Macrophage Responses to Silica Nanoparticles are Highly Conserved Across Particle Sizes

TOXICOLOGICAL SCIENCES 107(2), 553–569 (2009) doi:10.1093/toxsci/kfn250 Advance Access publication December 10, 2008

Macrophage Responses to Silica Nanoparticles are Highly Conserved Across Particle Sizes Katrina M. Waters,*,†,1 Lisa M. Masiello,*,‡,1 Richard C. Zangar,*,‡ Barbara J. Tarasevich,*,§ Norman J. Karin,*,‡ Ryan D. Quesenberry,*,‡ Somnath Bandyopadhyay,*,† Justin G. Teeguarden,*,{ Joel G. Pounds,*,‡ and Brian D. Thrall*,‡,2 *Environmental Biomarkers Program; †Computational Biology and Bioinformatics; ‡Cell Biology and Biochemistry; §Materials Chemistry; and {Biomonitoring and Modeling Groups, Pacific Northwest National Laboratory, Richland, Washington 99352 Received September 29, 2008; accepted November 28, 2008

Concerns about the potential adverse health effects of engineered nanoparticles stems in part from the possibility that some materials display unique chemical and physical properties at nanoscales which could exacerbate their biological activity. However, studies that have assessed the effect of particle size across a comprehensive set of biological responses have not been reported. Using a macrophage cell model, we demonstrate that the ability of unopsonized amorphous silica particles to stimulate inflammatory protein secretion and induce macrophage cytotoxicity scales closely with the total administered particle surface area across a wide range of particle diameters (7–500 nm). Whole genome microarray analysis of the early gene expression changes induced by 10- and 500-nm particles showed that the magnitude of change for the majority of genes affected correlated more tightly with particle surface area than either particle mass or number. Gene expression changes that were particle size-specific were also identified. However, the overall biological processes represented by all gene expression changes were nearly identical, irrespective of particle diameter. Direct comparison of the cell processes represented in the 10- and 500-nm particle gene sets using gene set enrichment analysis revealed that among 1009 total biological processes, none were statistically enriched in one particle size group over the other. The key mechanisms involved in silica nanoparticle-mediated gene regulation and cytotoxicity have yet to be established. However, our results suggest that on an equivalent nominal surface area basis, common biological modes of action are expected for nano- and supranano-sized silica particles. Key Words: amorphous silica; nanoparticle; nanotoxicology; macrophage; inflammation.

Many of the envisioned societal benefits of nanotechnology take advantage of the fact that the chemical reactivity and physical properties of many materials significantly change at 1

These authors contributed equally to this work. To whom correspondence should be addressed at Cell Biology and Biochemistry Group, Pacific Northwest National Laboratory, Richland, Box 999, Mail Stop P7-56, WA 99352. Fax: (509) 376-6767. E-mail: [email protected]. 2

the nanoscale. However, the possibility that the properties that make nanomaterials attractive for commercial and medical use may also enhance their biological reactivity and toxicity is an important concern. Indeed, materials classically considered biologically inert, such as amorphous silica (AS), titanium dioxide (TiO2), and gold have been reported to cause exacerbated biological responses when used at the nanoscale (Bermudez et al., 2004; Ferrin et al., 1992; Goodman et al., 2004; Kaewamatawong et al., 2005; Oberdorster et al., 2005). Unfortunately, the biological effects of nanomaterials cannot be extrapolated from first principals based on bulk chemical and physical properties alone. Thus, experimental determination of how the biocompatibility of a material changes as a function of particle size is critical to advancing its application in nanotechnology. For a given mass of particles, the total surface area increases with decreasing particle diameter. Particle size is thereby an essential determinant of the fraction of reactive groups on particle surface (Nel et al., 2006). The empirical relationship between particle size and biological response provides important insight into whether the particle surface properties that dictate biocompatibility fundamentally change at the nanoscale. However, systematic investigations of this relationship across a broad range of particle sizes and biological responses are still quite limited. Most previous dose-response studies employing nanoparticles have focused on only a small number of biological endpoints, and evaluated only a few particle sizes. For instance, studies have found that on a mass dose basis, nanoscale AS, TiO2, and carbon black particles (< 100 nm) induced much greater lung inflammation than chemically similar particles of larger diameter (Kaewamatawong et al., 2005; Oberdorster, 2000; Stoeger et al., 2006). To explain these apparent differences in inflammatory potency, various metrics of particle dose have been evaluated. For TiO2 or ultrafine carbon, particle surface area dose correlates more closely with lung neutrophil influx responses than either particle mass or particle number as dose metrics (Oberdorster, 2000; Oberdorster et al., 2005; Stoeger et al., 2006). Based on


The Author 2008. Published by Oxford University Press on behalf of the Society of Toxicology. All rights reserved. For permissions, please email: [email protected]

554

WATERS ET AL.

a re-analysis of these results however, Wittmaack disputed this interpretation and suggested that with the exception of particle surface area, either particle number or mass worked well as dose metrics for comparing inflammatory potential of different particles sizes or types (Wittmaack, 2007). A study using polystyrene particles also found a proportional relationship between particle surface area dose and pulmonary neutrophil influx (Brown et al., 2001). In contrast, other endpoints measured in the same study, such as cytokine secretion, did not closely correlate with surface area dose. Clearly, accurate interpretation of size-dependent biological responses to nanoparticles depends on the choice of dose metric used for comparison. However, because many of the potential cellular pathways activated by nanomaterials are still unknown, it is not clear that a common dose metric should be expected for all biological effects or for all materials. Some effects may be initiated through a general site-independent mechanism such as through redox reactions at the cell membrane, and thus be highly dependent on the surface reactive sites of the particle. Conceivably, other biological responses may be mediated downstream of specific receptor binding events (Chao et al., 2001), and thereby more dependent on a specific number or size of particles interacting at those sites than on particle surface reactivity per se. In this study, we used a genome-wide approach to investigate whether the cellular pathways activated by AS are size dependent. AS was chosen as an experimental material because of its increasingly important role in nanotechnology. AS is commonly used in production of high efficiency photovoltaics, tires, and in products directly intended for human exposure (Brumfiel, 2006) including drug delivery vehicles, toothpaste, sunscreen, cosmetics, and nutraceuticals (Son et al., 2005; Wen et al., 2006). In contrast to crystalline silica which can cause pulmonary fibrosis and some forms of lung cancer in humans, epidemiological evidence of pathologies associated with AS is less certain. Micron-sized AS is regarded as less pathogenic than crystalline silica, although at sufficient doses it induces substantial, albeit transient lung inflammation (Wiethoff et al., 2003). In contrast to micronsized particles, the potential toxicity of nanoscale AS has not been adequately evaluated (Fond and Meyer, 2006). It was reported that AS nanoparticles cause cytotoxicity, oxidative stress and induce protein aggregates that impair nuclear function (Chen and von Mikecz, 2005; Lin et al., 2006). Others found that on an equal mass basis AS nanoparticles (14 nm) cause more severe bronchial cell necrosis and lung inflammation than similar particles of larger diameter (230 nm) (Kaewamatawong et al., 2005). Because of the sporadic information on size-dependent particle-mediated effects, the present study was conducted to determine if global expression changes induced in macrophages by AS differ between nanoand supranano size particles, and to determine how broadly particle surface area as a dose metric correlates with these responses.

