High content analysis at single cell level identifies different cellular responses dependent on nanomaterial concentrations

A mechanistic understanding of nanomaterial (NM) interaction with biological environments is pivotal for the safe transition from basic science to applied nanomedicine. NM exposure results in varying levels of internalized NM in different neighboring cells, due to variances in cell size, cell cycle phase and NM agglomeration. Using high-content analysis, we investigated the cytotoxic effects of fluorescent quantum dots on cultured cells, where all effects were correlated with the concentration of NMs at the single cell level. Upon binning the single cell data into different categories related to NM concentration, this study demonstrates, for the first time, that quantum dots activate both cytoprotective and cytotoxic mechanisms, resulting in a zero net result on the overall cell population, yet with significant effects in cells with higher cellular NM levels. Our results suggest that future NM cytotoxicity studies should correlate NM toxicity with cellular NM numbers on the single cell level, as conflicting mechanisms in particular cell subpopulations are commonly overlooked using classical toxicological methods.

Scientific RepoRts | 5:13890 | DOi: 10.1038/srep13890 of different mechanisms, for instance by induction of autophagy 9 , intracellular degradation and loss of toxic ions 19 , binding important signaling molecules (ligands/receptors) and hereby affecting both intraand intercellular communication 20 . An important factor in bio-nano interaction studies is the formation of the protein corona around NMs. The protein corona will determine how the NM will be presented to the cells when present in physiological conditions and hereby affect the final biological outcome of cellular NM exposure 21 . Recent studies have shown that the composition of the protein corona determines where the NMs will finally end up within the cells 10 . Therefore, various methods have been set up to enable quantitative profiling of the protein coronas 22 . Much work has been put into determining the physicochemical properties of NMs and how they influence the composition of the protein corona [23][24][25] .
Recently, it has also been shown that temperature plays a vital role in determining protein corona composition and cellular NM uptake 26 .
To date, NM toxicity studies are commonly performed in a manner similar to chemical toxicity studies, where for every parameter tested, a biochemical assay is used, providing a single representative value for the entire cell population. Dose-response curves are then generated by exposing cultured cells to a wide range of concentrations of NMs or chemicals. For chemicals, this has been proven to be a suitable procedure, to test their reactivity on cells, as they typically cross membranes more easily. For NMs, this procedure is more questionable as toxicity is mostly linked to the intracellular presence of NMs, apart from more rare events such as interaction with cell surface receptors or plasma membrane permeabilisation 27 . Cellular NM levels can however vary greatly and are dependent on the efficiency of endocytotic NM uptake. Various groups have therefore stressed the importance of determining cellular NM concentrations to accurately determine NM toxicity [28][29][30] , as various NM-related parameters, such as the nature of the NM coating, can influence NM toxicity as a secondary effect caused only by altered cellular NM uptake levels 28 . However, the currently used methods still link cellular effects to the average cellular NM level for the entire cell population, based, for instance, on colorimetric or inductively coupled plasma-mass spectrometry assays. Cellular NM levels have been shown to vary widely, even between closely neighboring cells 31 . Therefore, even though calculating average cellular toxicity and NM uptake levels are, to date, the most accepted methods for analyzing NM exposure yet these methods do not provide a comprehensive overview of all the processes involved in the cells of a specific population, rather they provide an average effect elicited by the NMs.
Averaging effects over a population cloaks distinct effects in multiple subpopulations. Many NM-elicited cellular responses can therefore be overlooked and only gross effects will be picked up. In order to provide a mechanistic understanding of how biological systems interact with NMs, it is therefore essential that individual cellular responses are collected. Hence, in this study, we employ a recently set up and validated methodology based on high-content imaging to evaluate a wide range of cellular responses to NM exposure 11 . We have modified the methodology to perform analysis on a single cell basis, where the consequences of exposure to fluorescent quantum dots (QDots) were correlated to the level of NM in the same cell. The data were then binned in 10 different categories, depending on the cellular QDot level, after which cellular parameters were analysed per category, to represent a substantial subcategory of the entire cellular population. At the same time, analysis was also performed without taking cellular NM levels into account, where the data provided by the two analysis methods can be immediately compared to one another.
