Assessing the Relevance of in vitro Studies in Nanotoxicology by Examining Correlations between in vitro and in vivo Data

Abstract

There is an urgent need for in vitro screening assays to evaluate nanoparticle (NP) toxicity. However, the relevance of in vitro assays is still disputable. We administered doses of TiO₂ NPs of different sizes to alveolar epithelial cells in vitro and the same NPs by intratracheal instillation in rats in vivo to examine the correlation between in vitro and in vivo responses. The correlations were based on toxicity rankings of NPs after adopting NP surface area as dose metric, and response per unit surface area as response metric. Sizes of the anatase TiO₂ NPs ranged from 3 to 100 nm. A cell-free assay for measuring reactive oxygen species (ROS) was used, and lactate dehydrogenase (LDH) release, and protein oxidation induction were the in vitro cellular assays using a rat lung Type I epithelial cell line (R3/1) following 24 hr incubation. The in vivo endpoint was number of PMNs in bronchoalveolar lavage fluid (BALF) after exposure of rats to the NPs via intratracheal instillation. Slope analyses of the dose response curves shows that the in vivo and in vitro responses were well correlated. We conclude that using the approach of steepest slope analysis offers a superior method to correlate in vitro with in vivo results of NP toxicity and for ranking their toxic potency.

1. Introduction

The exciting field of nanotechnology has seen a rapid expansion in recent years. However, there is a growing concern about potential risks posed by the technology due to exposure of humans and the environment (Stern and McNeil, 2008).

Current toxicological research on nanotechnology mainly focuses on the adverse effects of one of the building blocks of the technology – nanoparticles (NPs). There are several challenges that seem to be special to nanotoxicology when evaluating the adverse effects of NPs. One challenge is whether to use mass as a dose metric for assessing NP toxicity. Since the activity of NPs is determined to a large extent by NP surface properties, there has been efforts to determine if a dose metric based on surface area is preferable rather than one based on mass (Brown et al., 2001; Oberdörster et al., 2005a, 2005b, and 2007; Teeguarden et al., 2007; Monteiller et al., 2007; Wittmaack, 2007; Stoeger et al., 2007).

Another challenge is to validate in vitro assays that have been well established for assessing toxicity of chemicals. There is an urgent need for quick and reliable in vitro screening assays to replace or reduce the slow, costly and ethically controversial animal testing that might be required due to the rapid development and commercialization of nano-enabled products. However, some traditional in vitro assays have been found to generate misleading results because some of the nanomaterials could interfere with the assays (Wörle-Knirsch et al., 2006; Casey et al., 2007; Belyanskaya et al., 2007; Han et al., 2011). In addition, NPs, because of their large specific surface area, could adsorb essential nutrients in cell culture medium, making it difficult to interpret some cytotoxicity results (Guo et al., 2008).

When evaluating NP toxicity using in vitro assays, in vivo relevance is an essential criterion for accepting their utility. The in vivo relevance can be questioned because of the differences between in vitro and in vivo conditions. These differences warrant developing novel methods to define equivalent doses between in vitro and in vivo exposures in order to improve correlations between the two testing systems.

One difference is the high concentrations/dose used in most traditional in vitro studies. An extremely high dose rate (dose administered per unit of time) is another issue of in vitro studies because the full dose is delivered as a bolus in traditional in vitro assays. The dose rate in such studies is much higher than in in vivo inhalation studies in which animals are exposed to a low concentration of chemical or particle via inhalation for an extended period of time (hours, days, weeks, or longer). Another important difference is the wide use of dispersants in vitro but not necessarily in vivo (e.g., inhalation of pristine NPs generated from dry powders) (Park et al., 2009). NPs dispersed in cell culture medium would adsorb some components in the medium while NPs inhaled into the lung would adsorb components of pulmonary surfactant. However, the use of high doses and high dose rates in vitro by itself does not invalidate in vitro assays. The in vitro data could still be useful as long as the data are confirmed to correlate well with in vivo results. In particular, in vivo dosing of the respiratory tract by intratracheal instillation or oro-pharyngeal aspiration are also bolus-type delivery methods and results should correlate better with in vitro dosing.

