Characterizing nanoplastics-induced stress and its SERS fingerprint in an intestinal membrane model

Epithelium membranes provide important barrier functions, and it is important to understand how nanoparticle (NP) exposure affects their barrier function. In this manuscript, we investigate NP-induced stress in a Caco-2 intestinal epithelial membrane model and its effect on the vibrational spectrum of the extracellular medium that can be sampled and investigated without perturbation of the cells. Monolayers of Caco-2 cells were incubated with 50 nm diameter polystyrene (PS) NPs functionalized with amine or carboxylic acid groups and concentrations of 1 × 10 12 –1 × 10 14 PS NPs mL –1 for 6 and 18 hours. Reactive oxygen species (ROS) generation, cell viability, and intestinal membrane integrity measurements were performed to detect and quantify PS NP-induced membrane damage under the acute exposure conditions. After identifying conditions that result in NP-induced stress, Surface Enhanced Raman Spectroscopy (SERS) was applied to monitor the composition of the medium in direct contact with the intestinal cells and to detect potential PS NP-induced changes in the cellular metabolism in real time and in a minimally invasive fashion. The analysis of the SERS spectra through artificial intelligence algorithms and chemometric tools revealed concentration, exposure time-, and surface chemistry-dependent differences in the cellular metabolism in response to PS NPs. The SERS spectral analysis identifies the ring breathing mode of hypoxanthine (C 4 H 4 N 4 O), as a spectroscopic marker for the PS NP-induced loss in membrane integrity.

to the elements, macroscopic objects from these plastics undergo photo-degradation as well as thermo-oxidative and mechanical degradation and gradually fragment into smaller and smaller pieces. [2] Synthetic polymer fragments with dimensions smaller than 5 mm are collectively referred to as microplastics, independent of their exact chemical composition. [3,4] The weathering of plastics does not stop with micrometer-sized particles but continues on sub-micrometer length scales to generate plastic nanoparticles (NPs). Importantly, the environmental mobility, as well as the uptake and fate of these nanoscale particles in animals and humans can differ from larger microplastics of the same chemical composition. [5][6][7] Due to their large surface-to-volume ratio, NPs show unique interactions with cells and can serve as carriers for adsorbed toxicants. [8][9][10] However, even materials that are benign in bulk form and free of adsorbed toxicants can show toxicity when formulated as NPs. [11][12][13][14][15][16] Lambert et al. investigated NP generation from seven macroscopic plastic materials (PE and PP pellets, PP films, PP sheet, PS coffee lids, PET water bottles, polylactic acid (PLA) cups) exposed to accelerated weathering conditions and found that PS and PLA generated the largest amount of NPs. [17] PS NPs can be synthesized with sharp size distributions and defined surface properties, which makes them ideal model systems for characterizing the effect of nanoplastics on cellular systems under defined conditions in this work. [18][19][20][21][22][23] As the most probable uptake route of PS NPs is via ingestion of contaminated food or water, we investigate in this work the interactions of PS NPs with an intestinal membrane model. NPs with diameters < 5 nm can cross the intestinal membrane barrier through paracellular diffusion. [24,25] Larger NPs can cross the intestinal membrane by persorption through degrading epithelial cells [7] or through energy-dependent transcytosis. [26][27][28][29][30][31][32][33][34][35] Another concern, in addition to NPs crossing the membrane, is that interactions between NPs and cells perturb cellular functions, damage the intestinal membrane integrity, and compromises its barrier function. Indeed, Thubagere et al. observed in an intestinal cell monolayer model that PS NPs uptake can trigger oxidative stress and that the associated NP-induced hydrogen peroxide formation triggers a spread of apoptosis across the entire monolayer. [30] In another study it was shown that 4 hours exposure to PS NPs disrupts iron transport across the intestinal epithelium, effectively inhibiting the absorption of the essential nutrient. [36] Although some previous studies have found no indications of PS NP-induced genotoxicity or loss of structural integrity in intestinal membranes models, [37] there is also prior evidence that PS NPs can trigger genotoxicity in epithelial cells and macrophages. [38] Overall, the work performed so far suggests that nanoplastics can have a detrimental effect on the intestinal membrane, but that the effect depends on NP concentration, surface charge, size, and exposure duration. A fundamental understanding of the interactions between nanoplastics and the intestinal membrane as well as the availability of ideally label-free markers of cell stress and membrane damage are key for an accurate risk assessment and to develop effective strategies to mitigate potential detrimental effects.
As interactions between NPs and intestinal cells are complex and depend on a large number of potential variables, in vitro intestinal membrane models are frequently used to investigate the effect of plastic NPs on the membrane integrity under defined conditions. [39][40][41][42] In these model systems, transepithelial electrical resistance (TEER) measurements facilitate a quantitative monitoring of membrane integrity and permeability. [43][44][45] Cytotoxicity and cell viability are quantified using MTT and Lactate dehydrogenase (LDH) assays, and the oxidative stress induced by NPs is determined through reactive oxygen species (ROS) measurements. [45][46][47] Optical and electron microscopies facilitate a characterization of transepithelial NP transport with subcellular spatial resolution. [48][49][50] NPs also affect the cellular metabolism, and metabolomics is another potentially powerful approach for both detecting and characterizing cellular responses to NPs at a molecular level. [51][52][53] NP-induced differences in the cell metabolism cause changes in the composition of cells as well as the extracellular medium due to variations in the nature and quantity of the molecules uptaken by and released from the cells. [54] Although metabolomics is traditionally performed with mass spectroscopy, Raman spectroscopy, in particular surface enhanced Raman spectroscopy (SERS), has shown promise for the metabolic profiling of the cellular activity of bacterial cells and human cancer cells, [55][56][57][58][59][60] as well as for the detection of biomarkers in serum. [55,61,62] SERS is a vibrational spectroscopy that provides unique opportunities for signal amplification in aqueous media as water has low Raman cross-sections while the signal from Raman-active molecular vibrational modes in analytes can be enhanced by multiple orders of magnitude due to an electromagnetic enhancement effect in the vicinity of nanostructured noble metal surfaces. [63][64][65][66][67] This amplification of molecule-specific vibrational information facilitates a sensitive detection and identification of metabolites rapidly without the need for elaborate sample preparation or enrichment. In this work, we exposed an in vitro intestinal cell membrane model with PS NPs in a concentration range that warrants NP-induced cell stress and combined TEER, ROS and cell viability measurements with SERS measurements to validate, characterize and identify nanoplastics-induced changes in the vibrational spectrum of the extracellular medium under conditions of increased intestinal membrane permeability.

