Changes in Caco-2 cells transcriptome pro ﬁ les upon exposure to gold nanoparticles

(cid:1) Biological effects (toxicity, uptake and changes in gene expressionpatterns) on Caco-2 cells of citrate-stabilized 5nm and 30nm AuNPs were compared. (cid:1) Exposure to 5nm AuNPs had much stronger effect on gene expression as compared to treatment with 30nm AuNPs. (cid:1) Nrf2 signaling stress response was among the highly activated pathways. Higher ef ﬁ cacy and safety of nano gold therapeutics require examination of cellular responses to gold nanoparticles (AuNPs). In this work we compared cellular uptake, cytotoxicity and RNA expression patterns induced in Caco-2 cells exposed to AuNP (5 and 30nm). Cellular internalization was dose and time-dependent for both AuNPs. The toxicity was observed bycolony forming ef ﬁ ciency (CFE) and not by Trypan blue assay, and exclusively for 5nm AuNPs, starting at the concentration of 200 m M (24 and 72h of exposure). The most pronounced changes in gene expression (Agilent microarrays) were detected at 72h (300 m M) of exposure to AuNPs (5nm). The biological processes affected by smaller AuNPs were: RNA/zinc ion/transition metal ion binding (decreased), cadmium/copper ion binding and glutathione metabolism (increased). Some Nrf2 responsive genes (several metallothioneins, HMOX, G6PD, OSGIN1 and GPX2) were highly up regulated. Members of the selenoproteins were also differentially expressed. Our ﬁ ndings indicate that exposure to high concentration of AuNPs (5nm) induces metal exposure, oxidative stress signaling pathways, and might in ﬂ uence selenium homeostasis. Some of detected cellular responses might be explored as potential enhancers of anti-cancer properties of AuNPs based nanomedicines. .


Introduction
Nanotechnology and nanomedicine bring novel approaches into clinics, revolutionizing diagnosis and treatment options for diverse groups of patients. Engineered nanomaterials (ENMs) are also widely used in other consumer products and applications increasing the likelihood of human exposure to nanomaterials (Staggers et al., 2008). The exposure to ENMs can take place not only during their synthesis, production and usage, but also at other stages of the life cycle (e.g., waste deposition/combustion, material recycling) of a given nanomaterial. Therefore, in order to assure the safety of human population and environment, toxicological profiling of cellular responses to ENMs and proper risk assessment of nanomaterials should be performed (Maynard, 2012). The (nano) toxicological hazard identification and characterization are also required for improvement of the safety and efficacy of nanomedicines, including nano gold based formulations used in biomedical applications. Moreover, studies focused on the consequences of the long-term exposures to ENMs, including gold nanoparticles (AuNPs), are needed (Kunzmann et al., 2011).
Gold belongs to the group of noble metals and has been used in medical applications for centuries (Bhattacharya and Mukherjee, 2008). In recent years, nano gold based applications were designed and AuNPs synthesized with the idea that noble metal characteristics (in bulk) of gold would be preserved even when synthesized and used at nano scale, thus being bio-compatible (Connor et al., 2005). AuNPs and their derivatives undergo continuous development for their use in clinical diagnostics, as drug delivery systems or therapeutic agents (Dykman and Khlebtsov, 2012). In general, AuNPs are well tolerated in vivo (Hainfeld et al., 2006), although some toxicity of AuNPs was also reported in vitro (Pernodet et al., 2006;Chuang et al., 2013;Coradeghini et al., 2013;Mironava et al., 2014).
As methodologies develop, "omics" based research platforms can complement the classical tools for cytotoxicity testing in (nano) toxicology field. Therefore, identification and characterization of biological responses can be achieved at molecular levels compromising biomarkers/endpoints that can include whole or parts of transcriptome, proteome, metabolome, epigenome and/or genome of the in vitro/in vivo system under investigation (Hamadeh et al., 2002). Although toxicogenomics and other "omics" methods are currently easily accessible and cost effective, to date only a few studies applied transcriptomics or proteomics tools in the study of AuNPs induced/mediated cellular responses (Esther et al., 2005;Yang et al., 2010;Li et al., 2011;Qu et al., 2013;Gioria et al., 2014).
All living organisms respond to diverse stress stimuli. Depending on the biological context of given exposure/stressor, its severity and duration, cells within tissue/organ react to stimuli by evoking stress or adaptation responses, or die (Fulda et al., 2010;Chovatiya and Medzhitov, 2014). One of the key regulators of cellular responses to endogenous and/or exogenous stress is the nuclear factor E2-related factor 2 (Nrf2) protein. Nrf2 is a redox and xenobiotics sensitive transcriptional factor, assuring the expression of proteins involved in cellular adaptation to oxidative stress, adjustment of metabolism and detoxification of drugs (Kensler et al., 2007). Under normal cellular conditions, Nrf2 functions as a tumor suppressor. However, it has been shown that Nrf2 might also act as an oncogene (Shelton and Jaiswal, 2013). Among gene targets regulated by Nrf2 are metallothioneins (MTs), which are small, cysteine rich and metal ion binding proteins. MTs play an important role in the protection of cells against metal toxicity, but are also involved in responses to oxidative stress, and exposure to glucocorticoids and cytokines (Miles et al., 2000).
The aim of this study was to investigate how the colorectal adenocarcinoma epithelial cell line , an in vitro model of the intestinal route of exposure to nanoparticles, responds to AuNPs treatments at RNA gene expression level. We selected undifferentiated cells mainly because in this way we looked at populations of heterogenic pool of Caco-2 cells prior to differentiation, therefore having wider plasticity in the cellular physiology, sensing and adaptation/response to stress, as compared to mature and polarized enterocytes (Tadiali et al., 2002). Keeping this in mind, when running both the cytotoxicity and RNA transcript profiling experiments, we explored Caco-2 cells (in undifferentiated stage) as potentially more suitable model for detecting alterations in gene expression patterns upon exposure to NPs. For that purpose, we tested citrate-stabilized spherical AuNPs of two sizes (5 nm and 30 nm, synthesized and characterized in-house) for their cytotoxic potential, internalization and induction of changes in mRNA and long non-coding RNAs expression patterns using transcriptomics approach (Agilent microarrays platform). The microarray data were further validated with quantitative real-time reverse-transcriptase PCR (qPCR) for a set of mRNAs which were differentially expressed upon exposure of Caco-2 cells to AuNPs. Several biological pathways affected by treatment of Caco-2 cells with AuNPs (5 nm, 300 mM, 72 h) were identified in microarray datasets using bioinformatical tools. The novelty of this work is the integration of classical in vitro testing with the chemico-physical characterization of NPs, Omics and advanced analytical techniques to better understand the mechanisms of potential nanomaterials' toxicity. Their use has the potential to provide a description of the bio-responses and allows highlighting critical biochemical pathways affected by nanoparticle exposure. Huge progresses have been made in the genomic field and here we integrated this discipline to investigate in detail modifications caused by AuNPs exposure at gene level. The results of this work established a starting point in understanding and disseminating which cellular processes can be linked to the observed cytotoxic effect of 5 nm AuNPs on cancer Caco-2 cells.