Cell nanoparticle interaction studies. For high-content imaging studies, all cell types were seeded at 7500 cells/well in a 24 well plate (Nunc, Belgium). Cells were allowed to attach overnight in a humidified atmosphere at 37 °C and 5% CO 2 , after which the cells were incubated with the carboxylated QDots for 24 h in full growth medium. For cellular exposure studies, cells were incubated with the QDots at 2.5, 5, 7.5, 10, 12.5, 15 or 20 nM. For intracellular QDot distribution studies, MSCs were incubated with the QDots at 10 nM, while MEF cells were incubated with 12.5 nM QDots. For mechanistic studies, MEF, MEF Atg 5 KO or MEF Bax/Bak DKO cells were used and incubated with the QDots for 24 h at 12.5 nM. MSCs were either exposed to the QDots directly (10 nM) or were co-incubated with autophagy inhibitor 3-methyladenine (3-MA; 1 mM; Sigma-Aldrich, Belgium) or pan-caspase inhibitor Z-VAD-fmk (100 μ M, Bachem AG, Switzerland). Each experiment was performed in three independent repeats and Scientific RepoRts | 5:13890 | DOi: 10.1038/srep13890 data were analyzed using full data sets of the different repeats. The specific experimental procedures for the high-content imaging experiments are described in the Supplementary Information. Statistical analysis. All data are expressed as the mean + standard error to the mean (SEM). For all experiments, any statistical significance between a single condition and untreated control cells were analyzed using the t-test statistical method.

Average effects of quantum dots on cultured cells. The QDots used are commercially available
CdSe/ZnS core/shell nanoparticles (Invitrogen, Belgium) with maximal emission at 655 nm, which were coated using a carboxylated amphiphilic polymer, generating negatively charged NPs (Supp Section S1). When exposing mouse mesenchymal stem cells (MSCs) and mouse embryonic fibroblasts (MEFs) to these QDots, a clear endosomal localization can be observed, as the QDots are efficiently endocytosed and do not merely adhere to the plasma membrane (Fig. 1a, Supp Fig. S2), which is in line with other reports on carboxylated QDots 28 . Next, MSCs and MEFs were exposed to a series of QDot concentrations, and the effect on various parameters was evaluated ( Fig. 1b-d). The data reveal significant concentration-dependent effects of the QDots on cell viability (Supp Fig. S3  and MEF (right) cells that were transduced with CellLight Lysosomes-GFP (green) and subsequently exposed to 10 nM QDots (red) for 24 h. yellow/orange dots represent colocalized QDots and lyosomes. Scale bars = 25 μ m. The area in the white rectangle is shown as a magnified view below the image. (b) Heat maps of high-content imaging-based data for MSCs and MEFs exposed to varying concentrations of QDots and analyzed for; cell viability, cell membrane damage, mitochondrial ROS, size of the mitochondrial network, area of the cell, skewness of the cell, and level of autophagy. Data are shown as relative values after z-normalization compared to untreated control cells (= 1) where the foldchange is indicated by the respective color-code. Data have been acquired for minimum 5000 cells/condition which were gathered from three independent experiments. (c,d) Representative InCell high-content images of control MSCs or MSCs exposed to the QDots at 10 or 20 nM for 24 h, after which the cells were stained for (c) β -actin (green) or (d) mitochondrial ROS (green). Cells were counterstained with Hoechst nuclear stain (blue). Scale bars = 100 μ m, the area in the white rectangle is depicted in a magnified view below the original image. Effects of quantum dots on cultured cells at the single cell level. When considering average results over the entire visualized population, our data reveal that the effects the QDots induce in MSCs at 10 nM are similar to those observed in MEFs at 12.5 nM. At these concentrations, no significant effects could be observed, apart from the induction of