RESULTS AND DISCUSSION
The size and surface charge of nanoplastics generated in the environment is determined by the weathering process and can vary depending on the specific conditions. We used colloidal PS NPs with amine (NH 2 -PS) or carboxylate (COOH-PS) surface ligands as model system in this work to investigate the interactions between PS NPs and cells under well-defined conditions. The hydrodynamic diameter and zeta potential of the NPs in water and growth medium are summarized in Table 1. In water, the hydrodynamic diameter is 57 ± 5 nm for NH 2 -PS, and 64 ± 8 nm for COOH-PS, and the zeta potential (ζ) of NH 2 -PS is 48.7 ± 2.1 mV compared to −50.7 ± 2.3 mV of COOH-PS. In growth medium, the initial hydrodynamic diameter (D H ) increases to 96 ± 12 nm for NH 2 -PS and 89 ± 13 nm for COOH-PS, and we measured a zeta potential of −6.1 ± 1.2 mV for NH 2 -PS and −9.6 ± 0.7 mV for COOH-PS. We attribute the nearly identical zeta potentials for both NH 2 -PS and COOH-PS and its low absolute value in growth medium that is rich in proteins and other biopolymers to non-specific protein adsorption, that is, corona formation. [68][69][70] The hydrodynamic diameter of both NH 2 -PS and COOH-PS in growth medium gradually increases as function of time and converges against approximately 350 nm in the case of NH 2 -PS and 150 nm in the case of COOH-PS after ∼6 hours ( Figure S1a). SEM images ( Figure S1b) reveal that this additional increase in the hydrodynamic diameter is due to some self-association of the PS-NPs. The different NP stabilities (i.e., hydrodynamic diameters) point to differences in the composition and structure of the corona formed around NH 2 -PS and COOH-PS. Caco-2 cells are human colon carcinoma cells that grow into a monolayer on permeable membranes. The cells can be induced to differentiate into mature absorptive enterocytes with well-developed microvilli, tight apical junctions, and a polarized distribution of membrane components. [39][40][41][42][71][72][73][74] Due to the resemblance of Caco-2 cells with intestinal enterocytes, differentiated Caco-2 cell monolayers are a common model system to study the transepithelial membrane transport of various NPs, including polymeric [71][72][73] and plastic NPs, [33][34][35][36][37][38] at a fraction of the cost and without the ethical implications of a small animal experiment. In this work, NH 2 -PS and COOH-PS were incubated with differentiated monolayers of Caco-2 cells for 6 or 18 hours, which are typical incubation times for mechanistic barrier studies of Caco-2 cells [34][35][36][37]75] Subsequently, the effect of the PS NPs on the intestinal membrane health was evaluated, and a SERS analysis of the medium was performed to identify changes in the cellular metabolism. Figure 1 summarizes the experimental approach of this work.