AuNPs synthesis and characterization
AuNPs of approximately 5 nm and 30 nm size in diameter were synthesized and concentrated as described in literature (Coradeghini et al., 2013;Turkevich et al., 1951). Detailed description of the chemicals used for AuNPs synthesis, description of the equipment used for particles size distribution and characterization, sample preparation, image acquisition and processing, can be found in Supplementary methods.
The stability of AuNPs dispersions was studied both using a centrifugal particle sedimentation (CPS) instrument and by zeta-potential measurements in representative media. Citratestabilized AuNPs (5 and 30 nm) were incubated at two different concentrations (100 and 300 mM) in three different media (milliQwater, serum-free cell culture medium and complete cell culture medium). Measurements were made immediately after mixing AuNPs samples with the corresponding medium (time 0) and then again after 24 and 72 h incubation times at 37 C.
AuNPs dispersions for cell exposure experiments were freshly prepared by diluting the AuNPs suspensions (after concentration step, as described in the Supplementary methods) in complete culture medium, and were added directly to Caco-2 cell cultures. The concentration step produced two solutions containing all the reagents used for the NPs synthesis and they were used as solvent controls for biological testing.

Cell culture
Human epithelial colorectal adenocarcinoma cells (Caco-2) are from the European collection of cell cultures (ECACC) and were purchased from Sigma (Catalogue number: 86010202; passage 45). Experimental cultures were prepared from deep-frozen stock vials and maintained in a sub-confluent state. They were grown in complete cell culture medium composed of high glucose (4.5 g/mL) Dulbecco's modified Eagle medium (DMEM), supplemented with 10% (v/v) fetal bovine serum (FBS), 4 mM L-glutamine and 1% (v/v) pen/strep (all from Invitrogen, Italy). Cultures were maintained in cell culture incubator (HERAEUS, Germany) under standard culture conditions (37 C, 5% CO 2 and 95% humidity). Caco-2 cells used in this work were screened for the presence of Mycoplasma with quantitative PCR (qPCR) using the Venor 1 GeM Prime kit and were Mycoplasma free.

Cytotoxicity assays
Cytotoxic effect of AuNPs in Caco-2 cells was studied by colony forming efficiency (CFE) and Trypan blue exclusion assays.
For the CFE, cells were seeded at the density of 200 cells per 60 Â 15 mm dish (Corning, Italy) in 3 mL complete culture medium (three replicates) for each treatment. Twenty four hours after plating, the medium was changed and 3 mL of fresh medium containing the AuNPs was added to obtain the final AuNP concentrations of 10, 50, 100, 200 and 300 mM corresponding to 0. 02, 9.85, 19.70, 39.40 and 59.10 mg/mL, respectively. After 24 and 72 h of exposure, the medium was replaced with fresh complete culture medium. On the 13th day from seeding, cells were fixed with 3.7% (v/v) formaldehyde solution (Sigma-Aldrich, Italy) in phosphate-buffered saline (PBS) without calcium, magnesium and sodium bicarbonate (Invitrogen, Italy) and stained with 10% (v/v) Giemsa GS-500 solution (Sigma-Aldrich) in ultra pure water. Solvent controls (cells exposed to two different solvents obtained after AuNPs of 5 nm and 30 nm filtration, as used for the highest concentration tested) and a positive control (1 mM) sodium meta chromate (Sigma-Aldrich) were included. Colonies were scored using a cell colony counter GelCount (Oxford Optronix Ltd., UK).
For Trypan blue exclusion assay, 2.5 Â 10 5 and 0.9 Â 10 5 cells were seeded for 24 h and 72 h exposures, respectively, in 6 well plates (Falcon, Italy) with 2 mL of complete cell culture medium. Twenty four hours after plating, cells were exposed to 50, 100 and 300 mM of 5 and 30 nm AuNPs, corresponding to 9.85, 19.70, and 59.10 mg/mL. At the end of the exposure time (24 or 72 h), cells were washed twice with PBS, detached with 0.5 mL of 0.05% trypsin sodium ethylene diamine tetraacetic acid (Trypsin-EDTA) from Invitrogen, and harvested with 1 mL of complete culture medium. Next, 30 mL of each sample was stained with 30 mL of Trypan Blue (Sigma-Aldrich) and cells were counted with the TC10 automated cell counter (Biorad, Italy) according to the supplier's protocol. Negative control, solvent control and positive control were also included in the test, as described for CFE assay.
In the CFE and Trypan blue assays, the results were normalized to the solvent control. For the CFE assay, they are expressed as CFE (%) ([average number of treatment colonies/average number of solvent control colonies] Â 100). For the Trypan blue assay, they are expressed as viability (%) ([number of cells in treatment/number of cells in solvent control] Â 100). One-way analysis of variance (ANOVA) with post-hoc test (Dunnett's multiple comparison test) for comparing groups of data against one control group was used. Data are reported as mean values AESEM (standard error mean = standard deviation/ p number of replicates). The data represents mean value of three independent biological replicas, where p values of less than 0.05 (*) and 0.001 (***) were considered statistically significant. Statistical calculations were carried out using GraphPad Prism Version 5.0 (GraphPad Software, USA).