oxidative stress, which does not appear to induce any cell death at these levels due to natural cellular defense mechanisms 32 . Exposure of both cell types to the QDots (MSC at 10 nM; MEF at 12.5 nM) however resulted in a broad distribution in cellular QDot levels ( Fig. 2), as is typical for in vitro cultured cells exposed to NMs 31 . NM levels between the two cell types were highly similar which explains the high similarity in their toxicity profiles. In order to evaluate the impact of cellular NM distribution on its resulting toxicity, the fluorescent properties of the QDots were exploited to determine relative cellular NM levels, and consequently every cellular parameter was linked to a subpopulation of cells with a certain NM level. NM distribution in cells followed a near-Gaussian profile and cells were subdivided into 10 different categories, category 1 being the lowest NM level and category 10 the highest (Fig. 2c-f). The data reveal that the majority of cells are in the medium categories (c4-c7; 68%). The lower categories (c1-c3) represent merely 12% of the entire population while the higher categories (c8-c10) represent 20% of the population. Interestingly, for MSCs exposed to 10 nM QDots and MEFs exposed to 12.5 nM QDots, the distribution of cellular QDot levels is quite similar, suggesting that the high similarity in cellular effects under these conditions (Fig. 1) is due to the similar levels of intracellular QDots. In general, MEF cells appear to be slightly more resistant to QDot-elicited cytotoxic effects as a secondary mechanism caused by the intrinsic lower internalization efficiency of these cells.
Following this categorization, the results showed a clear impact of cellular NM levels on both the nature and degree of the cellular response (Fig. 3). For cells containing low to medium QDot levels,  (a) Heat maps of high-content imaging-based data for MSCs and MEFs exposed to QDots at 10 and 12.5 nM concentrations, respectively, at which slight but non-significant effects had been observed for nearly all the tested parameters based on the total cell population. Cells were analyzed for; cell viability, cell membrane damage, mitochondrial ROS, size of the mitochondrial network, area of the cell, skewness of the cell, and level of autophagy. For every cell, the relative level of QDots was also calculated based on the cellular QDot intensity and area, after which cells were divided into 10 categories based on their cellular QDot levels, ranging from c1 (lowest) to c10 (highest). Observed cellular effects were grouped based on the different categories of intracellular QDot concentrations. The data are shown as relative values after z-normalization compared to untreated control cells (= 1) where the fold-change is indicated by the respective colorcode. Data were acquired for minimum 5000 cells/condition and were gathered from three independent experiments. (b,c) Representative InCell high-content images of (b) MSCs and (c) MEFs exposed to the QDots at 10 and 12.5 nM, respectively. Red: QDots, Blue: Hoechst nuclear stain, Green: dead cells (left), mitochondrial ROS (middle), β -actin (right). Scale bars = 100 μ m, the area in the white rectangle is depicted in a magnified view below the original image. no significant effects were observed for any of the assessed parameters. However, at higher intracellular NM levels (categories 8-10), significant effects were observed, which, however, did not all correlate with intracellular NM levels. Mitochondrial stress and ROS (Supp Figs S09, S10), cell morphology and skewness (Supp Figs S11, S12) and autophagy (Supp Fig. S13) showed higher significance with an increase in intracellular QDot levels, whereas the highest levels of cell death and associated membrane damage were seen in categories 8 and 10, but less pronounced in category 9 (Supp Figs S14, S15).

Mechanistic understanding of quantum dot-induced cytotoxic effects at single cell level.