NH2-PS and COOH-PS Uptake by Caco-2 Cells
Throughout this manuscript, we used PS NPs that contained fluorescent dyes embedded in their core as these are easily tractable through fluorescence microscopy. We validated that fluorescence dye release on the time scale of our experiments (up to 18 hours) is negligible (Table  S1). Confocal microscopy scans confirmed that both NH 2 -PS and COOH-PS bind to and get internalized by Caco-2 cells (Figure 2A). The NPs were incubated with the cells for 6 hours before the number of uptaken NPs was determined by measuring their fluorescence signal. The effect of PS NPs on the intestinal membrane is expected to be concentration and time dependent. Concentrations of up to 1.5 × 10 12 NPs mL -1 of PS NPs were shown to have negligible effect on membrane integrity within an exposure time of 24 hours. [47] In this work significantly higher PS NPs concentrations in the range of 1 × 10 12 -1 × 10 14 NPs mL -1 were investigated to ensure conditions that induce a measurable damage to the membrane within the limited observation time of our experiments (18 hours). The chosen NP concentrations emulate oral administrations of 10 and 1000 mg kg -1 of 50 nm PS NPs to a 70 kg human (assuming the total surface area of the human small intestine is 2 × 10 6 cm 2 [76] and the presence of microvilli increases the surface area by a factor of 20 [77] ). In this concentration range, uptake was well described by a Hill-Langmuir like expression of the form where NP assoc.
is the number of NPs per cell, NP max is the maximum number of NPs uptaken by the cells, [NP] is the input concentration of NPs, and K is a NP-specific parameter that describes binding and uptake in differentiated Caco-2 cell monolayers ( Figure 2B, 2C). A smaller value of K indicates a higher binding affinity and increased uptake. Our fits yielded K = 2.6 × 10 12 NPs mL -1 for NH 2 -PS compared to 2.6 × 10 13 NPs mL -1 for COOH-PS. The higher uptake of  (Table 1), we attribute the differences in binding and uptake to differences and the corona assembled around NH 2 -PS and COOH-PS. Positively charged NPs have a higher affinity for serum proteins, [78] so that the surface properties of NH 2 -PS and COOH-PS can result in the formation of coronas with different composition in biological media. [79] It was shown previously that NH 2 -PS contain a higher number of apolipoproteins in the corona than COOH-PS, which can have a direct effect on NP uptake. [69,80,81] The difference in particle size due to self-association observed for longer incubation times ( Figure S1) may also contribute to differences in the uptake between NH 2 -PS and COOH-PS.

Effect of PS NPs on ROS levels and cell viability
Reactive oxygen species (ROS) generation is a cellular stress response to the uptake of non-biodegradable NPs, including nanoplastics, [82,83] and NP-induced increases in ROS can result in a loss of Caco-2 membrane integrity. [30] We quantified the ROS levels in differentiated Caco-2 monolayers incubated with NH 2 -PS and COOH-PS and concentrations C1 and C2 at four time points (1, 3, 6, 18 hours) after PS NPs addition ( Figure 3A). C1 achieves increased ROS levels for both NH 2 -PS and COOH-PS at 6 hours. In case of C2, significantly increased ROS levels are detected at 1, 3, 6 hours for NH 2 -PS and at 3 and 6 hours for COOH-PS. For both flavors of NPs, the ROS levels increase with NP input concentration and peak at around 3 to 6 hours. The effect of NH 2 -PS and COOH-PS on cell viability was characterized through microculture tetrazolium (MTT) assays ( Figure 3B) performed 6 and 18 hours after NP addition. In the case of NH 2 -PS, a significant reduction in cell viability relative to the no treatment control was detected for concentrations C1 and C2 at both time points. COOH-PS show a similar behavior with the exception that the MTT value for concentration C1 at 6 hours is not yet significantly reduced relative to the no treatment control. For both flavors of NPs the cell viability decreased with input concentration and exposure time. We applied LDH assays as a second strategy to characterize the cytotoxicity of NH 2 -PS and COOH-PS at 6 and 18 hours after addition of the PS NPs ( Figure 3C). For concentration C1, NH 2 -PS and COOH-PS, show only modest differences in LDH levels relative to the no-treatment control, but for concentration C2, a strong increase of LDH levels is detected at 6 and 18 hours for NH 2 -PS and COOH-PS. Both MTT and LDH assays confirm that NH 2 -PS and COOH-PS can reduce the viability of Caco-2 cells and that the effect increases with NP input concentration and incubation time. Overall, we find that NH 2 -PS and COOH-PS have similar effects on the cell viability for input concentrations that yield similar NP uptake.
To test for a potential PS-NP induced damage to the Caco-2 cell monolayer, we mapped dead cells with a live and dead stain through fluorescence microscopy (Figure 4A). We performed these experiments with both NH 2 -PS and COOH-PS and concentrations C1 and C2 6 and 18 hours after addition of PS NPs. Images of differentiated Caco-2 cell monolayers not exposed to NPs were included as controls. In the absence of PS NPs and for PS NPs concentration C1 the fluorescence images show an intact membrane of living (green) cells with only a few dead (red) cells whose frequency is higher for 18 hours than for 6 hours. The number of dead cells increases for NH 2 -PS and COOH-PS with concentration C2 and large areas void of cells emerge, presumably because dead cells have detached from the membrane support in these areas. Figure 4B summarizes the area in % in the cell monolayer that is void of cells. The plot confirms that the "holes" in the cell monolayer increase in size with time and concentration of NPs. One notable exception is NH 2 -PS at high concentrations, for which essentially identical areas void of cells were detected at 6 and 18 hours.
The detachment of dead cells from the membrane support accounts for the holes in the Caco-2 cell monolayers observed in Figure 4B and also provides an explanation for the steep drop in the ROS levels observed for NPtreated cells between 6 and 18 hours in Figure 3A. Dead and detached cells are removed during the wash step, and a lower number of cells with increased ROS levels after 18 hours of NP treatment result in ROS levels that are comparable to those of a higher number of healthy control cells that generate less ROS per cell. Indeed, if we normalized the ROS levels by the number of cells on the substrate, we found significantly enhanced ROS levels relative to the no treatment controls for all PS NPs conditions at t = 18 hours ( Figure S3).