MATERIALS AND METHODS Nanoparticles and characterization. AS particles were obtained in water suspension from either Polysciences, Inc. (Warrington, PA) or from W.R. Grace and Company (Columbia, MD) (Table 1). The crystalline silica reference material (Min-U-Sil 5) was obtained from U.S. Silica Company (Berkely Springs, WV). All particles were tested for contaminating endotoxin using the Pyrogene Recombinant Factor C Assay (Cambrex Bioscience, Walkersville, MD). Particles which tested positive (0.01 EU/ml detection limit) were excluded from further study. The precise concentrations of AS nanoparticles in suspension were determined gravimetrically. Particle suspensions were diluted and known nominal weights were placed into sterile SpectraPor7 dialysis tubing (MWCO 10k, Spectrum, Houston, TX) and dialyzed against water to remove soluble contaminants. The suspensions were weighed after dialysis and known volumes were dried under vacuum at 200C for 2 days. The dried AS was weighed and suspension concentrations were determined as % g silica/ml. These independent measurements were essential because in some cases the measured particle concentration varied significantly from the information provided by the supplier. Primary particle diameter was determined by transmission electron microscopy (TEM). The agglomeration states of the particles in suspension were studied using dynamic light scattering (DLS) performed at Particle Characterization Laboratories, Inc. (Novato, CA). Hydrodynamic diameters were obtained using a Brookhaven Instruments 90 Plus equipped with a 657-nm, 35-mW laser. Time dependent fluctuations in the scattered intensity were measured using a Brookhaven Instruments Avalanche Photodiode Detector digital correlator. Nanoparticle suspensions in serum-free Dulbecco’s modified Eagle’s medium (DMEM) media at 1, 10, and 100 lg/ml concentrations were analyzed in triplicate using a 90 scattering angle at 25.0C. The buffer solutions were filtered through 0.2- and 0.02lm filters and were also analyzed by DLS. Standard National Institute of Standards traceable polystyrene 40 ± 1.8-nm latex standards and a blank, 0.02-lm filtered deionized ultrapure water (VWR, West Chester, PA) were run as reference standards. The autocorrelation functions were deconvoluted to obtain size distributions using both the non-negatively constrained least squares fit (multiple pass NNLS) and the regularized LaPlace inversion (Contin) algorithms. The volume weighted size distributions obtained from the algorithm with the best fit (primarily the Contin algorithm) are presented. The average hydrodynamic diameter was determined from the translational diffusion constant obtained from the autocorrelation function using the Stokes-Einstein equation. It was not possible

TABLE 1 AS Particle Characteristics

Particle type Polysciences 10 Polysciences 50 Polysciences 150 Polysciences 300 Polysciences 500 Ludox HS-30 Ludox SM Min-U-Sil 5 a

Primary diameter (nm)

Hydrodynamic diametera (nm)

11.5 ± 2.3c 35.0 ± 0.3 53.0 ± 12.6c 85.1 ± 0.1 150.0 ± 30.0d 147.5 ± 0.1 300.8 ± 16.3c 318.9 ± 0.1 500.0 ± 70.0d 420 ± 15 16.0d n.d.e n.d. 7.0d ~1700c n.d.

f potential (mV)
25.0 ±
19.3 ±
27.9 ±
26.2 ± n.d. n.d. n.d. n.d.

2.6 1.8 1.6 1.6

Silanol free/Hbond ratiob 1.06 2.36 1.74 1.76 n.d. 1.07 1.59 n.d.

Mean (± SD) DLS measurement conducted in serum-free culture medium. Ratio of free:hydrogen-bonded silanol FTIR band intensities measured in dried particles. c Mean (± SD) diameter determined by TEM at PNNL. Min-U-Sil 5 particle diameters varied broadly. d Data provided by supplier. e Not determined. b

MACROPHAGE RESPONSES TO SILICA NANOPARTICLES to accurately determine particle surface area by Brunauer, Emmett, and Teller technique analysis because AS dissolves and reprecipitates upon drying due to capillary condensation, resulting in a large reduction in surface area. Thus, surface areas were calculated from the measured particle diameters assuming spherical particle shapes (confirmed by TEM images). Zeta potential measurements were obtained using the ZetaPALS accessory of the Brookhaven Instruments 90 Plus instrument (Long Island, NY). Electrophoretic mobility measurements were obtained for the suspensions in serum-free DMEM at 100 lg/ml as well as blank water, DMEM buffer, and a Brookhaven Bl-ZR3 electrophoretic mobility standard as controls. The electrophoretic mobility was converted into zeta potential values using the Henry equation. Particles were also analyzed by fourier transform infrared (FTIR) absorbance spectroscopy to estimate the relative degree of free and hydrogen-bonded silanols (Iijima et al., 2007) and potential presence of contaminants. Cell culture and treatment. The RAW 264.7 murine macrophage cell line was obtained from the American Type Culture Collection (Rockville, MD) and cultured as we have previously described (Weber et al., 2006). For viability assays, cells were plated in 24-well culture plates at a density of 75,000 cells per well and allowed to attach and grow for 24 h in complete medium (DMEM supplemented to 10% fetal bovine serum and penicillin/streptomycin). For all results reported, particle treatment was conducted in serum-free medium to minimize protein opsonization effects. For viability assays, particles were suspended by vortexing in serum-free DMEM at the concentrations indicated and treatments (0.5 ml final volume) were conducted for 24 h. Cell viability measurements were performed using the (3-(4,5-dimethylthiazol-2-yl)-2,5diphenyltetrazolium) bromide (MTT) assay (Sigma Chemical, St. Louis, MO), according to manufacturer recommendations. Similar viability measurement results were observed using the MTT assay or the neutral red assay, and we found no evidence that the AS particles used interfered directly with the MTT measurements (data not shown). Protein microarray enzyme-linked immunosorbent assay analysis. Protein secretion in response to nanoparticle exposure was determined using a custom protein microarray enzyme-linked immunosorbent assay (ELISA) platform, as previously described (Woodbury et al., 2002). The same cell culture treatment conditions used in the viability assays were used for these experiments, and conditioned medium was collected 24 h after AS exposure for analysis. Individual ‘‘sandwich ELISAs’’ were developed for 10 different mouse proteins, which were selected based on literature evidence of their involvement in lung disease and inflammation: eotaxin, G-CSF, IL-1a, IL-12, KC, MMP2, chemokine ligand 5 (RANTES), Tarc, TNF-a, and VEGF. Capture antibodies, biotinylated detection antibodies and antigens for the 10 sandwich ELISA’s were purchased as a lyophilized powder from R&D Systems (Minneapolis, MN). Capture antibodies were suspended in water to 500 lg protein/ml, aliquoted, and lyophilized. Biotinylated detection antibodies were dissolved in water to 100 lg protein/ml, aliquoted, and lyophilized. Antigens were suspended in 1% casein in buffered saline (BioRad, Hercules, CA) at a concentration of 100 ng/ml and were lyophilized as single-use aliquots. Aliquots of all reagents were stored at
20C until use, at which time they were resuspended at the original concentration. ELISA microarray chips were prepared and processed essentially as described previously (Varnum et al., 2003; Woodbury et al., 2002). Briefly, 25 3 75-mm, amino-silanated glass slides (Erie Scientific, Portsmouth, NH) were activated with (bis)sulfosuccinimidyl suberate immediately prior to printing of the capture antibodies. Capture antibodies were printed at 50% humidity using a GeSiM NanoPlotter 2.0 (Quantum Analytics, Foster City, CA) equipped with noncontact piezo tips that print ~0.4 nl per spot. Spot pitch was 500 lm. To facilitate sample throughput and reproducibility, 16 identical chips were printed on each slide using a hydrophobic barrier to separate the samples, as described previously (Zangar et al., 2006). Each of the 10 capture antibodies was printed three or four times on each chip, along with Cy3-labeled protein that served as an orientation marker. After printing, chips were blocked in 1% casein and stored dry at
20C until use. All 10 antigen standards were combined prior to serially diluting threefold in 0.1% casein to create seven standards that spanned a 729-fold concentration range. A blank solution of 0.1% casein was also analyzed. Prior to