Nanoparticle internalization
The uptake of AuNPs, their association with cellular membrane/ matrix and internalization were quantified by Inductively Coupled Plasma Mass Spectrometry (ICP-MS). The uptake of AuNPs, association of AuNPs with cellular membrane/matrix and the internalization of AuNPs was quantified by ICP-MS. For the ICP-MS, 7 Â 10 5 and 3 Â 10 5 CaCo-2 cells were seeded in dishes from Corning (100 mm Â 15 mm) for 24 and 72 h exposure time respectively, in 5 mL of complete culture medium. Twenty four hours after seeding, cells were treated for 24 and 72 h with 5 and 30 nm AuNPs (100 and 300 mM). At the end of the exposure time, the medium was removed and collected for each sample, cells were washed twice with PBS and each wash collected in a separate tube. Cells were then detached with 1 mL of Trypsin-EDTA (Invitrogen, Italy) and harvested with 5 mL of complete culture medium. After this, cells suspension was centrifuged at 200 Â g for 5 min (4 C), and the supernatant was transferred into a new tube. The cellular pellet was suspended in 1 mL of culture medium. An aliquot of each sample (20 mL) was taken for cell count with TC10 automated counter. The total content of Au in the cells (cellular pellet) as well as in all the fractions collected from each cell culture sample: (i) medium at the end of the treatment (24 and 72 h), (ii) supernatants from the two washes, (iii) cellular pellet, and (iv) the final supernatant (representing a third wash), were analyzed by ICP-MS after mineralization with Aqua Regia (HNO 3 :HCl at 1:3 ratio) and microwave digestion (2 cycles at 950 W for 10 min) in a Discover-Explorer microwave instrument (CEM Corporation, USA). These extensive washing of cells and sample collection steps for Au content measurements were applied to remove weakly bound AuNPs and to assure that the total starting amount of Au given to the cells was recovered among different fractions (including both, strongly associated with and/or internalized by cells and not taken up by cells). Three independent experimental replicas were carried out and the results were expressed as pg Au/cell, % of the external exposure concentration and as a number of NPs/cell. Calculations were performed as described in Coradeghini et al. (2013).

Exposures to AuNPs and RNA extraction
For the gene expression experiments, 2.5 Â 10 5 and 0.9 Â 10 5 Caco-2 cells were seeded in 2 mL complete cell culture medium, in a 6 well/plate (Falcon, Italy), and incubated for 24 and 72 h with 5 nm or 30 nm AuNPs, at either 100 and 300 mM concentrations. At the end of the treatment, the exposure medium was removed and cells were washed twice with PBS, and harvested with 300 mL of RLT plus RNeasy lysis buffer (Qiagen, USA). Cell lysates were collected from wells and stored at À80 C until the RNA extraction step was performed. The details of RNA isolation are available in Supplementary methods.

Microarray expression profiling
The microarray (Agilent Whole Human Genome Oligo Microarray: 4 Â 44 k 60 mer slide format) experiments were designed to perform three biological replicates for each time point (24 and 72 h), AuNPs size (5 and 30 nm) and concentration (100 and 300 mM) with complete medium and solvent controls. All the cRNA synthesis/sample-labeling, hybridization, washing, and scanning steps were conducted following the manufacturer's specifications (Agilent Technologies Inc., USA). Procedures for microarrays based data generation are described in Supplementary methods.

Quantitative PCR validation of microarray data
A real-time quantitative PCR (qPCR) analysis has been done on the same RNA samples that were used for the microarray hybridization, with the aim to validate the microarray results. All reagents for the qPCR and equipment were from Applied Biosystems, USA. The details of qPCR validation process and data analysis using delta-delta Ct method (Livak and Schmittgen, 2001) are available in Supplementary methods.

Gene expression data analysis
Quality control and array normalization were performed in the R statistical environment using the Agi4x44PreProcess (v 1.18.0) package, downloaded from the Bioconductor web site (Gentleman et al., 2004). The normalization and filtering steps were based on those described in the Agi4x44PreProcess reference manual. Briefly, Agi4x44PreProcess options were set to use the mean signal and the BG median signal as foreground and background signals, respectively. Data were normalized between arrays by the quantile method (Bolstad et al., 2003). In this approach, the distribution of intensities of different microarray chips are transformed to become equal so that intensities can be compared between each other. In order to detect expression changes among different treatment conditions, the moderated t test was applied. Moderated t statistics were generated by Limma Bioconductor package. Modulated genes were chosen as those with a fold change greater than 1.5 log 2 fold change and a false discovery rate (Benjamini and Hochberg's method) corrected p-value smaller than 0.05 (Smyth, 2004). Microarray results have been submitted to NCBI's gene expression omnibus (GEO) repository and are available under accession number GSE55349.

Bioinformatics
Up regulated and down regulated genes were analyzed in the Kyoto Encyclopedia of Genes and Genomes (KEGG) database (http://www.genome.jp/kegg/) in order to identify genes with similar functions. Expression analysis systematic explorer (EASE) biological theme analysis was conducted online using DAVID bioinformatics resources server (http//david.abcc.ncifcrf.gov/). Phylogenetic tree creation (rVISTA), co-expression network visualization (FunCoup) and in silico miRNA binding sites prediction (DIANA-mirExTra) are described in Supplementary methods (Alexeyenko et al., 2011;Alexiou et al., 2010;Dubchak et al., 2013).