Characterizing the effect of PS NPs on Caco-2 membrane integrity
We measured the TEER of monolayers of differentiated Caco-2 cells incubated with NH 2 -PS or COOH-PS with concentration C1 or C2 every hour for a total of 24 hours  Figure 4C). All TEER values are normalized by the no treatment control for the same time point. For NH 2 -PS with an input concentration of C1, the TEER remains constant for the first 2 hours before it transiently decreases by approximately 30%. After approximately 10 hours, the TEER starts to increase again and eventually converges against the starting value prior to NP exposure. If the NH 2 -PS concentration is increased to C2, a fast, continuous and non-reversible drop of the TEER value by nearly 80% of its initial value is observed within the first 10 hours. In the case of COOH-PS, the TEER does not decrease over the entire observation time for C1, but an increase of the input concentration to C2 results in a rapid and continuous decrease in TEER of similar magnitude as observed for NH 2 -PS. The strong decrease in TEER for concentration C2 of NH 2 -PS and COOH-PS indicates an increased permeability of the cell monolayer and is consistent with the substantial detachment of dead cells from the membrane observed for these conditions in the live/dead staining experiments in Figure 4A.

Vibrational characterization of the growth medium of Caco-2 cell monolayers through SERS
After identifying PS NPs conditions that result in a decrease of the Caco-2 cell monolayer integrity, we apply in this section SERS for a vibrational characterization of the medium in which the Caco-2 cells are cultured to screen for metabolomic changes induced by the PS NPs. The underlying hypothesis is that PS NP-induced changes in the cellular metabolism lead to changes in the extracellular medium through changes in the cellular uptake of nutrients from the medium or release of metabolites into the medium, and that these changes can be sensitively detected with no special sample preparation through SERS. The electromagnetic amplification of the Raman signal in SERS depends on the fourth power of the electric (E-)field (∼E 4 ). SERS spectra are consequently typically recorded on nanostructured noble metal surfaces that provide strong local E-fields through excitation of Localized Surface Plasmon Resonances (LSPRs). [63][64][65][66][67]85] In this work, we prepared monolayers of gold NPs with a diameter of 78 ± 1 nm by the Langmuir-Blodgett (LB) technique and transferred them onto a glass coverslip as SERS substrates (see Methods). An optical darkfield image of a substrate is shown in Figure 5A. SEM micrographs show a layer of closely packed NPs with nanoscale crevices and junctions between individual NPs ( Figure 5B). Under resonant excitation, these nanogap structures form electromagnetic "hot-spots" that provide strong E-fields to amplify the Raman signal of molecules localized to these gaps. [86] The scattering spectrum of the gold NP film (Figure 5C) shows a broad plasmon band that is strongly broadened and red-shifted when compared with the colloidal gold NP building blocks (inset in Figure 5C). The strong spectral shift of the plasmon confirms strong electromagnetic coupling between the NPs in the film. All SERS measurements reported in this study were performed with a laser excitation at 785 nm as it overlaps with the plasmon resonance of the gold NP film and because the fluorescence background is generally low in this wavelength range.
We recorded SERS spectra of growth medium collected from the apical compartment of the Caco-2 monolayers incubated with NH 2 -PS and COOH-PS with concentration C1 and C2 6 and 18 hours after NP addition. At both time points we also measured SERS spectra of the growth medium collected from cells not incubated with NPs as well as of NH 2 -PS and COOH-PS incubated in medium without cells as controls. SERS measurements for each condition were performed with five independent experiments: each on a new SERS substrate and from a new Caco-2 cell culture. Within each independent SERS measurement, five spectra were collected from different locations on one substrate. Figure 5D shows three representative spectra of medium samples collected from NH 2 -PS (C2: 6 × 10 12 NPs mL -1 ) treated cells after 18 hours of incubation. A close inspection of the three spectra recorded under identical conditions reveals some variability between the spectra. For some peaks the relative intensities differ between the spectra, while other peaks are only detectable in selected spectra. Due to the strong Raman signal intensity enhancement provided by electromagnetic hot-spots, the SERS signal recorded at a particular location of the SERS substrate depends on the chemical composition of the hot-spots. The biological medium investigated in this work has a complex composition, containing amino acids, lipids, carbohydrates, nucleic acids and other nutrients, as well as metabolites released from the cells. This chemical complexity can result in differences in the composition of the hot-spots and, thus, the recorded spectra. [87][88][89][90][91] In addition to the compositional heterogeneity, the hot-spots also have a morphological heterogeneity that results in differences in the E-field enhancement between individual hot-spots. [65] These two effects in addition to differences in the orientation of analyte molecules in the hot-spots are sources of variability in the recorded SERS spectra. [56] Given the heterogeneity of the SERS spectra, each experimental condition is best described by an ensemble of spectra, and a systematic comparison between experimental conditions should be based on the analysis of the ensembles. [92] To obtain an overview of the spectral ensembles of this work, we calculated the second derivative spectra and determined the maxima, defined F I G U R E 6 A, Histograms of the second derivative spectra collected from media incubated with NH 2 -PS (C1: 1 × 10 12 NPsmL -1 and C2: 6 × 10 12 NPs mL -1 ) and COOH-PS (C1: 6 × 10 12 NPs mL -1 and C2: 1 × 10 14 NPs mL -1 ), and no particles (cell control) for 6 hours. The spectra for NH 2 -PS (6 × 10 12 NPs mL -1 ) and COOH-PS (1 × 10 14 NPs mL -1 ) incubated in medium with cells are also included. B, Same data for 18 hours as all features with a negative second derivative whose absolute value was at least 2.5 standard deviations higher than the noise in each spectrum. The second derivatives spectra were calculated with the PLS toolbox (Eigenvector Research) in MatLab. The fraction of spectra containing the identified maxima provide a measure of the relative frequency of that feature in the ensemble. The histograms of the peak positions obtained in this way are summarized in Figure 6 for the different experimental conditions recorded 6 or 18 hours after addition of the NPs. A visual inspection of the histograms suggests differences in the spectra recorded for PS NPs treated cells and no treatment controls as well as between the different NP conditions. To corroborate this hypothesis and to identify the spectral features that are responsible for the differences, we next applied artificial intelligence and chemometric data analysis strategies. T-distributed stochastic neighbor embedding (t-SNE) is a machine learning algorithm for dimensionality reduction and visualization. [93] Each point in the t-SNE plot in Figure 7 represents a single SERS measurement. At 6 hours ( Figure 7A), the spectra recorded for NH 2 -PS and COOH-PS with concentration C2 separate clearly from each other and the no treatment and NP controls. The no treatment and NP controls also cluster at 6 hours, but the individual clusters are less well defined and separated. With increasing time, the clustering overall improves and the individual clusters become more distinct at 18 hours ( Figure 7B). The clustering of the spectra in the t-SNE plots confirms that the spectra of one condition share similarities with each other and systematic differences with other conditions.
