate the metabolic responses on graphenetreated HepG2 and detected twelve metabolites as potential biomarkers. The authors also determined three KEGG pathways including arginine and proline metabolism, purine metabolism, and glycophospholipid metabolism.
Adamson et al. 287 studied the metabolic change caused in macrophages by graphene nanoplatelets. The number of compounds changed following exposure to graphene was determined to be both concentration and time dependent. The identied metabolites are related to several metabolism pathways, such as GSH metabolism, pantothenate and CoA biosynthesis, sphingolipid metabolism, purine metabolism, arachidonic acid metabolism and others. Graphene oxide (GO) also has some biomedical applications but a greater    understanding of its cytotoxicity and efficiency as a drug carrier is needed. Raja et al. 288 used NMR-based metabolomics to assess the metabolic effect of GO nanosheets on MCF-7 breast cancer cells. The treatment affected arginine metabolism, proline metabolism, and aminoacyl-tRNA biosynthesis, including anabolism and catabolism. Moreover, GO increased the number of metabolic disturbances in cancer steroids in a dosedependent manner. 5.7.2 Carbon black NPs (CBNPs). Carbon black NPs (CBNPs) are the core component of ne particulate matter in the atmosphere, which make its exposure to the respiratory system easy. It was reported that CBNPs can induce inammation, oxidative stress, and changes in cell signalling and gene expression in mammalian cells and organs. Hou et al. 289 used MS-based metabolomics to reveal this mechanism in A549 cells. Their study identied a total of 32 differential metabolites between the CBNP exposure and control groups. The pathway analysis showed that the metabolic changes were involved in tricarboxylic acid (TCA) cycle, alanine, aspartate, glutamate, and histidine metabolism. This suggests that CBNPs act by affecting the normal process of energy metabolism and disturbing several vital signalling pathways in the cells, nally leading to cell apoptosis. Other studies performed in vivo experiments and assessed the effect of carbon-based NMs on the ecosystem by studying some models such as earthworms. For instance, Xu et al. 290 studied the impacts of three carbon NMs, i.e., carbon black (CB), reduced graphene oxide (RGO), and single-wall carbon nanotubes (SWCNTs), on Eisenia fetida, an early warning soil invertebrate for pollution events. They concluded that the soil environmental risk of C-NMs was related to their particle morphology, contributing to a comprehensive understanding of nano-agriculture. Table 8 summarizes the studies that used the metabolomics technique to assess the effect of other C-based NPs in vitro on different cells.

Polymeric NPs
Polymeric NPs such as polystyrene (PS) are gaining considerable attention because of their growing accumulation in the environment and the high probability of human and animal exposure. Therefore, more research must be done to increase our understanding of their potential effects. Kim et al. 303 studied the metabolic effects of PS NPs on human epithelial colorectal cells (Caco-2). The authors designed two methods to investigate the exposure of Caco-2 cells to NPs, where the rst is by exposing cells to a high concentration of 50 nm PS NPs for 24 h (acute), and the second is by exposing them to a relatively lower concentration for over 1.5 months (chronic). The biological assays were performed using specic NP concentrations, which were 10 and 80 mg mL −1 for the acute model and 0.1 mg mL −1 for the chronic exposure model. Aer acute exposure, untargeted metabolic proling was performed and the change in lipid metabolic pathways determined, including steroid and arachidonic acid metabolism. Alternatively, chronic exposure induced relatively minor changes. However, there was still a potential effect on fatty acid biosynthesis, indicating that acute and chronic exposure to PS NPs may disturb lipid   Glutamate, asparagine, choline, and creatine Table 9 Summary of polymeric NP-induced perturbation of metabolic pathways and their biological impact on different cells homeostasis. They also conrmed the changes in the expression levels of lipid transcriptional regulatory factor coding genes, namely, sterol regulatory element-binding transcription factors 1 and 2. The total fatty acid composition was further studied to verify metabolic disturbance by chronic exposure. Su et al. 304 investigated the effect of poly(L-lactic acid) (PLLA) nanobers on PC12 cell differentiation at the metabolic level. Many differential metabolites were identied and two pathways and three metabolites critical to PC12 cell differentiation were inuenced by the nanobers. Table 9 summarizes the studies that used the metabolomics technique to assess the effect of other polymeric NPs in vitro on different cells.

Conclusions
Currently, it is obvious for many researchers that nanotechnology provides countless benets, and consequently its demand is increasing daily. It is very important to assess the safety of every NP that is being produced and to test its bene-cial and disadvantageous effects. Understanding the interactions between NPs with cells and how NPs are internalized in cells are the rst step to assess their toxicity. Here, not only the type of the NP matters, but also its physicochemical properties such as size, shape, and surface properties. It was proven that NPs with different properties have different effect on cells. Some sizes of NPs are not toxic, but others are severely harmful to cells. In general, conventional assays are the most used strategy to assess the effect of NPs on cells. However, these assays were found to interfere with NPs, giving false results in some cases, and they are unable to reveal the molecular information of the toxicity or effect of NPs. Thus, an increasing number of researchers are heading towards the use of other analytical techniques. Metabolomics is a powerful technique that provides a full picture of the toxicity of NPs by analyzing the functionality of an existing living system by measuring the metabolic change induced by NPs. Unlike conventional assays, this tool does not interfere with NPs and provides information at the molecular level about the toxicity of NPs. Furthermore, it forms an additional bridge that connects the in vitro with the in vivo models, as proven by several references. It was shown in this review that NPs can harm the cell through different ways, including cell viability and proliferation perturbation, inammatory response, oxidative stress, ROS generation, and cell death via apoptosis or necrosis. Moreover, using metabolomics, NPs were shown to perturb the metabolic pathways of cells, including the TCA cycle, DNA and protein synthesis, glycolysis, glutathione, and amino acid pathways. Thus far, metabolomics has been used in many studies to assess the effects of different NPs on living organisms. However, more research needs to be done to identify and validate specic biomarkers of the effects of NPs on cells. Reaching this point will introduce a huge step in determining the toxicity of NPs and how to avoid or multiply this toxicity. This will help in designing better NPs for biomedical applications and producing safer NPs for industrial applications. Nevertheless, long-term targeted studies should also be performed to ll many gaps in this eld. Also, the combination of metabolomics with other techniques is Review Nanoscale Advances required in some cases to provide a bigger picture on the events occurring in the cell.

Conflicts of interest
There are no conicts to declare.