Characterization of AuNPs
Fig . 1A and B shows representative TEM images of the two nano gold samples used in this study. The morphologies observed were mainly spherical, although for 30 nm AuNPs it could also be observed a certain degree of faceting. Size distributions determined by TEM image analysis were narrow 4.8 AE 1.0 and 32.2 AE 11.0 nm, and the absorption band maximum measured by UV-vis spectrometry were in agreement with expected diameters of 5 nm and 30 nm AuNPs (Fig. 1C). It has been shown that the position of the surface plasmon resonance peak of Au NPs can be used to determine both size and concentration of gold nanoparticles (Haiss et al., 2007). Dynamic light scattering size measurements were also in agreement with the rest of characterization techniques (Suppl. Fig. 1) showing a mean size distribution, expressed as intensity (%), of 7.5 nm and 37.3 nm for 5 nm and 30 nm AuNPs, respectively.
Disc centrifuge sedimentation analysis using CPS technology was considered as an appropriate technique to monitor and measure the size distributions of AuNPs suspension samples in water, culture medium without serum and in complete cell culture medium, supplemented with 10% of serum (Table 1). The particles size distributions in water when measured by CPS are in agreement with those determined by TEM image analysis. As expected, in the absence of serum in cell medium, AuNPs aggregated, showing increase in AuNPs diameters. Aggregation was more noticeable at higher concentrations and for longer incubation times. However, in complete cell medium supplemented with serum proteins, no aggregation was observed and the NPs size was slightly larger to the one obtained in water due to the formation of a protein corona.
Zeta-potential measurements (Suppl. Fig. 2) were also performed in order to determine the evolution of the AuNP surface charge after incubation in cell culture media (with or without serum) and in water. Zeta-potential of AuNPs showed initial values between À30 and À40 mV in water. After incubation with cell culture medium in the presence of serum proteins, for both sizes (5 and 30 nm) and concentrations (100 and 300 mM) of AuNPs, a sudden change of the negative value of zeta-potential already occurring after mixing (time 0), is observed. At longer incubation times, zeta-potential values evolved toward the average charge of serum proteins (zeta-potential about À13.5 mV in our experimental system), which is indicative of the absorption of proteins at the surface of the NPs.
Finally, the concentration of ions released from the AuNPs suspensions, when measured by ICP-MS for up to 72 h in complete culture medium, was for both sizes of AuNPs below the detection limit of the technique (<1 ppb).

Cytotoxicity
Nano gold cytotoxicity in Caco-2 cells was evaluated using two methods: CFE assay and Trypan blue exclusion assay. The CFE is a standard test, already optimized for studying the toxicity of NPs (Ponti et al., 2010) and which is undergoing an inter-laboratory comparison of performance and reproducibility testing in the frame of the OECD working party of manufactured nanomaterials (work in progress). Cells were exposed to 5 and 30 nm AuNPs for 24 and 72 h, at concentrations ranging from 10 to 300 mM, as described in Supplementary methods. Statistically significant cytotoxicity was observed only for 5 nm AuNPs when tested by CFE ( Fig. 2A and B). Under the same exposure conditions, small decrease in cell viability at 100 and 300 mM was detected by Trypan blue assay. Although the observed drop in the membrane integrity was statistically not significant ( Fig. 2C and D), its biological effects as a results of exposure of Caco-2 cells to 5 nm AuNPs, in particular to the highest concentration tested (300 mM), were evident when looking at detected changes in gene expression patterns. Interestingly, the cytotoxicity end-points based on changes in the membrane permeability detected by Trypan blue assay or efficiency in colony formation observed in CFE, were not statistically significant in Caco-2 cells exposed to 30 nm AuNPs, even for the highest concentration (300 mM) and at the longest (72 h) exposure time tested (Fig. 2). No significant difference in the number of colonies scored in the negative controls was observed in comparison to the solvent controls. As expected, for both in vitro cytotoxicity assays, treatment of Caco-2 cells with 1 mM sodium meta chromate (positive control) resulted in complete cell death (data not shown).

Nanoparticles internalization
The interaction of 5 and 30 nm AuNPs with Caco-2 cells was studied and quantified by ICP-MS after exposure to 100 and 300 mM nano gold suspensions (Fig. 3).
Internalization of gold in Caco-2 cells is expressed as pg/cell of Au, as % of Au vs total or number of NPs/cell. After 24 and 72 h of exposure at both 100 or 300 mM of Au 5 nm, cells incorporated approximately the same amount of Au; while we observed a dose-dependent increase of the Au 30 nm internalisation after both 24 and 72 h of exposure (Fig. 3A).
Interestingly, when the results are expressed as % of Au vs total exposure, for both AuNPs 5 and 30 nm a time-dependent increase in cell interaction and uptake was observed (Fig. 3B); and when expressing the data as a number of NPs in each cell the internalization of AuNPs 5 nm seems to be more efficient as compared to AuNPs 30 nm and no dose or time-dependent uptake manner was observed (Fig. 3C). Table 1 Mean size distribution of AuNPs as measured by centrifugal sedimentation (CPS). Mean sizes of AuNPs samples (5 and 30 nm) incubated in water, Caco-2 cell culture medium without and with 10% (v/v) serum for 0, 24 and 72 h. Incubations took place in a cell culture incubator (in the dark). Abbreviations -HW: half width; PdI: polydispersity index; nd: not determined.