There have been some studies that addressed the issue of the in vivo relevance of some in vitro assays for evaluating the toxicity of NPs (Sayes et al., 2007) or ambient particulate matter (Seagrave et al., 2005). Both found poor correlations. However, these findings do not necessarily indicate intrinsic flaws of the in vitro assays for predicting in vivo toxicity. Instead, there can be multiple reasons for the differences between in vitro and in vivo results (Seagrave et al., 2005). Our group (Rushton et al., 2010) has proposed an alternative approach of slope analysis and found a good in vitro – in vivo correlation when applied to the data in the paper of Sayes et al. (2007).

The objective of this work is to test the hypothesis that results of in vitro assays correlate with acute in vivo effects when an appropriate response metric is used. Therefore, in this work, we continue to address the relevance of in vitro assays by comparing results of in vitro and in vivo studies for NPs of different sizes based on a slope analysis of dose-response curves. In addition to the graphical method used in our previous work to determine the maximum response per unit of the dose (steepest slope) (Rushton et al., 2010), we introduce a new method to derive the steepest slope – a mathematical method. We report in this work that using this approach showed a good correlation of acute toxicity between in vitro and in vivo results. Future work needs to also consider extension to long-term effects.

2. Methods

2.1. Materials

NPs used for the main study of this work includes different TiO₂ NPs (for correlation study), and for assay validation only a silver NP, and a copper NP. A series of anatase TiO₂ NPs of different sizes (3, 7, 10, 16, 30, 50, 53, and 104 nm) were used for the correlation study. These NPs were synthesized from titanium tetra-isopropoxide (TTIP, 97%, Aldrich) either in a diffusion or premixed flame aerosol reactor (Jiang et al., 2007). These two methods for preparing NPs are commonly used, although it was unknown if both methods could generate particles with different toxicological activities. Thus, NPs generated by both methods were considered as different NPs even if they had the same average size. The other NPs (P-25, TIO₂, Ag-40, and Cu-40) were only used for validating the protein carbonylation/oxidation assay (see Supplemental Information).

2.2. Physicochemical characterization of NPs

The hydrodynamic sizes of the anatase TiO₂ NPs in different media were measured using dynamic light scattering (DLS) technology (intensity distribution of suspended NPs) with a Malvern ZetaSizer (ZetaSizer Nano ZS). Number or volume distribution of the particles would generally incur larger errors because they are not direct measurements; instead, they are derived from the directly measured intensity distribution. The NPs were sonicated using a cuphorn sonicator for 15 sec twice (Sonics VCX 750 with maximum output 750 watts and a frequency of 20 kHz, Newtown, CT) at 29% maximum output. Suspension and cell culture media included saline (pH 5.9), phosphate buffered saline (PBS) with a pH of 7.4 (1.14 g of Na₂HPO₄, 0.72 g of NaH₂PO₄, and 8.18 g of NaCl in deionized water, pH adjusted to 7.4 using 10 N NaOH), RPMI 1640 cell culture medium with no phenol red (Invitrogen GIBCO® 11835), and RPMI medium containing 1% fetal bovine serum (FBS, Hyclone, SH30070.03). For each size of anatase TiO₂ NP, the same concentration was used (2 μg/ml). Measurements were repeated at least 6 times to ensure a stable reading; if not, measurements would be repeated further until there is no obvious time trend of the measured sizes as judged by examining the last 6 measurements. In addition, primary particle sizes and agglomeration state were also measured using transmission electron microscope. BET isotherms (Autosorb-1, Quantachrome) was used to determine the specific surface area of the NPs (Brunauer et al., 1938).