555

analysis, conditioned medium was centrifuged to remove particulates and was diluted 0.1-fold and 11-fold such that the final solution for both dilutions contained 0.1% casein. Both dilutions were analyzed in duplicate. To avoid any potential slide-related bias in signal intensity, the technical duplicates were placed in different positions on different slides. Sample positioning was also blocked based on treatment group so that typically no more than two samples of the same treatment group were located on a single slide. Following incubation with either a sample or a standard mixture, each chip was incubated with the detection antibody mixture. The biotin signal was amplified using the biotinyltyramide amplification procedure followed by incubation with streptavidin conjugated to Cy3 (Varnum et al., 2003; Woodbury et al., 2002). Slides were washed in buffered saline (pH ¼ 7.2) containing 0.05% Tween-20 before and after all incubation steps. Slides were washed with water and dried prior to fluorescent analysis. The fluorescent images for the slides were obtained using ScanArray ExpressHT laser scanner and the spot intensity quantified using ScanArray Express (Perkin Elmer, Wellesley, MA). Standard curves were generated and sample antigen concentrations were calculated using the Protein Microarray Analysis Tool (ProMAT), which is a custom freeware program that we developed specifically for this use (White et al., 2006). Gene expression and pathway analysis. Whole genome microarray analysis was performed using Affymetrix Mouse Genome 430A 2.0 chips (Affymetrix, Santa Clara, CA; 22,690 probesets). RAW 264.7 cells were plated in 60-mm plates (7.5 3 105 cells) overnight and then exposed to 10 nm (5, 20, 50 lg/ml) or 500 nm (250, 500, 1000 lg/ml) AS for 2 h in serum-free medium. This time point permitted identification of early gene regulatory changes that occur prior to any observable indications of cytotoxicity. The concentrations used were chosen such that the total administered surface area of the particles was the same for the low concentration of each particle and for the middle and high concentration of the 10- and 500-nm particles, respectively. Total RNA was collected using RNeasy (Qiagen, Valencia, CA) and the integrity and purity evaluated by gel electrophoresis and absorption spectroscopy using an Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA). Complementary DNA was synthesized from 5 lg of total RNA in the presence of an oligo-dT primer containing a T7 RNA polymerase promoter, and an in vitro transcription reaction was performed in the presence of mixture of biotin-labeled ribonucleotides to produce biotinylated cRNA from the cDNA template, according to manufacturer’s protocols. Biotin-labeled cRNA (15 lg) was fragmented to a size range between 50 and 200 bases and used for array hybridization. After hybridization, the arrays were washed, stained with streptavidin–phycoerythrin, and scanned at a resolution of 2.5 lm using an Affymetrix GeneChip Scanner 3000. Quality control parameters were assessed throughout the experimental process to measure the efficiency of transcription, integrity of hybridization, and consistency of qualitative calls. The synthesis of the cDNA and cRNA, and the fragmentation of cRNA were assessed using the Agilent 2100 Bioanalyzer. Spike-in control transcripts also were monitored to verify hybridization integrity. Each biological sample was hybridized against three independent arrays. Raw intensity data were quantile normalized (Bolstad et al., 2003) and subjected to analysis of variance (Kerr et al., 2000) with Tukey’s post hoc test and 5% false discovery rate calculation (Benjamini and Hochberg, 1995). Unsupervised hierarchical and Kmeans clustering analyses of microarray data were performed using an Euclidean distance metric and average linkage clustering to create the hierarchical trees that describe similar (and distinct) patterns of gene expression between 10- and 500nm particle treatments. The clustering algorithms, heat map visualizations and centroid calculations were performed based on log10 expression ratio values using Multi-Experiment Viewer software (Saeed et al., 2003). To measure the level of significance for gene ontology (GO) biological processes represented by the gene expression changes identified, gene enrichment statistics were derived using the Fatigo tool (Al-Shahrour et al., 2004). Unless otherwise stated, only biological process terms that passed a p value of < 0.05 and were represented by at least 10 genes were considered significant. mRNA quantitation by quantitative reverse transcriptase polymerase chain reaction analysis. Real time quantitative reverse transcriptase polymerase chain reaction (qRT-PCR) was used to verify gene expression changes

556

WATERS ET AL.

measured by Affymetrix microarray analysis. Complementary DNA was synthesized from total RNA via reverse transcription with oligo-dT priming (Promega, Madison, WI). Primer pairs designed to amplify cDNA encoding the target genes were prepared using the Universal ProbeLibrary Assay Design Center (Roche Applied Science, Indianapolis, IN) (see Supplementary Table S1). All primer pairs spanned an intron to eliminate potential amplification of contaminating genomic DNA. PCR reactions were carried out using FastStart DNA Master SYBR Green I reagents and 3mM MgCl2 according to the manufacturer’s instructions (Roche Applied Science) in a Lightcycler II (Roche Applied Science). Cycle parameters were: 95C for 10 s, 55C for 5 s, and 72C for 10 s for 45 cycles. Melting curve analyses were performed from 60 to 95C in 0.5C increments. qRT-PCR data for each gene product were normalized to the level of cyclophilin A transcript levels and are reported as mean (± range) relative changes compared with untreated controls from duplicate analyses. Statistical analyses. Statistical differences between treated and untreated groups was determined by one-way ANOVA (Systat, San Jose, CA). Response to experimental treatments that was different than untreated control cells was determined at p  0.05 by the Dunnett’s post hoc test. To determine the degree of linear correlation between calculated particle surface area or mass dose and the magnitude of gene expression measured by microarray or RT-PCR analysis, the Pearson’s correlation coefficient (r) was calculated using Microsoft Excel software.

RESULTS

Particles and Characterization The particles used in these studies along with their physicochemical characteristics are listed in Table 1. A summary of the distribution of particle sizes determined by TEM and

DLS measurements is also shown in Figure 1. All particles consisted of spherical AS without additional surface chemistry modifications. TEM measurements conducted on sets of particles used in the study generally verified the TEM size measurements provided by the suppliers, within reported variances. DLS measurements conducted in protein-free DMEM indicated the hydrodynamic diameters of the Polyscience 150-, 300-, and 500-nm particles were within the size range estimates measured by TEM. The average hydrodynamic diameter estimates for Polysciences 10- and 50-nm particles were larger than TEM measurements. These differences likely reflect the formation of short chain aggregates consisting of two to three particles connected by a small neck region, which was observed in some cases by TEM (not shown). These hard aggregates are generally formed during the particle synthesis step when the silica precursor deposits between adjacent particles to form necks. Hard aggregates tend to be more common during the synthesis of smaller particles compared with larger particles. This moderate degree of aggregation of 10-nm particles in culture medium is consistent with previous studies of these particles from our group (Dutta et al., 2007). Because of the limited surface comprising these interconnecting regions, we estimate that the total exposed particle surface area is minimally affected although hydrodynamic size is increased. It is also apparent from the DLS measurements in Figure 1 that only

FIG. 1. AS particle size distributions as determined by TEM and DLS. Primary particle diameter (red bars) and hydrodyanamic diameters (black bars) were determined using TEM and DLS measurements, respectfully, as described under ‘‘Materials and Methods.’’ Serum-free culture medium was used as the diluent for the DLS measurements. The particles shown include Polysciences 10 nm (A), Polysciences 50 nm (B), Polysciences 150 nm (DLS only, C), and Polysciences 300 nm (D).

MACROPHAGE RESPONSES TO SILICA NANOPARTICLES

557

a minor fraction of the 10-nm particles formed larger agglomerates in serum-free DMEM. The lack of agglomeration of the silica particles in suspension is consistent with the relatively high negative zeta potentials expected for unmodified silica as shown in Table 1. Colloidal stability is promoted by zeta potentials in the 20–30 mV range as well as the relatively dilute suspensions used in this study. High zeta potentials were obtained even in the relatively high ionic strength media solutions used. No apparent size-dependent trends in zeta potential were observed among the particles tested. FTIR absorbance spectroscopy was used to measure the ratio of peak intensities for free silanol (SiOH) groups (3741 cm
1) to hydrogen-bonded groups (3660 cm
1) in oven-dried particles. A previous study reported that the fraction of free silanols, which are more reactive than hydrogen-bonded silanols, increase as particle diameter decreases (Iijima et al., 2007). However, this trend was not observed with the particles used in this study, and no consistent relationships between silanol isotypes and particle manufacturer were discerned. Macrophage Cytotoxicity Through their phagocytic activity, macrophages play an important role in determining the biopersistence of foreign particles and initiating inflammatory responses. We first conducted dose-response experiments using AS particles (7– 300 nm diameter) to determine whether the macrophage cytotoxic potential of AS is size dependent. These experiments were performed in serum-free medium to minimize particle opsonization effects which may modify biological activity mediated by particle surface chemistry. For comparison, the results shown in Figure 2 are expressed using three different particle dose metrics. When expressed on a mass concentration (lg/ml) basis, the dose-response relationship for cytotoxicity (measured after 24 h exposure) supports a size-dependent effect, with increased cytotoxic potential for a given mass of AS as particle diameter decreases (Fig. 2A). Calculated LC50 concentrations (concentration where 50% viability is expected at 24 h) based on mass dose varied by more than an order of magnitude, ranging from 20 lg/ml for the smallest diameter particle (7 nm) to 592 lg/ml for 300-nm particles. For reference, we found that under the same assay conditions the LC50 for Min-U-Sil 5 (crystalline silica, mean diameter 1.7lM) was 253 lg/ml (data not shown). When the dose-response relationship is expressed as a function of total particle number, fewer numbers of particles are required to induce cytotoxicity as particle size is increased (Fig. 2B). However, when the results are presented based on the total particle surface area as the dose metric, the dose-response curves for all particles converge into a consistent relationship with surface area (Fig. 2C). Under these experimental conditions, the calculated mean LC50 for cytotoxicity for all particle sizes based on total administered particle surface area was 85 cm2/ml, with a coefficient of variation of only 21.0%. This overall low level of variability strongly supports the hypothesis that

FIG. 2. Dose-response relationships for RAW 264.7 cell cytotoxicity by different sizes of AS particles. The data are presented as a function of mass (A), particle number (B), or total administered particle surface area (C) as alternative dose metrics. Values represent the mean (± SE) of at least three replicates normalized to untreated controls and were determined after 24-h treatment. *Differs from untreated controls (p < 0.05).