NM-mediated autophagy has been described to be a general cellular response to exposure to a wide range of NMs, including QDots 9 . The physiological impact of NM-induced autophagy remains however somewhat unclear, as some studies have indicated clear cytoprotective effects by impeding cellular apoptosis signaling, while other studies showed direct autophagy-mediated cell death 33,34 . The induction of autophagy was therefore selected as the potential key mechanism to explain the lower correlation of intracellular NM levels and cell death. We tested two transgenic MEF cell lines, displaying either compromised autophagy, because of the deletion of the essential autophagy gene Atg5 (i.e. MEF-Atg5 KO) or mitochondrial apoptosis due to the double deficiency in pro-apoptotoc Bax and Bak (i.e. MEF-Bax/ Bak DKO), a defect known to blunt apoptosis to a variety of stress signals 35,36 . Along with these genetic approaches, we chemically inhibited autophagy (3-methyladenine (3-MA)) and apoptosis (Z-VAD-fmk) in MSCs. Figure 4 reveals clear involvement of apoptosis and autophagy in QDot-mediated cellular response. Inhibition of apoptosis did not affect the levels of cellular autophagy (Supp Fig. S16), whereas inhibition of autophagy resulted in a clear increase in cellular apoptosis levels at higher intracellular QDot concentrations (Supp Fig. S17). Interestingly, inhibition of either mechanism could not prevent cell death, yet in apoptosis-deficient cells, cell death only occurred at the highest intracellular QDot concentration, whereas in autophagy-deficient cells, cell death induction followed a clear concentration-dependent profile (Supp Fig. S18).

Discussion
In the present study, we have demonstrated a clear impact of intracellular NM concentration and NM distribution throughout the entire cell population on the cellular response to NM exposure. Upon using biochemical assays that commonly generate mean values for an entire cell population, more subtle effects that are only present in a certain subpopulation of the cells are normally overlooked, although they can have serious consequences and limitations for the final use of the cells. Biochemical assays are often used as they are generally easy to perform, amenable to high-throughput and quantitative. Imaging-based methods provide visual information on the impact of NMs, and are therefore generally qualitative. Some parameters, such as cell morphology can however only be properly studied using visual analysis. To overcome these limitations, several research groups have turned to the use of high content imaging 37 . One of the main advantages of high content imaging is the automated generation of a high number of images that can then be analyzed and turned into quantitative information over thousands of cells per condition 38 . The use of high content imaging has gained high interest from many fields, including cell biology and drug discovery as a powerful quantitative method that enables the generation of multidimensional (multiple parameters) and hierarchical data (i.e. in this study: all parameters linked to cellular QDot levels) 38 . Additional advantages of high content imaging are the ability to discriminate between real and off-target effects, non-specific fluorescence (i.e. interference of NMs), or the ability to study additional parameters (i.e. NM distribution).
The results obtained in this study are in line with previous reports on the toxicity of CdSe/ZnS QDots, coated with amphiphilic polymers, where clear effects on cell viability could be observed at concentrations ranging from about 2-20 nM 28,39,40 . This wide range of safe upper concentrations is mainly linked with the generation of disparate data due to differences in cell types, QDot surface chemistry and analysis method used. For cadmium-containing QDots, their toxicity has been mainly associated with the presence of free Cd 2+ ions, that are known toxic agents. The presence of Cd 2+ ions can result from remnants of cadmium after the synthesis and purification, where in a hydrophilic oxidative environment, some Cd 2+ will also leach from the QDot surface, as a result of the inequilibrium of Cd 2+ ions in solution and within the QDots. In a cellular environment, the QDots will be exposed to the degradative acidic microenvironment of the endosomes in which they will locate, which will further increase the loss of Cd 2+ ions due to acid etching of the surface 19,28,41 . The present work shows that the second mechanism is more predominant in affecting cellular viability, as toxicity is clearly linked to intracellular QDot levels and thus Cd 2+ ions being released intraendosomally. If toxicity was caused by free Cd 2+ ions in solution, then any toxic effects would be more homogeneously spread and not linked so closely to the intracellular QDot levels. This of course also depends on the quality and purity of the QDot sample, where "aged" QDots or QDots that have been poorly purified and contain high Cd 2+ levels in the stock suspension, may be more prone to toxicity induced by the higher levels of free Cd 2+ ions 42 .