2.3. Rationale of choosing in vitro and in vivo endpoints

There is increasing evidence that the induction of oxidative stress through generation of reactive oxygen species (ROS) plays a critical role in the mechanism of toxicity induced by NPs (Brown et al., 2001; Xiao et al., 2003; Hussain et al., 2005; Nel et al., 2006; Long et al., 2006; Sayes et al., 2006; Federici et al., 2007; Limbach et al., 2007). Based on this, most of the endpoints we selected for the in vitro and in vivo analyses were related to oxidative stress or inflammation. The oxidative stress related endpoints in three testing systems were ROS generation in a cell-free system, protein carbonylation in cultured cells, and the number of polymorphonuclear neutrophils (PMNs) in lung lavage fluid. In addition, we also included measurement of released lactate dehydrogenase (LDH) from cultured cells. Though LDH release is caused by cellular membrane damage which may or may not be related to oxidative stress, the LDH assay is a very commonly used assay and warrants evaluation for rapid screening of NP toxicity. TiO₂ particles could interfere with the LDH assay by adsorbing LDH molecules (Han et al., 2011), however, there is only minor interference at low and intermediate concentration so the assay will still be useful as long as its in vivo relevance can be shown.

The endpoints chosen for this study are general and broad-based because the purpose is to choose relatively simple screening assays that are widely applicable so that the toxicity of a large number and variety of different NPs can be evaluated.

2.4. Cell-free ROS assay

The intrinsic capability of the NPs in generating reactive oxygen species (ROS) was evaluated using a cell-free ROS assay (Jiang et al., 2008). The assay was performed in the dark to avoid any interference from photocatalytic activities of the NPs. The dye 2′,7′-dichlorofluorescein diacetate (DCFH-DA) (CalBiochem, 287810) is deacetylated and oxidized into a fluorescent product in the presence of NP-generated free radicals. The fluorescence was detected using a spectrofluorometer (absorbance/emission maxima, 485 nm/535 nm).

Three to six groups (including controls) were used for these NPs with concentration ranges from 0 to 100 μg/ml and a dose spacing factor of 2 or 3. When expressed as NP surface area, the highest concentration of 100 μg/ml is equivalent to 30 cm²/ml for the largest TiO₂ NP and 426 cm²/ml for the smallest NP. Since ROS level in the cell-free assay increased linearly with increasing NP concentration (see “Results” below), the slope of this increase for each NP was obtained using linear regression. In addition to mass concentration (μg/ml), surface area concentration (cm²/ml) was also used to express the slopes. After calculating the mass based slope for each NP, the surface area based slope was calculated by dividing the mass based slope by the specific surface area (SSA) of the NP.

2.5. Cell culture

A rat lung epithelial Type I like cell line, R3/1 (Koslowski et al., 2004), was used for the cellular in vitro studies. Cells were maintained in RPMI 1640 cell culture medium (without phenol red) containing 10% fetal bovine serum (FBS) and passaged every three or four days. For in vitro assays, cells with a density of 10⁵ per ml were seeded in plates, grown to subconfluency (70–80% confluency), and exposed to NPs in RPMI 1640 with 1% FBS for different time durations. Though plates with wells of different sizes (6-, 12-, or 24-well plate) were used, when seeding cells into a plate, the same cell density was used and a constant depth of cell culture medium was maintained. NPs were probe sonicated in RPMI 1640 with 1% FBS on ice for 10 sec twice.

2.6. LDH assay

The LDH assay was performed according to the Sigma LDH assay kit. The assay works by measuring the catalytic activity of LDH in the reaction: NADH + Pyruvate ↔ NAD⁺ + Lactate, where NADH stands for reduced β-nicotinamide adenine dinucleotide. The decreasing rate of NADH – measured as decreasing absorbance at 340 nm over two minutes – is used as an indicator of LDH activity.

Eight groups (including controls) were used for this study with concentrations ranging from 0 to 200 μg/ml for larger particles (50 nm), 0 to 100 μg/ml for medium sized particles (10 and 30 nm), and 0 to 50 μg/ml for smaller particles (3 and 7 nm). The spacing factor was 2 or 2.5.