To further characterize the degree of similarity and dissimilarity between the different conditions, as well as to identify the spectral features that are responsible for the differences, we next performed Principal Component Analysis (PCA) [94] and principal component discriminant analysis (PCA-LDA). [95] PCA reduces the number of variables of a data set by projecting the data onto a lower dimensional space spanned by so-called principal components (PCs) that are weighed linear combinations of the original variables. In the reduced presentation, the individual spectra are defined by "scores" for the individual PCs. In PCA-LDA, linear discriminant analysis (LDA) is subsequently applied on the reduced data presentation to maximize the separation between the individual groups of observations, yielding a set of scores on a new set of linear discriminant (LD) axes. Importantly, the PCs and LDs that can separate the SERS spectra of different groups contain information about the spectral features that are responsible for the systematic differences. PCA of the second derivative spectra was performed using PCA as integrated in Matlab. The first 20 PCs were used as input for an LDA analysis that was performed with a home written code based on Chen and Hsu [96] in Matlab (see Methods). Figure 8A,B shows PCA score plots in the PC2-4 space for the spectra recorded (a) 6 and (b) 18 hours after PS F I G U R E 7 t-Distributed stochastic neighbor embedding (t-SNE) analysis of the second derivative SERS spectra of the medium collected from Caco-2 cells incubated with COOH-PS (C1: 6 × 10 12 NPs mL -1 , C2: 1 × 10 14 NPs mL -1 ), NH 2 -PS (C1: 1 × 10 12 NPs mL -1 , C2: 6 × 10 12 NPs mL -1 ), no particles cell controls (NT control), and NH 2 -PS and COOH-PS (NPs incubated in medium without cells) collected (A) 6 hours and (B) 18 hours after NP addition NPs addition. After 6 hours, the spectra of COOH-PS and NH 2 -PS with concentration C2 are well separated from each other and all other conditions, which implies i.) that these two conditions have a particular strong effect on the cells and ii.) that the cellular responses to these two NPs with different surface chemistries differ. The spectra obtained for COOH-PS and NH 2 -PS with concentration C1 form clusters that lie close together. Likewise, the clusters of the no treatment control and of COOH-PS and NH 2 -PS incubated with growth medium in the absence of cells also overlap, albeit in a separate region of the PC2-4 space than the spectra collected from cells incubated with PS NPs. Based on this clustering pattern, we can conclude that both flavors of PS NPs induce detectable cellular changes already after 6 hours even for C1, but that the spectral differences between COOH-PS and NH 2 -PS are still modest. After 18 hours, the separation between the individual groups has overall improved, but the clusters NH 2 -PS with C1 and the no treatment control still overlap in the PCA score plot as do the clusters for the NH 2 -PS and COOH-PS no cells controls. PCA-LDA yields overall similar trends as PCA but with further improved separation between the clusters in the LD1-3 space ( Figure 8C, D). For the data set recorded after 6 hours, PCA-LDA achieves a clear separation of both concentrations of COOH-PS and NH 2 -PS from the no treatment and NP controls. For the data recorded after 18 hours, PCA-LDA achieves more compact clusters and better separations between the clusters when compared with PCA performed at the same time point. At 18 hours, the only clusters whose 90% confidence intervals still overlap in the LD1-3 space are the COOH-PS and NH 2 -PS no cell controls, which are conditions for which no change in the growth medium is expected.
Overall, the multivariate data analysis is consistent with the t-SNE analysis and confirms that both COOH-PS and NH 2 -PS induce changes in the composition of the growth medium and that these changes increase with the concentration of the NPs and incubation time. Intriguingly, although COOH-PS and NH 2 -PS were compared under conditions that yield comparable uptake, the SERS spectra of the medium collected from cells treated with these particles can be distinguished not only from the no treatment control but also from each other. The spectra of all PS NPs treated cells separate from the no treatment controls along LD1 in Figure 8D, while the spectra of COOH-PS and NH 2 -PS separate along LD3. This finding suggests that all PS NPs elicit a significant cellular response, but it also implies that the responses to COOH-PS and NH 2 -PS are not completely identical. Information about the spectral features that are responsible for separating the different conditions in PCA or PCA-LDA score plots are contained in the loading spectra. These spectra plot the contribution of each wavenumber to a PC or LD. We focus here on the analysis of LDs as PCA-LDA provides the clearest separation between the different experimental conditions. The loading spectra for LD1-3 after 6 and 18 hours are plotted in Figure 9A, 9B. The spectral features that have prominent contributions in multiple loading spectra include peaks at (+/-3 cm -1 ) 728 cm -1 (LD1 6h , LD2 6h , LD1 18h ), 1002 cm -1 (LD3 6h , LD2 18h , LD3 18h ),1032 cm -1 (LD1 6h , LD2 6h , LD1 18h , LD2 18h , LD3 18h ), 1115 cm -1 (LD2 6h , LD1 18h , LD2 18h , LD3 18h ), 1320 cm -1 (LD1 6, LD3 6h , LD2 18h ), 1343 cm -1 (LD1 18h , LD2 18h ), and 1457 cm -1 (LD2 6h , LD1 18h ). In LDx y , x specifies the number of the LD and y the time of sampling. The detailed peak assignments of all loading spectra are summarized in Table S2. The prominent feature at 728 cm -1 in LD1, which separates the spectra of the no treatment control and PS NPs incubated cells, is characteristic of the ring-breathing modes in purines. Likewise, the peak at 1457 cm -1 lies in the range of C-N ring stretching modes of purines. The purine vibrational features could indicate adenine, coenzyme A or hypoxanthine. We tested these three components under the identical experimental conditions and found coenzyme A ring-breathing peaks appeared at 734 and 1430 cm -1 ; hypoxanthine shows SERS peaks at 728 and 1457 cm -1 ; adenine has SERS peaks at 737 and 1461 cm -1 ( Figure S4, S5). So we determined that hypoxanthine provides the best fit.
While the modes at 1002 , 1032 , 1115 , and 1320 and 1340 cm -1 , indicative of phenylalanine, C-C or C-N stretching modes, and C-O and C-H stretching modes of lipids (Table S2) are common in the growth medium, the 728 and 1457 cm -1 modes are rarely detected in the spectra of plain growth medium ( Figure S6). The 728 cm -1 feature is, however, very prominent in the second derivative spectra of medium collected from cells treated with NH 2 -PS (C1 and C2) for 18 hours, with NH 2 -PS (C2) for 6 hours, and with COOH-PS (C2) for 18 hours (Figure 6). The integrated 728 and 1457 cm -1 peak intensity, which correlate with the hypoxanthine concentration in the medium, are increased relative to the no treatment control for all PS-NP conditions ( Figure 9C, 9D). The 728 and 1457 cm -1 signal intensity are particularly high for NH 2 -PS and COOH-PS with concentration C2 at 18 hours and to a somewhat lower degree for NH 2 -PS (C2) at 6 hours as well as for NH 2 -PS (C1) at 18 hours. The observation of increased 728 and 1457 cm -1 peak intensities for conditions that result in increased Caco-2 membrane damage in Figure 4B and 4C suggest a relationship between hypoxanthine concentration in the medium and Caco-2 cell damage. To formally test this hypothesis, we correlated the peak intensity at 728 and 1457 cm -1 with the TEER ( Figure 9E) and the area in the Caco-2 cell monolayer void of cells ( Figure 9F). The Pearson correlations coefficients for the two peaks are R = -0.93 and -0.72, and R = 0.84 and 0.86, respectively. The strong (anti)correlation confirms that increased levels of hypoxanthine in the medium are associated with PS NPs induced cell stress and that they are particularly high under conditions that lead to cell death and increased permeability of the Caco-2 cell monolayer. Hypoxanthine acts as a marker for nanoplastics-induced damage to the Caco-2 intestinal membrane model. This interpretation is corroborated by previous studies that identified F I G U R E 9 Loading spectra of the first three discriminants LD1-3 (top to bottom) for the data set recorded after (A) 6 hours and (B) 18 hours. C, Integrated area of the 728 cm -1 , (D) 1457 cm -1 peak in the SERS spectra of Caco-2 monolayers treated with NH 2 -PS (C1: 1 × 10 12 NPs mL -1 and C2: 6 × 10 12 NPs mL -1 ), COOH-PS (C1: 6 × 10 12 NPs mL -1 and C2: 1 × 10 14 NPs mL -1 ) for 6 or 18 hours, and Caco-2 cells without treatment as the control group. The peak areas were determined by Lorentzian fits. E, Peak area at 728 and 1457 cm -1 as function of TEER value. Pearson correlation coefficient: -0.93 and -0.72. F, Peak area at 728 and 1457 cm -1 as function of area of Caco-2 membrane void of cells in % at 18 hours. Pearson correlation coefficient: 0.84 and 0.86. The data represent an average of n = 25 independent experiments. * indicates a significant difference at the P < .05 level hypoxanthine as a marker for the dysregulation of epithelial cell membranes. [97] We determined a SERS detection limit of 10 nM for hypoxanthine in medium under our experimental conditions ( Figure S7), which underlines the potential of SERS as sensitive diagnostics for intestinal stress metabolites with potential applications for monitoring intestinal membrane health both in vitro and in vivo.