Gene expression profiling
As the major goal of our work was to get a better understanding of which molecular pathways might be affected by gold nanoparticles, we investigated the gene expression profiles of Caco-2 cells treated for 24 and 72 h with 5 and 30 nm nano gold, at two concentrations (100 and 300 mM), and compared them with the RNA transcripts levels of untreated cells (negative control).
First, a comparison of gene expression in cells treated with solvent (solvent control) vs the negative control was run and since almost no change was observed in the solvent control of Caco-2 cells (data not shown), we used untreated cells as reference control for gene expression profiling experiments. Thereafter, an induction or repression greater than 1.5 log 2 fold change (that corresponds to an increase/decrease of 50%) and with a false discovery rate (FDR) corrected p value smaller than 0.05, were used to compare control data sets vs different treatment conditions for AuNPs exposures.
At the lower concentration of AuNPs (100 mM), a minimal number of differentially expressed genes was found. There were no gene transcripts that were regulated by the 5 nm AuNPs neither at 24 nor at 72 h. However, there were four mRNAs induced in Caco-2 [ ( F i g . _ 2 ) T D $ F I G ] Fig. 2. Cytotoxic effects of 5 nm and 30 nm AuNPs on Caco-2 cells. The cytotoxicity of 5 nm AuNPs (black bars) and 30 nm AuNPs (gray bars) were estimated by colony forming efficiency (CFE) and Trypan blue exclusion assays. CFE assay: Caco-2 cells were exposed to increasing concentrations (10-300 mM) of AuNPs for 24 (A) and 72 h (B). In this range of concentrations and time points tested, no cytotoxicity was found in Caco-2 cells exposed to 30 nm AuNPs; while statistically significant cytotoxicity was observed for 5 nm AuNPs after 24 and 72 h of exposure to 200 mM (*p < 0.05) and 300 mM (***p < 0.001). Trypan blue exclusion assay: cell viability was tested on Caco-2 cells exposed for 24 (C) and 72 h (D) to increasing concentrations of 5 and 30 nm AuNPs (50-300 mM). Results of both assays represent a mean of three independent experiments (three replicates each) AE standard error of the mean (SEM) and are expressed in (A) and (B) as CFE (% of solvent control), while in (C) and (D) as viability (% of solvent control), respectively. cells exposed to 30 nm nano gold (100 mM) after 24 h, and none of these genes was further affected after 72 h (Fig. 4A). Among these genes which mRNAs are increased, we find neurotensin receptor 2 (NTSR2) and two zinc finger containing transcriptional factors: human immunodeficiency virus type I enhancer binding protein 2 (HIVEP2) and zinc finger, DHHC-type containing 11 protein (ZDHHC11) involved in palmitoylation of proteins (Evers, 2006;Fukuda et al., 2002;Oku et al., 2013). Moreover, a gene locus LOC644662, which encodes yet uncharacterized novel long intergenic non-coding RNA (Zhang et al., 2011) (TSC22D3) and cysteine-rich PAK1 inhibitor (CRIPAK). The CDH16 gene product is a member of the cadherin family of cell adhesion, calcium dependent, membrane associated glycoproteins. Another mRNA induced by larger in size nano gold particle tested (30 nm), is a gene product of CPA2, which encodes a secreted protein involved in catabolic and digestion processes of other proteins (Vendrell et al., 2000). The third differentially expressed gene TSC22D3 encodes a leucine zipper transcription factor. Several studies showed that TSC22D3 gene, also known as GLIZ/DIP/TSC-22R, is induced by glucocorticoids (GCs), and plays a very important role as mediator in the anti-inflammatory and immunosuppressive action of GCs (Ayroldi and Riccardi, 2009). The fourth up regulated gene, CRIPAK is a novel endogenous inhibitor of p21-activated protein kinase 1 (Pak1). This kinase plays an important role in cytoskeleton organization, promotion of the cell survival responses and estrogen receptor (ER) mediated signaling (Talukder et al., 2006). Among these four genes, the mRNA of TSC22D gene was also down regulated in response to 5 nm AuNPs exposure (300 mM, 72 h). Thus, the other three genes (CDH16, CPA2, CIRPAK) showed decrease in their corresponding mRNAs levels only when the Caco-2 cells were treated with 30 nm nano gold particles.

Validation of microarray data with PCR
A subset of genes identified during microarray profiling was chosen for validation with real-time quantitative PCR (qPCR). The target selection of mRNAs (12 targets) for validation study was based on manual screening of differentially expressed genes, with an aim to select mRNAs covering diverse biological and biochemical functions which might be altered in Caco-2 cells when exposed to AuNPs.

[ ( F i g . _ 4 ) T D $ F I G ]
Several members of genes involved in responses to oxidative stress, metal exposure and changes in cellular redox status were chosen for qPCR validation. To this group of validation targets belong mRNAs of the following genes: oxidative stress induced growth inhibitor 1 (OSGIN1), HMOX1, MT2A and GPX2 (Gozzelino et al., 2010;Li et al., 2006;Vašák and Meloni, 2011;Wingler et al., 1999). As the data from the CFE assay indicated cytotoxic effect of 5 nm AuNPs (300 mM, 72 h), we included for qPCR validation two genes encoding regulators of apoptosis: the inhibitor of apoptosis baculoviral IAP repeat containing 2 (BIRC2) and the apoptosisinducing, DNA binding C1D protein (Dubrez-Daloz et al., 2008;Rothbarth et al., 1999).
When selecting mRNAs for validation, we also took into consideration the trends in gene expression (up regulation and down regulation) observed in microarray assays, therefore genes which were up (4) and down (8) regulated were included in the qPCR validation test ( Table 2). As internal reference genes (RGs), the mRNA transcripts of the c-abl oncogene 1 (ABL1), a non-receptor tyrosine kinase and a mitochondrial ribosomal protein L19 (MRPL19) were used. Under given experimental conditions, these RGs showed to be not affected by exposures of Caco-2 cells to neither size/concentration/exposure time of studied AuNPs (data not shown).
Generally, as shown in Table 2, gene expression levels and trends of regulation (increase/decrease) detected with microarrays vs detection and quantification based on qPCR are in agreement with each other. However, expression levels of mRNAs of HAT1 and SRSF10 were lower when detecting them with qPCR, indicating that for these RNA transcripts, the microarray data (300 mM, 72 h) over estimated the levels of their down regulation. On the contrary, for gene BIRC2, HAT1 and MT2A, the qPCR was more sensitive in detecting the alterations of these particular mRNA expression levels (100 mM, 24 and 72 h), where microarrays data was not indicating their decreased RNA transcripts abundances (Table 2).