558

WATERS ET AL.

particle surface area is a more predictive dose metric for comparing cytotoxicity induced by different particle sizes than either mass dose or total particle number. Proinflammatory Protein Secretion To investigate how functional parameters associated with macrophage activation are affected following 24-h exposure to AS nanoparticles, we used a custom protein microarray ELISA

platform to measure the secretion of proinflammatory proteins. The microarray ELISA approach allowed us to evaluate protein secretion profiles across particle dose and size in a multiplexed manner. Following exposure to AS particles ranging from 10 to 300 nm in diameter, increases in secreted levels of four of the ten proteins included on the array were detected, including RANTES, TNF-a, G-CSF, and VEGF (Fig. 3A). Similar to the cytotoxicity results, protein secretion profiles supported a

FIG. 3. Protein microarray ELISA analysis of proinflammatory protein secretion patterns in response to different size AS particles. (A) A representative image of the raw microarray data for a control and 10-nm particle sample. The pseudocolor scale indicates the level of antibody binding. (B–E) The quantitative data obtained based on standard curves generated with each antigen.

MACROPHAGE RESPONSES TO SILICA NANOPARTICLES

size-dependent effect when expressed on a mass dose or particle number basis (data not shown). In contrast, the profiles for protein secretion also tend to converge onto a common dose-response curve across all particles sizes tested when particle surface area is used as the dose metric (Figs. 3B–E). In terms of absolute protein concentration, TNF-a and RANTES showed the highest level of secretion in response to AS, reaching nanomolar concentrations over 24 h. Although absolute concentrations of G-CSF protein were significantly less than that of either TNF or RANTES, the relative level of induction of G-CSF was > 400-fold over untreated controls. In general, increases in protein secretion were observed at AS concentrations starting between 10 and 20 cm2/ml, and began to plateau at higher concentrations. The plateau in the protein secretion dose response at higher AS concentrations may be a consequence of loss of cell regulation over longer treatment times. However, the response to AS treatment was clearly regulated selectively, because extracellular levels of IL-1a, KC, IL-12, MMP2, and Tarc remained below detection limits at all concentrations tested (not shown). Transcriptome Analysis To determine whether the macrophage response to AS particles differs depending on the particle size in a more comprehensive manner, transcriptome analyses were performed using Affymetrix microarrays. To capture the early response pathways activated prior to onset of cytotoxicity, these experiments measured gene regulatory changes occurring at 2 h after exposure to either 10 or 500 nm AS. Three nominal concentrations of each particle size were used, and the experiment was designed such that the surface area dose was equivalent between particles diameters for two of the three concentrations. The microarray data were subjected to analysis of variance along with Tukey’s post hoc test and 5% false discovery rate calculation. This analysis resulted in 503 and 502 genes whose expression was significantly altered by at least twofold with either 10- or 500-nm particles, respectively, in at least one dose group. The union of these gene sets included 753 genes, with 252 genes overlapping between the two particle size groups. A complete listing of these genes is provided in the Supplemental Information (Table S2). Several of the genes whose expression was highly affected by particle treatment are known to play important roles in lung inflammation. For instance, Cxcl2 (also known as MIP-2a), a C-X-C chemokine that exhibits potent neutrophil chemotactic activity in rat lung following quartz exposure (Driscoll, 2000), was induced by over 30-fold compared with controls in both particle size groups. Additional chemokines found to be induced by greater than fourfold (in at least one dose group) included Ccl4 (MIP-1b) and Ccl3 (MIP-1a), along with the chemokine receptor Cxcr4. mRNA levels for TNF, a cytokine required for initiation of particle-mediated inflammation in the lung (Driscoll, 2000), were also elevated within 2 h following

559

AS treatment. These results are consistent with our findings obtained by protein microarray ELISA. Also consistent with the protein ELISA results, levels of VEGF mRNA were elevated compared with controls, although the level of induction (~1.8-fold) did not meet the more stringent twofold criteria applied in this analysis. In contrast, mRNA levels for G-CSF were not different from control after 2 hr exposure, despite a dose-dependent increase in protein levels observed at later time points. The latter observation may indicate G-CSF is proteolytically shed from cells in response to AS, rather than under transcriptional control. In addition to cytokine family members, transcription factors involved in regulation of cytokine and stress responses were among the largest class of genes induced by AS. For instance, mRNA levels for ATF-3, which is reported to regulate expression of CCL4 (Khuu et al., 2007), were induced up to 10-fold by both particle sizes. Other transcriptional regulators whose expression was increased include nuclear receptor subfamily 4a (Nr4a), N-myc downstream regulated gene 1 (Ndrg1), v-maf masculoaponeurotic fibrosarcoma oncogene F, and the negative growth regulator GADD45b (see Supplemental Information). The induced expression of these transcriptional regulators is consistent with a hypothesized role of oxidative stress mediating AS-induced responses. To explore expression patterns and dose-response relationships at the global level, the union set of 753 genes was subjected to K-means clustering. This analysis resulted in identification of six gene clusters representing three predominant dose-response patterns across the particle size and mass dose groups. The average expression profiles for each of these groups are represented in the heat maps and centroid plots shown in Figure 4. Pattern A, represented by a majority of the genes (404 total; Figs. 4A and 4D), showed robust doseresponse behavior in both the 10- and 500-nm particle groups. Within this pattern, 183 genes showed strong induction by both particle sizes and 221 genes showed dose-dependent suppression by both particle sizes. Patterns B (nanodominant, 21% of the total significant genes) and C (microdominant, 23% of total) also showed dose responsive behavior, but the average magnitude of response was more pronounced in either the 10-nm group (B) or the 500-nm group (C). Unsupervised hierarchical clustering was then performed on the gene sets that comprise the three predominant patterns. This analysis demonstrated that the genes in the majority pattern A naturally clustered according to the particle surface area dose (Fig. 5A). In contrast, the genes for patterns B and C continued to separate by particle size and mass dose (data not shown). Pearson correlation analysis for the majority pattern A genes indicated a strong overall correlation between the gene expression ratio and particle surface area dose across all treatment groups (mean r2 ¼ 0.85; standard deviation ¼ 0.15). In fact, for approximately half of the genes (204) in this subset the correlation coefficient for expression ratio versus particle surface area across all treatment groups was > 0.9. A direct

560

WATERS ET AL.

FIG. 4. Major dose-response profiles for gene expression changes induced by 10- and 500-nm AS particles. Heat map profiles for the three major doseresponse patterns of gene expression (2 h) identified by supervised hierarchical clustering are shown in (A)–(C). The centroid plots in (D), (E), and (F), represent the corresponding overall average patterns of expression at three different doses of each particle for pattern A (D), pattern B (E), and pattern C (F).

comparison of the expression ratios measured between the 10- and 500-nm particle groups at equivalent surface area doses is shown in Figures 5B and 5C. Comparison of the expression ratios for the pattern A genes between the 500-nm low-dose and 10-nm low-dose groups resulted in a Pearson’s coefficient (r2) ¼ 0.77. A similar comparison between the 500-nm highdose group and 10-nm mid-dose group showed greater correlation (r2 ¼ 0.92). Considering quantitative limitations of global microarray analysis, the correlation values observed between particle surface area dose and gene expression response are remarkably strong for this majority gene set. The overall correlation for gene expression changes between the 10- and 500-nm low-dose groups for pattern B and C genes was less strong (r2 ¼ 0.50 and 0.59, respectively). In many cases, these lower correlation values reflect genes that showed dose-response behavior in both particle size groups, but the slope of the dose-response tended to be steeper in either the

10- or 500-nm groups. For instance, mRNA expression of the tyrosine kinase receptor, EphA2 was induced by over threefold in the highest doses of 10-nm particles, but the level of induction in the 500-nm particle group only reached 1.6-fold in the highest dose group, despite an overall dose-dependent trend. However, genes that showed significant dose-response expression behavior in the 10-nm particle group with no apparent change at any dose within the 500-nm group (slope ¼ 0) were also identified. For example, several members of the Kruppel-like factor family of zinc finger transcription factors (Klf2, Klf3, Klf7) were selectively induced by 10-nm particles in comparison to 500-nm particles. Klf3, which is a transcriptional repressor involved in regulation of hematopoietic differentiation pathways (Turner and Crossley, 1999), was induced by over fivefold over controls in the 10-nm particle group, yet unaffected by treatment with 500-nm particles. Another example of a transcript selectively induced in the 10-nm particle is MALAT1, which has been previously