CONCLUSION
We characterized PS NPs induced cellular stress in a monolayer of differentiated Caco2 cells and validated the applicability of SERS to identify NP-induced perturbation of the intestinal membrane model through spectral analysis of the extracellular medium. Multivariate analysis of the data achieved a spectral "fingerprinting" of the cellular response to the different PS NPs conditions: COOH-PS and NH 2 -PS with concentrations C1 and C2 after 6 and 18 hours. At 18 hours, it was possible not only to distinguish the PS NPs treated cells from the no treatment controls but also to distinguish different PS NPs conditions, which differed in surface chemistry (NH 2 , COOH) and concentrations, from each other. Importantly, the SERS spectra of NH 2 -PS and COOH-PS for both investigated concentrations C1 and C2 could already be reliably distinguished from the no treatment controls after 6 hours, indicating a sensitive detection of PS NP-induced metabolomic changes even under conditions with little to no apparent intestinal membrane stress based on ROS, MTT, LDH, or TEER measurements. The spectral differences become more pronounced under conditions for which a continuous non-reversible decrease in TEER indicates a loss in cell monolayer integrity. The spectra identified hypoxanthine as a metabolite in the extracellular medium that is correlated with NP-induced intestinal membrane damage.
The PS NPs concentrations required to induce membrane damage within a time window of 18 hours investigated in this work are high, and uptake of these concentrations through food and water is unlikely. At the same time, the detection of spectral changes in the medium under conditions that do not show significant alterations in membrane permeability indicates the existence of more subtle NP-induced cellular changes that may give rise to detrimental effects on longer time scales than sampled in this study, especially under chronic exposure conditions. The ability to detect PS NP-induced changes in the metabolism of intestinal cells through SERS measurements of the extracellular medium provides a new tool for the quantitative characterization of NP-induced cellular stress that can be instrumental in assessing the risk associated with nanoplastics in future studies.