A clear difference between the two cell types in terms of their ability to deal with NM exposure has been observed. The MSCs and MEFs show similar effects when exposed to 10 and 12.5 nM QDots, respectively. This has then be shown to result in highly similar intracellular levels of the QDots, thus showing that the intracellular level is more meaningful in terms of comparing different cell types than the concentration of NMs used for initial cell exposure. To date, nearly all studies however express NM toxicity in terms of exposure concentrations, which, as shown here, is less meaningful. The difference in cellular NM concentrations between different cell types can be due to a lower intrinsic endocytic capacity of one cell type compared to the other, or a smaller surface area and hence less potential interaction area for NMs with the cell surface. Other studies have made similar observations, where the concept of "cell vision" has been introduced as a key factor in nano-bio interactions. In short, "cell vision" signifies the intrinsic capacity of a particular cell type to deal with NMs, which is determined by the cell-specific uptake and defense mechanisms 43,44 .
For mechanistic studies, the overall effect of the NMs will dominate, which requires high NM concentrations. By linking cellular responses to the intracellular NM concentration on the single cell level, we have shown that at apparent sub-cytotoxic NM incubation concentrations, various conflicting mechanisms are evoked, such as apoptosis and autophagy. Taken together, these data suggest that apoptosis occurs at medium to high intracellular QDot concentrations, but is then impeded by autophagy at higher concentrations, resulting in a partial recovery of cell viability. At the highest intracellular QDot concentrations, autophagy levels were too high, resulting in autophagy-mediated cell death. One possibility is that the level of Cd 2+ ions present in the QDot incubation media, are responsible for the induction of apoptosis as the release of Cd 2+ has been shown to be a major cause of QDot toxicity 41 . Particularly for cell-internalized QDots, degradation has been observed to occur at a higher rate, resulting in the generation of free Cd 2+ ions. Cd 2+ ions are known to be highly toxic and a sustained cellular exposure to high levels of these ions (partially derived from the initial incubation mixture and partially from intracellular QDot dissolution) will result in cell death, through the process of apoptosis. The precise mechanism underlying the induction of autophagy remains somewhat unclear, but oxidative stress-induced mitochondrial damage, which is known to result in a particular form of autophagy, called mitophagy 45 , could be involved. Autophagy in itself is known to be a cytoprotective mechanism 46 and induction of Scientific RepoRts | 5:13890 | DOi: 10.1038/srep13890 autophagy is known to inhibit apoptosis signaling 47 . Many different types of NMs, including various types of Cd 2+ − containing ones, have been shown to induce autophagy, which either results in cytotoxicity or cytoprotection 33,34,48 . Our findings reveal a clear induction of apoptosis for higher levels of cellular NMs, which is however inhibited by the concurrent induction of autophagy. Together, this results in variable levels of cell death, where at low levels of autophagy, apoptosis is more pronounced and cell viability decreases. When autophagy levels increase, apoptosis is more inhibited, resulting in a recovery of cell viability. At high autophagy levels, cell viability again decreases, although apoptosis is still strongly inhibited. These data suggest that at the highest levels of autophagy, cell death occurs through direct autophagy-mediated signaling 49 .
These findings can explain the high level of disparate data that have been generated on NM toxicity, where small differences in cell types, exposure conditions or NM properties can result in changes in cellular NM levels, and hereby affect the balance between the different cellular responses. We have, for the first time, demonstrated a robust method to evaluate NM toxicity on high cell numbers with single cell resolution, allowing to delineate cellular responses in specific cellular subpopulations. The data here demonstrate the clear need for correlating cellular responses with intracellular NM concentration to enhance our understanding of NM toxicity and the precise mechanisms by which NMs can interact with biological components. Additionally, the findings reveal that data obtained from average population analyses overestimate the safe concentrations of NMs, due to the added cellular variability. Although single cell analysis was achieved, binning the data into the different categories provided a nicer and more achievable overview of substantial parts of the entire cell population, as revealed by the data in this study where significant effects are observed in nearly 20% of the entire population, but this goes unnoticed in the entire cell population. This knowledge is imperative for the future progress and clinical translation of nanotechnology and the methodology demonstrated here can serve as a template for future nanotoxicological analyses.