MACROPHAGE RESPONSES TO SILICA NANOPARTICLES

561

FIG. 5. Relationship between mRNA abundance changes and particle surface area dose for 404 genes regulated in response to 10 and 500 nm AS. (A) Unsupervised cluster analysis of all pattern A genes orders the gene expression profiles according to surface area dose. The pseudocolor scale indicates the expression ratio (log10). (B and C) Scatter plots comparing the magnitude of expression change between 10- and 500-nm particles at equivalent surface area doses, as indicated.

identified to be elevated in nonsmall cell lung carcinoma (Ji et al., 2003). As a noncoding transcript, the functional role of MALAT1 is not understood, however recent whole genome tiling studies have identified several such noncoding transcripts, including MALAT1, with altered expression in several types of human cancer (Perez et al., 2008). To verify the microarray results, we also conducted quantitative real time RT-PCR measurements on the same experimental samples for a subset of genes representing each of the dose-response patterns. For all genes tested (20 total), the general dose-response trends observed in the microarray analysis were verified by RT-PCR (Fig. 6). However, in some cases the magnitude of gene induction measured by RT-PCR was greater than that measured by microarray analysis, as may be expected due to the lower dynamic range afforded by microarray analysis compared with RT-PCR. It is also apparent from these results that many of the gene regulatory responses were saturated at the highest dose of the 10-nm particle. A plateau in the gene induction response at the 10-nm highdose group was common to the majority of genes in this group, as also indicated by the centroid plots of Figure 4. Because this plateau effect was also observed in the RT-PCR results, it appears to be due to true saturation of the biological response at the highest 10 nm dose rather than limited dynamic range of the microarray platform.

For the subset of 20 genes for which RT-PCR analyses were conducted, Pearson’s correlation coefficients were also calculated to determine whether there was a linear relationship between AS surface area dose and level of gene induction. Because the gene expression response in the high-dose 10-nm group was saturated, this group was excluded from the correlation analyses to prevent biasing the results. The results in Table 2 demonstrate that with one exception (Cxcr4), all genes tested from pattern A showed a strong correlation between relative expression level and administered particle surface area. The poor correlation for Cxcr4 was observed in both the microarray and RT-PCR data, and is largely due to the fact that the gene is similarly induced by the majority of particle doses used (see Fig. 6). Genes from patterns B and C were verified by RT-PCR to have lower overall correlation to surface area dose compared with the pattern A genes. In fact, the correlation values for pattern B genes (across both particle groups) suggest an inverse dose-response profile when expressed on a mass dose basis (Table 2). When surface area was used as the dose metric, the correlations are converted to positive values, more accurately reflecting the observed doseresponse relationships. Thus, even though the overall correlation with surface area for this group is less strong than for pattern A genes, potentially due in part to greater variability in the measurements, the use of surface area as a dose metric

562

WATERS ET AL.

FIG. 6. Validation of microarray results by quantitative RT-PCR. Black bars indicate microarray data. Gray bars indicate mean fold change (± range) derived from duplicate RT-PCR reactions.

MACROPHAGE RESPONSES TO SILICA NANOPARTICLES

FIG. 6. Continued.

563

564

WATERS ET AL.

TABLE 2 Pearson Correlation Values for Gene Expression Dose-Response Patterns for 10 and 500 nm AS

Gene symbol (Pattern A) Atf3 Ccl3 Ccl4 Cxcl2 Cxcr4 Gadd45b Hbegf Plau Plk2 Tnf (Pattern B) Epha2 Ndrg1 Nr4a1 Klf2 Klf3 (Pattern C) Gdf15 Cxcl10 Ccrl2 Slc40a1 Irg1

Gene ID

Microarray (r-SA)a

RT-PCR (r-SA)a

RT-PCR (r-Mass)b

11910 20302 20303 20310 12767 17873 15200 18792 20620 21926

0.974 0.933 0.917 0.941 0.081 0.910 0.776 0.795 0.980 0.950

0.973 0.909 0.918 0.820 0.360 0.926 0.789 0.851 0.869 0.952

0.379 0.195 0.211 0.859 0.144 0.674 0.003 0.774 0.861 0.711

13836 17988 15370 16598 16599

0.588 0.600 0.550 0.492 0.431

0.489 0.791 0.566 0.501 0.452


0.422
0.017
0.336
0.398
0.441

23886 15945 54199 53945 16365

0.648 0.644 0.682 0.162 0.803

0.710 0.719 0.753 0.498 0.788

0.968 0.908 0.895 0.993 0.851

a

Pearson’s values were calculated to determine the correlation between calculated surface area dose and the magnitude of gene expression determined by microarray or RT-PCR analysis for 10- and 500-nm particle groups. b Pearson’s value based on RT-PCR data and particle mass dose. All particle dose groups except the highest dose 10-nm particle group were used in the analysis.

appears to capture the dose-response relationship more accurately. This general finding is not specific to either RAW 264.7 cells or the macrophage cell type. For instance, in studies using type II lung alveolar cells (C10), we have found that mRNA expression of the chemokine CCL2 is induced to similar levels by 50 and 300 nm AS when compared at equivalent nominal particle surface area doses, despite a sixfold difference in administered mass dose (Supplemental Fig. S1). Gene Set Enrichment Analysis One advantage of the high-dimensional data microarray analysis provides is that it can provide a snapshot of the overall cell processes initiated in response to a stressor. Thus, by comparing cell process profiles using the genes expression changes identified, we can assess whether there are major differences in the pathways activated by nano- and micronsized AS. To accomplish this comparison, gene set enrichment (GSE) analysis was performed separately for the entire 10- and 500-nm gene sets using the Fatigo tool (Al-Shahrour et al., 2004) to compare the degree of representation of biological process terms from the GO database. As a first approach, the

significance level for a biological process was calculated from the relative frequency of representation in a specific gene set, compared with the frequency of all biological processes represented by all remaining genes on the array (‘‘particle group vs. all genes’’). For a biological process term to be considered significantly enriched, we required that the process within a gene set was represented with a significance level for enrichment of p 0.05, and processes represented by fewer than 10 genes were filtered out to identify the most robustly perturbed pathways. Biological process terms that represented complete subsets of other biological processes were rolled up to the highest level of GO hierarchy to eliminate redundancy. This approach resulted in identification of seven major biological processes, three of which were strongly enriched in the microarray results from both particles sizes, including regulation of transcription, regulation of progression through cell cycle, and inflammatory response (Fig. 7). Four additional biological processes met the significance criteria in only one of the two gene sets. For example, apoptosis, cell differentiation and signal transduction processes were significant for only the 500-nm gene set, and the anatomical structure morphogenesis process was significant only for the 10-nm gene set. However, when the numbers of genes that correspond to categories are carefully scrutinized, it appears that differences in p values between the size classes are the result of only a small number of genes (5) in each case. Even for the apoptosis process, for which there are more gene expression changes identified in the 500-nm particles than 10-nm particles, a closer look at the data reveals that the ‘‘unique’’ apoptosis genes in the 500-nm data set follow the same trend following exposure to 10-nm particles but did not meet the predefined twofold threshold for inclusion. As a second approach we used GSE to directly compare the biological processes represented in the 10- and 500-nm particle gene sets (‘‘group vs. group’’). This direct comparison can be justified because the numbers of genes affected in each group were very similar (503 vs. 502). A list of the major biological processes identified for both particle sizes is provided in the Supplemental Information (Table S3). We found that among 1009 total biological processes represented by these genes, none of these processes were statistically enriched (p < 0.05) in one particle size group over the other, even when a minimum gene number filter was not imposed. When this analysis was repeated to compare only the subset of pattern B genes (‘‘nanodominant,’’ 162 total) with the pattern C genes (‘‘microdominant,’’ 178 total), only one biological process for each size class was identified that met statistical significance (p < 0.05). For the nanodominant genes, the neurogenesis process was enriched and represented by seven genes, including Nr4a2, an orphan nuclear receptor that has been shown to promote inflammation in activated macrophages (Pei et al., 2005, 2006), and Egr2, an immediate early response gene, which is induced in macrophages exposed to fibrous TiO2 particles (Hirano et al., 2000). For the microdominant genes, the cation transport

MACROPHAGE RESPONSES TO SILICA NANOPARTICLES

FIG. 7. Cell processes perturbed by AS particles identified by GSE. Each of the major nonredundant cell processes identified is plotted as a function of significance (
log p value). The dotted line indicates the threshold for p ¼ 0.05.

process was enriched. This process includes genes involved in regulating transport of metals such as iron (Slc40a1) and copper (ceruloplasmin). Thus, although some of the gene expression changes identified in this study may reflect sizedependent or particle-specific transcriptional signatures based on predefined thresholds of gene induction, at the biological process level, the cellular response pathways activated are predominantly conserved independent of particle size when compared at equivalent nominal particle surface areas.