Cell culture
The human colon carcinoma Caco-2 cell lines was purchased from the American Type Culture Collection (ATCC R , HTB-37). The cells were grown in advanced Minimum Essential medium (MEM) supplemented with 10% fetal bovine serum, 50 units mL -1 penicillin, and grown in an incubator at 37 • C, 5% CO 2 and 95% relative humidity. The Caco-2 cells were seeded at a density of 1×10 5 cells per well on to transwell polyester membrane cell culture inserts (12 mm diameter, 0.4 µm pore size, Corning). The growth medium was changed every 3 days. The cells were left to grow and differentiate for 21 days.

Preparation of SERS substrate
Aqueous solutions of 80 nm AuNPs (1 × 10 12 NPs mL -1 , 5 mL) were prepared as described above. 5 mL of hexane was added to form the water-hexane interface. Then ethanol was added dropwisely. The hexane layer on the top was removed, and the Au NP film formed on the water surface was carefully transferred to a glass coverslip. The prepared substrate was placed in the hood overnight to evaporate residual solvent.

Dark-field scattering characterization of SERS substrate
Scattering images of Au film were recorded using an upright microscope (Olympus BX51 WI). The scattering behavior the sample was characterized with unpolarized white light fitted with an air dark-field condenser in transmission mode. Light scattering from the Au NP film was collected using a 60x oil immersion objective (NA = 0.65). The microscope featured a 303 mm focal length imaging spectrometer (Andor Shamrock) and a back-illuminated CCD detector (DU401-BR-DD). The recorded spectra were corrected by subtracting the background signal originating from an adjacent area on the glass slide without Au NP film, and normalized by division through the excitation profile of the white light illumination source.

Quantification of PS uptake
Specified concentration of NH 2 -PS and COOH-PS in growth medium was placed into the apical chamber. Cells were incubated with particles at 37 • C and 5% CO 2 for 6, 18 hours. Cells were washed twice with Dulbeccos Phosphate-Buffered Saline (DPBS) and followed by detachment with cell dissociation buffer. The cell suspension were collected and move into a microplate, where the fluorescence intensity was measured at an excitation wavelength of 546 nm and emission wavelength of 561 nm for NH 2 -PS (488 and 520 nm for COOH-PS) using a microplate reader. Cells were stained with 0.2% trypan blue, and the cell number was counted with a cell counter. The number of polystyrene particles being uptaken was determined with a calibration curve of fluorescence intensity as function of NP concentration.