Bioinformatics
After validation of transcriptomics data with qPCR, we proceeded with a search for significantly enriched gene classes among differentially expressed genes, as defined by both gene ontology (GO) annotation and KEGG. This was applied to data sets obtained from microarray profiling experiments of Caco-2 cells treated with 5 nm AuNPs (300 mM) and harvested at 24 and 72 h after exposure.
The genes that were down regulated at 24 h time point, show an enrichment of categories related to transcription co-repressor/cofactors activities and transcription factor binding, binding of Table 2 Validation of selected mRNAs with qPCR. Fold change expresses the difference of the mean log control and mean log stimulated data. Genes identified as regulated (log 2 fold greater than 1.5 and with false discovery rate (FDR) corrected [ 2 6 _ T D $ D I F F ] p value smaller than 0.05) are coloured in red (down regulated) and blue (up regulated). The cells in the table are coloured in [ 4 _ T D $ D I F F ] gray, when log 2 fold change is greater than 1.5 but the [ 2 6 _ T D $ D I F F ] p value is not significant.  Table II). These groups of GO/KEGG categories contain proteins encoded by, e.g., genes of STAU2, SRP9, RNF219, BIRC2, DNAJB9, DNAJC21, C1D, DCP2. As shown in Table 5, the up regulated genes at 72 h after exposure of Caco-2 cells to 5 nm AuNPs (300 mM) turned out to be significantly enriched in GO biological process classes mainly related to ion (Cd and Cu) binding, amino acid/amine transmembrane transporter activity, peptide antigen binding (several MTs and SLC family members) and KEGG pathways, represented by GPX2 and G6PD, involved in the glutathione metabolism. The top highly induced transcripts with 3.55 log 2 fold change (increase in the mRNA levels of about 12 times) belong to genes encoding metallothioneins (MTs). In humans, there are at least 18 genes encoding four distinct isoforms of MTs proteins: MT1, MT2, MT3 and MT4 (Laukens et al., 2009). The sequence homology grouping of MTs mRNAs is shown in Suppl. Fig. 3A. For the construction of phylogenetic trees, all members of the human MT gene family were plotted and analyzed against the mRNA of MT2A as a reference sequence. The MTs which mRNAs were induced in Caco-2 cells treated with 5 nm AuNPs  Fig. 3B), with the most highly up regulated gene being the MT2A. Noteworthy, the increased mRNA level of MT2A gene detected in the microarray assay was confirmed with qPCR (Table 2).
When we analysed functions of genes which were among the highly down regulated (Suppl. Fig. 3C), the gene encoding selenoprotein T (SELT) was noticed once more. The SELT attracted our attention because another selenoprotein, namely GPX2 was highly up regulated (Tables 2 and 5) under the same exposure conditions (5 nm AuNP, 300 mM, 72 h).
The data analysis for information on mRNA co-expression and protein-protein interaction of highly up-regulated genes (MT2A, GPX2 and G6PD) and their connections with oxidative stress pathway [Hs_Oxidative_Stress_WP408_38774] revealed that MT2A is connected with 1 gene, while G6PD is connected with 5 genes belong to this biological network (Suppl. Fig. 4A). Notably, the MT1X gene, connected with MT2A, is also up regulated by 5 nm AuNPs (300 mM, 72 h). When looking at interactomes of most significantly up regulated and down regulated genes, the following was found: (i) in the set of genes which were up regulated, grouping of MTs genes with their mRNA and protein co-regulation was observed (Suppl. Fig. 4B); (ii) the density of mRNA/protein co-expression networks of up regulated genes was not as dense as for genes which were significantly down regulated (Suppl. Fig. 4C) and showed strong connections between each other.
The observation of a high number of down regulated RNA transcripts in Caco-2 cells exposed to 5 nm AuNPs, in particular at 72 h time point (300 mM), raised a question: at which level the repression of genes might have occurred? With the microarrays data in hand and with the freely available DIANA mirExTra tool for predicting potential microRNAs (miRNAs) binding targets sequences in differentially expressed mRNAs, we decided to run an in silico experiment. In this way, we were able to probe 25 most significantly down regulated genes, in parallel with 50 not affected by AuNPs (control genes), for analyzing ad hoc potential epigenetic regulators of gene expression, at miRNA-mRNA interface levels (Flynt and Lai, 2008). The results of the prediction test are presented in Suppl. Table III and in Fig. 5.
We found that, the following miRNAs: miR 340, miR 181a, miR 410 and miR 520 d 5p (Fig. 5A), are being frequently identified by Table 4 GO and KEGG enrichment of altered genes (down regulated) by 5 nm AuNPs treatment (300 mM, 72 h). The p value refers to how significant an association of a particular term has with the gene list. Where there are more than 10 genes regulated by AuNPs treatment, then only ten most regulated ones are displayed.

Term
Count P value Genes GO: 0003723 -RNA binding  37  0.0002 ZCRB1, KIN, STAU2, SNRPB2, DCP2, SRSF3, TRUB1, C1D, EIF1AY, SRP9,  GO:0008270zinc ion binding  82  0.0048 RNF219, ZNF124, MOBKL3, PTS, ZFAND6, THAP1, BIRC2, RFK, Table 5 The p value refers to how significant an association of a particular term has with the gene list. the DIANA mirExTra algorithm and might represent potential mediators of down regulation of several mRNAs in the analyzed data set (Fig. 5B). Interestingly, the most down regulated genes (ZCCHC10, MAD2L1, C13orf27, CNPK, COMMD8) were not among the best targets for miRNAs mediated post-transcriptional gene regulation in the given experimental data set prediction context (Suppl . Table III, Fig. 5B). On the contrary, gene encoding selenoprotein SELT has 12 predicted binding sites for miRNAs, in the same number range for predicted miRNA binding sites (10-17), as the other down regulated genes have (PLDN, TMEM38B, ARL6IP6, C14orf129 and GOLT1B). The gene that has the highest number of predicted miRNAs binding sites (27), encodes mesoderm induction early response 1, family member 3 protein (MIER3).