DISCUSSION

One of the central challenges facing the advancement of nanotechnology is developing effective strategies to determine whether the biocompatibility of a material is altered at the nanoscale. As an emerging field, the molecular mechanisms by which nanoparticles stimulate cellular responses are still poorly understood, yet it is clear that particle surface chemistry is an important driving factor (Oberdorster et al., 2005). Particle surface reactivity is impacted by many physicochemical parameters, including size, chemical composition, surface structure, shape, and solubility. The complexity of these physicochemical parameters and their potential interactions with biological systems make predicting the biocompatibility of a nanomaterial a significant challenge. Nonetheless, for nanomaterials which are increasingly used for diagnostic and biomedical purposes, such as AS, understanding how biocompatibility scales with particle size is critically important. To our knowledge, this is the first study to use genome-scale measurements to evaluate whether the cellular effects induced by nanomaterials are dependent on particle size. By identifying the early transcriptional events and pathways activated by AS in macrophages, our results significantly extend previous studies (Chen et al., 2004; Cho et al., 2007; Dutta et al., 2007; Lin et al., 2006; Sayes et al., 2007) that report AS nanoparticles stimulate inflammation and macro-

565

phage cytotoxicity. Many of the transcriptional changes stimulated by AS in RAW 264.7 cells recapitulate important events associated with the toxicity of particulates in vivo. As an example, the most robust transcriptional change observed was an induction of the Cxcl2 chemokine (MIP-2a). Studies demonstrating pretreatment of rats with anti-MIP-2a antibodies protects against crystalline silica-induced lung neutrophil recruitment indicate its importance as a early mediator of inflammation (Driscoll et al., 1996). TNF mRNA expression and protein shedding was also enhanced by AS nanoparticles in a dose-dependent manner. Neutralizing antibodies against TNF attenuate crystalline silica-induced lung inflammation in vivo (Driscoll, 2000), consistent with its role as an early initiator of chemokine expression. Our results are also consistent with a recent study that found lung mRNA and protein levels for MIP-2a, TNF, and MCP-2 (Ccl2) were significantly elevated in mice within 24 h after intratracheal instillation to 14 nm AS (Cho et al., 2007). Indeed, several chemokines that exhibit neutrophil chemotactic activity, including Ccl3 and Ccl4 were among the most robust early gene regulatory events associated with AS nanoparticle exposure in our study. The question of which metric of particle dose is most appropriate for evaluating biological effects of nanoparticles has stimulated recent debate (Lison et al., 2008; Oberdorster et al., 2007; Stoeger et al., 2007; Teeguarden et al., 2007; Wittmaack, 2007). Studies using rodents treated with TiO2 or carbonaceous ultrafine particulates have shown that markers of lung inflammatory response correlate more strongly with the particle surface area than with particle number or mass (Oberdorster et al., 2005; Stoeger et al., 2006). Wittmack (2007) challenged these interpretations and suggested particle number rather than surface area is a more suitable dose metric for evaluating dose-response relationships for particle-induced inflammation. The selection of appropriate in vitro dose metrics has also stimulated debate. Teeguarden et al. (2007) posited that nominal (total administered) particle media concentrations should be corrected for size- or density-dependent differences due to diffusion and gravitational settling which can impact delivery of particles to cell monolayers. This supposition is supported by studies which showed cellular uptake of small particles (25 nm) was consistent with diffusive delivery whereas large particle (320 nm) uptake was consistent with gravitational delivery to cells (Limbach et al., 2005). Based on studies using 29-nm AS nanoparticles, Lison and colleagues concluded that nominal dose (surface area, mass, and number) is proportional to the degree of cytotoxicity and cell uptake, and thus is an appropriate dose metric. Although diffusion and gravitational settling are known physical parameters that influence particle deposition, convection forces generated during typical particle administration practices used in vitro may have a greater impact on determining particle-cell interactions than gravitation or diffusion alone. In this regard, the alignment of a broad array of biological responses reported here with total administered surface area is supportive of the

566

WATERS ET AL.

appropriateness of nominal surface area as a dose metric for in vitro studies. The importance of understanding these relationships extends beyond providing guidance on how to effectively scale biological response data across experimental systems and particle types. Rather, understanding these dose relationships provides insight into whether the intrinsic surface chemistry of a material significantly changes as the material approaches the nanoscale. Our collective results are in agreement with Oberdorster et al. (2007), in that for AS, particle surface area is the most informative dose metric for comparing results across different particle sizes. Dose-response relationships for cytotoxicity and protein secretion responses based on particle number or mass varied by more than an order of magnitude across different particles sizes. In contrast, these relationships expressed on a surface area basis showed an excellent fit (coefficient of determination ¼ 21% for macrophage cytotoxicity) across a wide range of AS sizes and different particle manufacturers. Similarly, among more than 750 early gene expression changes identified, the majority (~76%) displayed a tighter correlation with surface area dose than with mass dose. These results suggest the inherent particle surface chemistry properties of AS that are responsible for these biological effects do not significantly change as particle size decreases to the nanoscale. Extrapolation of our in vitro dose-response results to the in vivo situation in the lung is challenging due to multiple cooperating cell types which contribute to inflammation processes. A recent study estimated that, for nonsoluble low-toxicity particles such as TiO2, the threshold for initiation of inflammation both within the proximal alveolar region in vivo and in cultured human lung cells is between 1- to 2-cm2 particle surface area per cm2 tissue/cell surface (Donaldson et al., 2008). Although evaluating response thresholds was not the focus of our study, our protein microarray results (Fig. 3) indicate a thresholdlike pattern of protein secretion in macrophages where the initial increase in secretion of several cytokines in response to AS begins at approximately 10 cm2/ml AS. Under the experimental conditions used, this equates to a nominal in vitro dose of 2.5cm2 particle surface area per cm2 of cell monolayer, which closely corresponds to the upper bound estimate for initiation of inflammation by particles in the proximal lung alveolar region (Donaldson et al., 2008). Although these estimates involve simplifying assumptions of the underlying biology, the correspondence between in vitro and in vivo reinforces the utility of surface area as a comparative dose metric. Whereas the majority of genes expression changes identified were affected by both 10- and 500-nm particles, we also identified genes that were preferentially regulated by specific particles. For instance, ~23% of the genes identified showed a greater magnitude of transcriptional change in response to 500 nm AS compared with 10-nm particles. In some cases, these differences may reflect true ‘‘particle-specific’’ differences in gene regulation. For instance, the preferential induction of the metal transport proteins ceruloplasmin and

Slc40a1 by 500-nm particles could indicate trace contamination of divalent metals in this particle that were not identified by our analyses. Several genes within the Kruppel-like factor family of transcription factors were preferentially induced by 10-nm particles. The function of this family of transcriptional suppressors in macrophages is unclear, although they have been associated with erythroid differentiation. Whether these transcription factors control different transcriptional networks in nanomaterials compared with larger particles will require identification and further study of the downstream genes they regulate in macrophages. In other cases, the differences in mRNA abundance identified between the 10- and 500-nm particle groups may simply reflect more subtle temporal differences in gene regulation that are not reflected at a single time point. Shifts in the time course for gene induction following exposure to nano- and supranano-sized particles could be expected for several reasons, including due to differences in the dynamics of particle internalization. For relatively large particles (500 nm), phagocytosis is expected to be a central internalization pathway in macrophages. The mechanism(s) for nanoparticle uptake are poorly understood, although processes ranging from diffusion-limited uptake to receptor-mediated endocytosis have been reported (Geiser et al., 2005; Kanno et al., 2007). A recent study found a critical cut-off for receptor-mediated endocytosis of nanoparticles to be 40–50 nm (Jiang et al., 2008). These distinct pathways occur not only with different dynamics, but would result in localization of the particles with different intracellular proteins and organelles that could modulate downstream signaling events. Elucidating these differences will require more extensive time course and cell uptake studies, which are underway in our laboratory. The unbiased and comprehensive nature of ‘omics platforms such as microarray can play an important role in defining potential toxicity pathways induced by emerging nanomaterials. However, interpretation of these high-dimensional data requires careful scrutiny and placing individual gene expression changes into a larger biological context. Biological interpretation of genome-level data can significantly depend on the arbitrary statistical thresholds for gene expression that are set (Shi et al., 2006). This issue is also evident in our study, where the differences in gene expression changes we observed between 10- and 500-nm particles were often subtle, but due to the significance threshold used led to different classifications despite similar directional dose-response trends. With this in mind and to put the gene expression changes identified into a broader context, we extended our analytical approach by combining methods for clustering dose-response profiles with GSE analysis to identify major biological processes and pathways represented. Although GSE analysis is limited by the accuracy of gene annotations, it has proven valuable in identifying patterns in pathway regulation and disease states that are otherwise difficult to discern from large gene lists (Huang et al., 2003). Using two different GSE comparison