Mapping intracellular particle distribution through confocal microscopy
Cells were fixed with 4% (w/v) Pierce Formaldehye (Ther-moFisher) for 15 minutes. Subsequently, the cells were incubated with ZO-1 Monoclonal Antibody, Alexa Fluor 647 (10 ug mL -1 , ThermoFisher), Hoechst 33342 (400 ng mL -1 , Invitrogen, USA) at room temperature for 1 hour. The cells were washed three times with DPBS after each step of the fixation and labeling procedure to remove excess reagents. Samples were imaged with scanning confocal microscope (Olympus FV 1000) at 405,488 and 647 nm.

Optical inspection of live and dead cells in Caco-2 monolayers
After NP treatment, Caco-2 cells were stained with Calcein-AM (2µM) and EthD-1 (4 µM) for 45 min at room temperature. The cells were imaged in an inverted microscope (Olympus IX71) through a 10x objective, and signal captured on an electron multiplying CCD (EMCCD, Andor Ixon + ). The fluorescent signal of live cell was measured with excitation/ emission wavelengths of 473 and 510 nm. The fluorescent signal of dead cell was measured at an excitation and an emission wavelength of 580 and 620 nm. The exposure time for each monochromatic image was 0.5 seconds, and the cycle time for a full set of ten images was 5 seconds.

Cell viability measurement
Cell viability was measured using the Vybrant MTT Cell Proliferation Assay kit (Invitrogen) following the manual instructions. After PS NPs treatment, cell were incubated with MTT working solution (1 mM) at 37 • C for 2 hours, and followed with 50 µL of DMSO as solubilization solution. The absorbance was measured at 540 nm. The percentage of cell viability change was compared with cell viability from the control group.

ROS level measurement
Intracellular ROS level was detected using 10 µM of 2′, 7′ -dichlorofluorescin diacetate (DCFH-DA; Sigma Aldrich, USA) at 37 • C for 45 minutes. The DCF intensity was measured at an excitation and an emission wavelength of 488 and 525 nm. H 2 O 2 (200 µM) treated Caco-2 cells were used as positive control. The percentage of ROS level change was compared with ROS level from the no treatment control group.

LDH measurements
LDH leakage was measured from Caco-2 cells using a CyQUANT LDH Cytotoxicity Assay kit (Invitrogen, C20300) following the manual instructions. After PS NP treatment, 50 µL of cell growth medium were collected from the apical compartment and 50 µL of LDH reaction mixture was added. The wells were incubated for 30 minutes at room temperature, followed by 50 µL stop solution.
The absorbance was measured at 490 nm. The percentage of LDH activity change was determined relative to the LDH release from the no treatment control group.

TEER measurements
The trans-epithelial electrical resistance (TEER) of the cell monolayers were measured using a Millicell-ERS Voltammeter (Millipore Sigma). To eliminate the influence of temperature, measures were performed within 2 minutes after taking the transwell cell culture plate out of the incubator. Within this time, samples did not show any reading drift. Prior to the measurements, electrodes were equilibrated and sterilized according to the manufacturer's recommendations. The TEER of the cell monolayer was measured every three days after seeding to assess cell confluency and tight junction functionality.
Only cells with an average TEER value 700 Ohms are qualified for further experiments. During particle treatment, TEER were recorded every hour for 24 hours. Three measurements were taken per cell sample. The normalized TEER was calculated by dividing the measurements with the TEER obtained before particle treatment (t = 0).

SERS measurements
Twenty µL of cell medium was pipetted onto the SERS substrate. Then a glass coverslip was placed on the top to prevent evaporation of cell medium during the SERS measurement. The spectra were obtained with a Renishaw Raman microscope (model RM-2000) using a 50 x objective. The sample were excited by a diode laser with 785 nm wavelength and an excitation power of 0.7 mW. All spectra were obtained with a 10 seconds integration time, and 10 spectra were accumulated

Data processing and analysis
All data were imported into MatLab (The Mathworks Inc., Natick, MA) and an initial data processing and analysis was performed with the PLS toolbox (Eigenvector Research, Wenatchee, WA). The spectra was baseline corrected using a Whittaker filter (λ = 200) and areanormalized (n = 1). The second derivatives spectra was calculated after a Savitzky-Golay smoothing (filter width ω = 15, polynomial order σ = 2) using the PLS toolbox. The obtained second derivative spectra were converted into a barcode spectra using a home-written program. Raman shifts with a negative second derivative whose absolute value was at least 2.5 standard deviations higher than the noise were assigned a value of 1. All other Raman shifts were assigned a value of 0. The bar code of all the spectra of one condition were added and normalized to generate histograms of the second derivative spectra. t-SNE of the second derivative spectra (no barcoding) was performed with the Matlab t-SNE function. PCA of the second derivative spectra was performed using the Matlab PCA function. The first 20 PCs were used as input for an LDA analysis implemented as described by Chen and Hsu in Matlab. [96] Spectral peak areas were obtained from baseline-and area-corrected spectra through fitting of Lorentzian peak function = 0 + 2 4(x−x ) 2 + 2 (A = area, ω = width, x c = spectral peak).

A C K N O W L E D G M E N T
Björn M. Reinhard acknowledges support from the National Science Foundations through grants CBET-1822246 and CBET-2032376.

D ATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.