Discussion
Several factors are involved in determining cytotoxicity of engineered nanomaterials (ENMs). Our work focused mainly on comparing cytotoxic potentials, in the context of inhibitory effect on colony forming efficiency (CFE) with changes in gene expression patterns (transcriptomics) induced in Caco-2 cells exposed to 5 nm and 30 nm citrate-stabilized AuNPs. Chemico-physical characterization tests of AuNPs used in this study confirmed that AuNPs were of expected size/shape and behaved in the experimental media as previously reported by Coradeghini et al. (2013).
In this work we selected undifferentiated Caco-2 cells an in vitro model of the intestinal route of exposure to nanoparticles. When Caco-2 cells are cultured over confluence for 21 days, these cells undergo spontaneous cell cycle arrest and differentiation. In his process, the enterocyte-like monolayer is formed serving as a model of the intestinal barrier (Sambuy et al., 2005;Natoli et al., 2012). Due to the intrinsic heterogeneity of the original parental cell line and culture-related conditions upon differentiation the expression of morphological and functional characteristics of mature enterocytes varies greatly between experiments and laboratories. Taking this into consideration, we have chosen to use in our work undifferentiated Caco-2 cultures to study their biological responses to AuNPs exposure. In this way we looked at populations of heterogenic pool of Caco-2 cells prior to differentiation, therefore having wider plasticity in the cellular physiology, sensing and adaptation/response to stress, as compared to mature and polarized enterocytes (Tadiali et al., 2002). Keeping this in mind, when running both the cytotoxicity and RNA transcript profiling experiments, we explored Caco-2 cells (in undifferentiated stage) as potentially more suitable model for detecting alterations in gene expression patterns upon exposure to NPs. We have to stress, that in that particular case, undifferentiated Caco-2 cells might have shown culture growth condition specific routes and efficiency in uptake of NPs. This in turn may contributed in part to alternative cellular responses to stress and gene expression patterns, as compared with possibly different responses of classical model of mature enterocyte-like Caco-2 monolayer cells (Tremblay et al., 2006;Natoli et al., 2011).
When exposed to AuNPs, Caco-2 cells efficiently uptake AuNPs at both incubation times tested (24 and 72 h). However, we did not detect further increase in cellular uptake at later time point (72 h), maybe due to reaching of a steady-state of the uptake. These findings are in line with data reported earlier, supporting the notion that extending the incubation times above 24 h might not be useful for enhancing cellular uptake of nanoparticles, as observed in primary HUVECs, C17.2 neural progenitor cells and rat PC12 cells (Soenen et al., 2012), but in contrast with results obtained with mouse Balb/3T3 cells (Coradeghini et al., 2013). In the later case, time dependent increase in uptake of AuNPs, even after 24 h exposure time, was observed. We have also observed that if we express the uptake in number of NPs/cell or % of AuNPs versus total exposure, at the same external dose of exposure, the 5 nm AuNPs cell internalisation is higher than 30 nm AuNPs. This result is probably due to the fact that cells are exposed to a number of small NPs that is approximately 100 times higher than bigger NPs. In addition, as previously observed for Balb/3T3 cells (Coradeghini et al., 2013) by TEM analysis, both 5 and 30 nm NPs are internalised by endocytic patway.
When testing cytotoxic effects of AuNPs at the cellular level, we observed inhibition of Caco-2 cell growth and decrease in colony forming efficiency, induced by smaller AuNPs (5 nm). We must admit that this cytotoxic effect was observed only at relatively high levels of exposure (200 mM and 300 mM), with the most evident inhibitory effect detected at concentration 300 mM at 72 h. At the molecular level, when measuring biomarkers related to AuNPs exposures by changes in RNA expression levels, broad range of responses which potentially mediate inhibition of cellular growth and other functions of Caco-2 cells in response to nano gold, were identified. These cellular processes affected by exposure to 5 nm AuNPs include, for the early time point examined (24 h), down regulation of genes involved in regulation of transcription (transcriptional factors), biogenesis of RNA and GTP binding among others. At the later time point studied (72 h), more dramatic changes in differentially expressed gene expression patterns evoked by 5 nm AuNPs were detected. Beside 708 down regulated genes, an increase in abundance of 103 RNA transcripts was also observed. Thus, our gene expression profiling experiments demonstrated that RNA/zinc ion/transition metal ion binding, heat shock protein binding, RNA degradation and splicing (decreased); cadmium/copper ion binding, amino acid/amine transmembrane transporter activity and glutathione metabolism (increased), were among the cellular response processes correlated with exposure of Caco-2 cells and were associated with cytotoxicity induced by smaller AuNPs (5 nm, 300 mM).
[ ( F i g . _ 5 ) T D $ F I G ] Fig. 5. Predicted miRs with the corresponding target genes that were decreased upon exposure of Caco-2 cells to 5 nm AuNPs (300 mM, 72 h). Group of miRs having 5-9 predicted target mRNAs (A). Set of genes with the highest number of predicted miRs binding sites in their mRNAs sequences (B). In gray are indicated mRNAs with no miRs binding site prediction. Asterics (*) denotes the most down regulated genes. Both word clouds were generated using the data presented in Suppl. Table 3. Interestingly, although the larger (30 nm) AuNPs had rather limited effect on Caco-2 cell mRNAs expression, these AuNPs were still able to induce expression of couple of genes already at 100 mM exposure level at 24 h time point (CDH16, CPA2, TSC22D3 and CRIPAK). Noteworthy, the CDH16 protein was shown to be exclusively expressed in the kidney (Thomson et al., 1995); however its mRNA transcript induction was evident when the Caco-2 cells, originating from the colon, were exposed to 30 nm AuNPs.
As mentioned earlier, treatment of cells with 5 nm AuNPs had very pronounced effect on increase in mRNAs levels of metallothioneins (MTs) genes belonging to the MT1 and MT2 isoforms only. Exposure to nano gold had no effect on the MT3 or MT4 mRNAs expression in Caco-2 cells (gastrointestinal origin) due to the fact that MT3 is usually expressed in neurons (Masters et al., 1994), while MT4 is exclusively found in stratified squamous epithelium .
Differential expression of selenoproteins (GPX2" while SELT, SEPP1, SELK and 15 kDa selenoprotein#), might be related to changes in levels of selenium. Selenoproteins belong to a group of proteins which play important role in, e.g., oxidative stress signaling and protection, redox homeostasis, thyroid hormone metabolism, protection of some forms of cancer among others (Papp et al., 2010). It has been shown that the mRNA stability of selenoproteins differs depending on the availability of selenium within the cell (Schomburg and Schweizer, 2009). Therefore, we have hypothesized that Caco-2 cells might have responded to high concentration of AuNPs as if they have faced selenium "deficiency". It is also possible that high concentration of gold within the cell or/and extra cellular milieu had an effect on selenium homeostasis or/and its availability for uptake by Caco-2 cells.
Lowered cellular selenium content can evoke Nrf2 and Wnt stress response signaling via disturbed redox state as demonstrated by Brigelius-Flohé and Kipp (2013). The changes we found in expression of genes belonging to Nrf2 mediated signaling cascades (metallothioneins, HMOX, OSGIN1, G6PD, GPX2 and other selenoproteins) are indeed well described targets of regulation by transcriptional factor Nrf2 (Kensler et al., 2007;Miles et al., 2000). Therefore, changes in expression of genes being under Nrf2 control and induced by small AuNPs (5 nm) might be the result of different triggers: metal exposure, oxidative stress, disturbed redox and selenium status, and/or combination of them. Noteworthy, it was recently observed that Nrf2 is activated in Caco-2 cells upon exposure to high concentrations of silver NPs (Aueviriyavit et al., 2014) supporting our reasoning that induction of mRNA levels of the genes mentioned above, involves Nrf2 mediated signaling events, at least to some extent.
The results presented here clearly indicate that the cancer cells used in this study were able to recognize the AuNPs not only as metal entities but also as potential "danger" signals and stress inducers. Although Caco-2 cells showed the capacity to activate Nrf2-mediated defense networks (Kensler et al., 2007), these responses were not sufficient in preventing toxic effect of 5 nm AuNPs that were observed in CFE assay. Consequently, stress resulting from cytotoxic level of smaller AuNPs had an effect on down regulation of genes important for RNA biogenesis (transcription, splicing) and post-translational modifications (ubiquitination, sumoylation) of proteins that are vital for protein stability, activity and turn-over (proteolysis), especially when the misfolding of proteins occurs, for example during oxidative stress. Therefore, Caco-2 cells exhibited slower growth and lower efficiency in forming colonies.
It is well known that gene expression can be regulated at chromatin level (histone code) and RNA level (transcription, splicing, post-transcriptional RNA modifications, RNA stability/ turn over of RNA transcripts, cellular localization of RNA molecules, their translation etc.). These processes quite often involve and are regulated by non-coding RNAs (ncRNAs), including small (20-22 nt long) non-coding microRNAs (Venkatesh et al., 2013;Ullah et al., 2014). For that reason, we tested in silico the microarrays expression data for the presence of miRNAs binding sites across highly down regulated mRNAs (300 mM, 72 h, 5 nm AuNP). Indeed, prediction analysis pointed out that some miRNAs (miR 340, miR 181a, miR 410, miR 520 d5p) could play to some extend a role in down regulating many of tested mRNA targets.
The explanation of the very weak response of Caco-2 cells at transcriptome level, upon exposure to larger AuNPs (30 nm), remains still unclear. When interpreting our results the following explanation(s) can be suggested. As already reported in literature, the cellular uptake of engineered nanoparticles depends on the size of nanoparticles (Wang et al., 2010;Labens et al., 2013;Li and Schneider, 2014). In our experimental settings, the number of AuNPs taken up by the cells is much higher in the case of smaller NPs (5 nm), and this by itself might represent more biologically significant "stimulus". Thereafter, combination of the frequency of cellular membrane association/engulfment, uptake and exocytosis of 5 nm AuNPs, may trigger much stronger effect on signaling events, as compared to 30 nm AuNPs that are present in lower number but at the same molar Au concentration (Bahrami et al., 2014;Oh and Park, 2014). Moreover, it is also possible that smaller AuNPs (5 nm), with higher surface area to volume ratio, resulting in higher reactivity toward biological molecules (Nel et al., 2009), are more "disruptive" to the homeostasis of exposed cells as compared to larger AuNPs (30 nm). The different expression profiles observed in Caco-2 cells exposed to the two different sizes of nano gold particles, gave an additional support to previously published findings demonstrating that isotropic AuNPs larger than 5 nm seem to be biologically inert .
We must stress however, that although the larger AuNPs showed very limited activity at transcriptional level, the picture of cellular responses might be different when looking at different exposure time (e.g., earlier sampling time), end-points/markers (protein expression, post-translational modifications, cellular metabolism, intracellular trafficking/secretion of bio-molecules, etc.). Also, the fact that Caco-2 cells were exposed to much higher number of small AuNPs as compared to larger AuNPs, points out that some cytotoxic effects might be observed even for 30 nm AuNPs, if given to cells in adequately higher numbers.
Interestingly, Caco-2 cells actually responded to 30 nm AuNPs already at 24 h by up regulating four gene transcripts after being exposed to this AuNPs (100 mM). Notably, under the same exposure conditions, presence of 5 nm AuNPs in the cell culture medium and/or as adsorbed/internalized AuNPs by cells, did not affect gene expression patterns in Caco-2 cells. One possible explanation is that the mass transfer of larger AuNPs (30 nm) to the cellular surface and interaction with cellular membrane, as a result of more rapid gravitational settling/sedimentation (Wittmaack, 2011;Li and Schneider, 2014) which is higher as compared to smaller AuNPs, triggering diverse signaling events, at different time points and with possible different effects on gene expression as compared to these evoked by exposure to 5 nm AuNPs.
The observed cytotoxic properties of smaller AuNPs might be of practical use for enhancement of already existing AuNPs based thermal therapies, as well as in other cancer treatment regimes (Schütz et al., 2013). Nevertheless, we should not ignore the fact that changes in gene expression induced by 5 nm AuNPs exposure might lead to potentially hazardous effect for healthy cells and/or tissues that still need to be estimated and characterized despite the absence of acute toxicity effects. This is of great concern, especially for individuals who are undergoing long-term AuNPs-based treatment.