MACROPHAGE RESPONSES TO SILICA NANOPARTICLES

approaches, we find that despite identified differences in the magnitude of change in expression for individual genes affected by different sizes of AS, the composite biological processes represented by these genes were nearly identical. These results provide additional credence to the hypothesis that the differences between the (cyto)toxicity of nano- and supranano-sized AS observed in our study, and potentially previous studies (Kaewamatawong et al., 2005), is primarily a function of effective surface area dose rather than a inherent shift in particle bioreactivity or biological mode of action. The high level of similarity in genes and cell processes modulated by 10 and 500 nm AS we observed suggests overlapping modes of action are involved. This raises intriguing mechanistic questions. For instance, it is uncertain whether the macrophage responses we observed are initiated nonselectively by particle contact with the cell membrane, or whether some gene regulatory events require particle internalization or specific receptor binding events. Previous studies have shown that macrophage apoptosis following exposure to unopsonized crystalline silica (micron diameter) is at least partially dependent on scavenger receptor function (Chao et al., 2001; Hamilton et al., 2006). Whether cellular responses to AS nanoparticles are receptor mediated has yet to be determined. However, prior studies from our group have found that pretreatment of AS with surfactants that prevent binding of scavenger receptor ligands to nanomaterials protects RAW 264.7 cells against the cytotoxicity of AS nanoparticles (Dutta et al., 2007). Future targeted investigations of the role of specific membrane receptors in mediating AS-mediated uptake and response will allow a better understanding of the initial cellular sites of action for this nanomaterial. Such studies should elucidate strategies for enhancing the beneficial applications of this material in biological systems, whereas reducing potential adverse effects mediated through macrophage interactions. In addition, the general strategy illustrated in this study can be extended to other nanomaterials and cell types both to identify modes of action and to prioritize more expensive and time-consuming in vivo toxicity studies. Such a strategy would be particularly valuable for materials for which toxicity data exists for larger size particles (such as with AS), but information is lacking regarding the potential effects of the material at the nano-scale.

SUPPLEMENTARY DATA

Supplementary data are available online at http://toxsci. oxfordjournals.org/

FUNDING

National Institutes of Health (ES016212) to B.D.T.; and Laboratory Directed Research and Development program at

567

Pacific Northwest National Laboratory (PNNL); and PNNL is operated by Battelle for the U.S. Department of Energy under contract (AC06-76RLO 1830).

ACKNOWLEDGMENTS

We thank Dr Colette Sacksteder for FTIR measurements and Mr Ruimin Tan for technical assistance with microarray analysis.

REFERENCES Al-Shahrour, F., Diaz-Uriarte, R., and Dopazo, J. (2004). FatiGO: A web tool for finding significant associations of Gene Ontology terms with groups of genes. Bioinformatics 20, 578–80. Benjamini, Y., and Hochberg, Y. (1995). Controlling the false discovery rate—A practical and powerful approach to multiple testing. J. Roy. Stat. Soc. B 57, 289–300. Bermudez, E., Mangum, J. B., Wong, B. A., Asgharian, B., Hext, P. M., Warheit, D. B., and Everitt, J. I. (2004). Pulmonary responses of mice, rats, and hamsters to subchronic inhalation of ultrafine titanium dioxide particles. Toxicol. Sci. 77, 347–57. Bolstad, B. M., Irizarry, R. A., Astrand, M., and Speed, T. P. (2003). A comparison of normalization methods for high density oligonucleotide array data based on variance and bias. Bioinformatics 19, 185–93. Brown, D. M., Wilson, M. R., MacNee, W., Stone, V., and Donaldson, K. (2001). Size-dependent proinflammatory effects of ultrafine polystyrene particles: A role for surface area and oxidative stress in the enhanced activity of ultrafines. Toxicol. Appl. Pharmacol. 175, 191–9. Brumfiel, G. (2006). Consumer products leap aboard the nano bandwagon. Nature 440, 262. Chao, S. K., Hamilton, R. F., Pfau, J. C., and Holian, A. (2001). Cell surface regulation of silica-induced apoptosis by the SR-A scavenger receptor in a murine lung macrophage cell line (MH-S). Toxicol. Appl. Pharmacol. 174, 10–6. Chen, M., and von Mikecz, A. (2005). Formation of nucleoplasmic protein aggregates impairs nuclear function in response to SiO2 nanoparticles. Exp. Cell Res. 305, 51–62. Chen, Y., Chen, J., Dong, J., and Jin, Y. (2004). Comparing study of the effect of nanosized silicon dioxide and microsized silicon dioxide on fibrogenesis in rats. Toxicol. Ind. Health 20, 21–7. Cho, W. S., Choi, M., Han, B. S., Cho, M., Oh, J., Park, K., Kim, S. J., Kim, S. H., and Jeong, J. (2007). Inflammatory mediators induced by intratracheal instillation of ultrafine amorphous silica particles. Toxicol. Lett. 175, 24–33. Donaldson, K., Borm, P. J., Oberdorster, G., Pinkerton, K. E., Stone, V., and Tran, C. L. (2008). Concordance between in vitro and in vivo dosimetry in the proinflammatory effects of low-toxicity, low-solubility particles: The key role of the proximal alveolar region. Inhal. Toxicol. 20, 53–62. Driscoll, K. E. (2000). TNFalpha and MIP-2: Role in particle-induced inflammation and regulation by oxidative stress. Toxicol. Lett. 112–113, 177–83. Driscoll, K. E., Howard, B. W., Carter, J. M., Asquith, T., Johnston, C., Detilleux, P., Kunkel, S. L., and Isfort, R. J. (1996). Alpha-quartz-induced chemokine expression by rat lung epithelial cells: Effects of in vivo and in vitro particle exposure. Am. J. Pathol. 149, 1627–37.

568

WATERS ET AL.

Dutta, D., Sundaram, S. K., Teeguarden, J. G., Riley, B. J., Fifield, L. S., Jacobs, J. M., Addleman, S. R., Kaysen, G. A., Moudgil, B. M., and Weber, T. J. (2007). Adsorbed proteins influence the biological activity and molecular targeting of nanomaterials. Toxicol. Sci. 100, 303–15. Ferrin, J., Oberdorster, G., and Penney, D. (1992). Pulmonary retention of ultrafine and fine particulates in rats. Am. J. Respir. Cell Mol. Biol. 6, 535–542. Fond, A. M., and Meyer, G. J. (2006). Biotoxicity of metal oxide nanoparticles. In Nanomaterials—Toxicity, Heath and Environmental Issues (C. S. S. R. Kumar, Ed.), Vol. 5, pp. 3–33. Wiley-VCH, Weinheim, Germany. Geiser, M., Rothen-Rutishauser, B., Kapp, N., Schurch, S., Kreyling, W., Schulz, H., Semmler, M., Im Hof, V., Heyder, J., and Gehr, P. (2005). Ultrafine particles cross cellular membranes by nonphagocytic mechanisms in lungs and in cultured cells. Environ. Health Perspect. 113, 1555–60. Goodman, C. M., McCusker, C. D., Yilmaz, T., and Rotello, V. M. (2004). Toxicity of gold nanoparticles functionalized with cationic and anionic side chains. Bioconjugate Chem. 15, 897–900. Hamilton, R. F., Jr., Thakur, S. A., Mayfair, J. K., and Holian, A. (2006). MARCO mediates silica uptake and toxicity in alveolar macrophages from C57BL/6 mice. J. Biol. Chem. 281, 34218–26. Hirano, S., Anuradha, C. D., and Kanno, S. (2000). Transcription of krox-20/ egr-2 is upregulated after exposure to fibrous particles and adhesion in rat alveolar macrophages. Am. J. Respir. Cell Mol. Biol. 23, 313–9. Huang, E., Ishida, S., Pittman, J., Dressman, H., Bild, A., Kloos, M., D’Amico, M., Pestell, R. G., West, M., and Nevins, J. R. (2003). Gene expression phenotypic models that predict the activity of oncogenic pathways. Nat. Genet. 34, 226–30. Iijima, M., Tsukada, M., and Kamiya, H. (2007). Effect of particle size on surface modification of silica nanoparticles by using silane coupling agents and their dispersion stability in methylethylketone. J. Colloid Interface Sci. 307, 418–24. Ji, P., Diederichs, S., Wang, W., Boing, S., Metzger, R., Schneider, P. M., Tidow, N., Brandt, B., Buerger, H., Bulk, E., et al. (2003). MALAT-1, a novel noncoding RNA, and thymosin beta4 predict metastasis and survival in early-stage non-small cell lung cancer. Oncogene 22, 8031–41. Jiang, W., Kim, B. Y., Rutka, J. T., and Chan, W. C. (2008). Nanoparticlemediated cellular response is size-dependent. Nat. Nanotechnol. 3, 145–50. Kaewamatawong, T., Kawamura, N., Okajima, M., Sawada, M., Morita, T., and Shimada, A. (2005). Acute pulmonary toxicity caused by exposure to colloidal silica: Particle size dependent pathological changes in mice. Toxicol. Pathol. 33, 743–9. Kanno, S., Furuyama, A., and Hirano, S. (2007). A murine scavenger receptor MARCO recognizes polystyrene nanoparticles. Toxicol. Sci. 97, 398–406. Kerr, M. K., Martin, M., and Churchill, G. A. (2000). Analysis of variance for gene expression microarray data. J. Comput. Biol. 7, 819–37. Khuu, C. H., Barrozo, R. M., Hai, T., and Weinstein, S. L. (2007). Activating transcription factor 3 (ATF3) represses the expression of CCL4 in murine macrophages. Mol. Immunol. 44, 1598–605. Limbach, L. K., Li, Y., Grass, R. N., Brunner, T. J., Hintermann, M. A., Muller, M., Gunther, D., and Stark, W. J. (2005). Oxide nanoparticle uptake in human lung fibroblasts: Effects of particle size, agglomeration, and diffusion at low concentrations. Environ. Sci. Technol. 39, 9370–6. Lin, W., Huang, Y. W., Zhou, X. D., and Ma, Y. (2006). In vitro toxicity of silica nanoparticles in human lung cancer cells. Toxicol. Appl. Pharmacol. 217, 252–9. Lison, D., Thomassen, L. C., Rabolli, V., Gonzalez, L., Napierska, D., Seo, J. W., Kirsch-Volders, M., Hoet, P., Kirschhock, C. E., and Martens, J. A. (2008). Nominal and effective dosimetry of silica nanoparticles in cytotoxicity assays. Toxicol. Sci. 104, 155–62. Nel, A., Xia, T., Madler, L., and Li, N. (2006). Toxic potential of materials at the nanolevel. Science 311, 622–7.

Oberdorster, G. (2000). Toxicology of ultrafine particles: In vivo studies. Philos. Trans. R. Soc. Lond. A 358, 2719–2740. Oberdorster, G., Oberdorster, E., and Oberdorster, J. (2005). Nanotoxicology: An emerging discipline evolving from studies of ultrafine particles. Environ. Health Perspect. 113, 823–39. Oberdorster, G., Oberdorster, E., and Oberdorster, J. (2007). Concepts of nanoparticle dose metric and response metric. Environ. Health Perspect. 115, A290. Pei, L., Castrillo, A., Chen, M., Hoffmann, A., and Tontonoz, P. (2005). Induction of NR4A orphan nuclear receptor expression in macrophages in response to inflammatory stimuli. J. Biol. Chem. 280, 29256–62. Pei, L., Castrillo, A., and Tontonoz, P. (2006). Regulation of macrophage inflammatory gene expression by the orphan nuclear receptor Nur77. Mol. Endocrinol. 20, 786–94. Perez, D. S., Hoage, T. R., Pritchett, J. R., Ducharme-Smith, A. L., Halling, M. L., Ganapathiraju, S. C., Streng, P. S., and Smith, D. I. (2008). Long, abundantly expressed non-coding transcripts are altered in cancer. Human Mol. Genet. 17, 642–55. Saeed, A. I., Sharov, V., White, J., Li, J., Liang, W., Bhagabati, N., Braisted, J., Klapa, M., Currier, T., Thiagarajan, M., et al. (2003). TM4: A free, opensource system for microarray data management and analysis. BioTechniques 34, 374–8. Sayes, C. M., Reed, K. L., and Warheit, D. B. (2007). Assessing toxicity of fine and nanoparticles: Comparing in vitro measurements to in vivo pulmonary toxicity profiles. Toxicol. Sci. 97, 163–80. Shi, L., Reid, L. H., Jones, W. D., Shippy, R., Warrington, J. A., Baker, S. C., Collins, P. J., de Longueville, F., Kawasaki, E. S., Lee, K. Y., et al. (2006). The MicroArray Quality Control (MAQC) project shows inter- and intraplatform reproducibility of gene expression measurements. Nat. Biotechnol. 24, 1151–61. Son, S. J., Reichel, J., He, B., Schuchman, M., and Lee, S. B. (2005). Magnetic nanotubes for magnetic-field-assisted bioseparation, biointeraction, and drug delivery. J. Am. Chem. Soc. 127, 7316–7317. Stoeger, T., Reinhard, C., Takenaka, S., Schroeppel, A., Karg, E., Ritter, B., Heyder, J., and Schulz, H. (2006). Instillation of six different ultrafine carbon particles indicates a surface area threshold dose for acute lung inflammation in mice. Environ. Health Perspect. 114, 328–33. Stoeger, T., Schmid, O., Takenaka, S., and Schulz, H. (2007). Inflammatory response to TiO2 and carbonaceous particles scales best with BET surface area. Environ. Health Perspect. 115, A290– A291; author reply A291–2. Teeguarden, J. G., Hinderliter, P. M., Orr, G., Thrall, B. D., and Pounds, J. G. (2007). Particokinetics in vitro: Dosimetry considerations for in vitro nanoparticle toxicity assessments. Toxicol. Sci. 95, 300–312. Turner, J., and Crossley, M. (1999). Basic Kruppel-like factor functions within a network of interacting haematopoietic transcription factors. Int. J. Biochem. Cell Biol. 31, 1169–74. Varnum, S. M., Covington, C. C., Woodbury, R. L., Petritis, K., Kangas, L. J., Abdullah, M. S., Pounds, J. G., Smith, R. D., and Zangar, R. C. (2003). Proteomic characterization of nipple aspirate fluid: Identification of potential biomarkers of breast cancer. Breast Cancer Res. Treat. 80, 87–97. Weber, T. J., Smallwood, H. S., Kathmann, L. E., Markillie, L. M., Squier, T. C., and Thrall, B. D. (2006). Functional linkage between tumor necrosis factor biosynthesis and calmodulin-dependent activation of iNOS in RAW 264.7 macrophages. Am. J. Physiol. Cell Physiol. 290, C1512–C1520. Wen, L.-X., Ding, H.-M., Want, J.-X., and Chen, J.-F. (2006). Porous hollow silica nanoparticles as carriers for controlled deliver of ibuprofen to small intestine. J. Nanosci. Nanotechnol. 6, 3139–3144. White, A. M., Daly, D. S., Varnum, S. M., Anderson, K. K., Bollinger, N., and Zangar, R. C. (2006). ProMAT: Protein microarray analysis tool. Bioinformatics (Oxford, England) 22, 1278–9.

MACROPHAGE RESPONSES TO SILICA NANOPARTICLES

569

Wiethoff, A. J., Reed, K. L., Webb, T. R., and Warheit, D. B. (2003). Assessing the role of neutrophil apoptosis in the resolution of particle-induced pulmonary inflammation. Inhal. Toxicol. 15, 1231–46.

Woodbury, R. L., Varnum, S. M., and Zangar, R. C. (2002). Elevated HGF levels in sera from breast cancer patients detected using a protein microarray ELISA. J. Proteome Res. 1, 233–7.

Wittmaack, K. (2007). In search of the most relevant parameter for quantifying lung inflammatory response to nanoparticle exposure: Particle number, surface area, or what? Environ. Health Perspect. 115, 187–94.

Zangar, R. C., Daly, D. S., and White, A. M. (2006). ELISA microarray technology as a high-throughput system for cancer biomarker validation. Expert Rev. Proteomics 3, 37–44.