Conclusions
A combination of standard cytotoxicity methods, such as CFE and Trypan blue exclusion assays, with gene expression profiling (transcriptomics) allowed us to identify cellular signaling and stress response pathways that might be associated with the cytotoxicity observed upon exposure of Caco-2 cells to citratestabilized AuNPs (5 nm). However, it is critical to point out that changes in gene expression at mRNA level do not necessarily correlate with observed trends (up/down regulation) in the amount of corresponding protein (Walker and Hughes, 2008). Therefore, validation(s) of alterations in mRNA expression, observed in Caco-2 cells as response to AuNPs exposures, requires further work to estimate the changes in the corresponding expression of encoded protein(s), their activity, proper localization and/or secretion. The very same approach also applies for testing whatever the predicted involvement of miRNAs and/or other ncRNAs takes place and might contribute to cellular responses evoked by treatment of Caco-2 cells with AuNPs.
Nevertheless, data presented here provide a starting point in exploration of possible mechanisms related to cytotoxicity observed in cells exposed to AuNPs and can be applied for designing more efficient and safe nano gold nanomedicines.

Funding
This work was supported by the European Commission DG Joint Research Centre (Work Program Action 15024 and 15014) for Nanobiosciences (NBS) Unit and Chemical Assessment and Testing (CAT) Unit respectively, of the Institute for Health and Consumer Protection (IHCP) at European Commission Joint Research Centre (JRC), Ispra, Italy.

Transparency document
The Transparency document associated with this article can be found in the online version.