2.7. Protein carbonylation study

In this study, enzyme-linked immunosorbent assay (ELISA) was used for detecting protein carbonyls following their reaction with dinitrophenylhydrazine (DNPH) (Buss et al., 1997).

The materials and their sources were as follows: protease inhibitors (Sigma, P2714), ascorbic acid (Sigma, 255564), ferrous ammonium sulfate (Sigma, 215406), bovine serum albumin (BSA, Sigma, A9056), NaBH₄ (Sigma, 213462), 2,4-dinitrophenylhydrazine (Sigma, D199303), 3,3,5,5-tetramethylbenzidine (TMB) tablet (Sigma, T-3405), hydrogen peroxide (H₂O₂, 30%) (Sigma, H-1009), guanidine hydrochloride (Sigma, G3272), anti-DNP antibody (Sigma, D-9656), anti-rabbit antibody HRP-linked IgG (Cell Signaling Technology, 7074), flat ELISA plate (high binding, Greiner bio-one, 756071), filter (Pall Life Sciences, 0.2 μm Super Membrane), EDTA disodium salt (J. T. Baker, 8993-1), phosphoric acid (85%) (EM Science, PX0995-13), 2N HCl diluted from concentrated HCl (Seastar), NaOH (J. T. Baker, 3722-04), Na₂HPO₄ (S0876), NaH₂PO₄ (S0751), and Citric acid (Sigma 251275). Phosphate buffered saline (PBS) contains 10 mM phosphate buffer (8 mM of Na₂HPO₄ and 2 mM of NaH₂PO₄) in 140 mM NaCl with a pH of 7.4. DNPH solution contains 10 mM DNPH in 6 M guanidine hydrochloride and about 0.29 M potassium phosphate buffer with a pH of 2.5.

The bicinchoninic acid (BCA) assay was used to measure protein concentration for the samples as described in the next section. Based on the protein concentrationprotein concentration was adjusted to 1 mg/ml for reaction with 10 mM DNPH in a volume ratio of 1:3. Subsequent ELISA steps follow the paper by Buss et al. (Buss et al., 1997), except that some antibodies and a few reagents were suggested in another paper (Alamdari et al., 2005).

Eight groups (including controls) were used for this study with dose range from 0 to 200 μg/ml for larger particles (50 nm), 0 to 100 μg/ml for medium sized particles (10 and 30 nm), and 0 to 50 μg/ml for smaller particles (3 and 7 nm). Dose spacing factor was 2 or 2.5.

2.8. The bicinchoninic acid (BCA) assay

The BCA assay was used to measure protein concentration for the carbonylation samples and total protein levels of lung lavage fluid samples. It was based on a kit from Thermo Scientific (Thermo Scientific Pierce BCA Protein Assay, PI-23235).

2.9. Animal studies

Male Fischer 344 rats were purchased from Harlan Laboratories and housed in the University Medical Center animal facilities. All procedures were approved by the University Committee on Animal Resources (UCAR) of the University of Rochester.

Rats had body weights ranging about 250–300 g and were divided into four groups (three dose groups and a control group) with a group size of three or four rats (See Figure 9 for doses). NPs were dispersed in a sterile physiological saline (Hospira, NDC 0409-4888-10), probe sonicated for 10 sec, and immediately given to animals via intratracheal instillation.

Fig. 9.

Slopes of PMN data from the anatase TiO₂ exposure study in rats as a function of particle size (graphical method). Each of the slopes was obtained using a graphical method in which the largest slope from a dose-response curve was picked directly from the slopes of adjacent pairs of dose groups. While there is a decreasing trend for slope expressed in mass (10⁶ PMNs per μg/rat) (a), surface area based slope (10⁶ PMNs per cm²/rat) increases with increasing size (b). These slopes will be used as in vivo data for correlating with in vitro slopes. These NPs were prepared using diffusion (marked with “D”) or premixed flame method (marked with “P”).

The rats were sacrificed 24 hr after intratracheal instillation to evaluate pulmonary inflammation. Euthanasia was induced with an i.p. injection of 0.3–0.5 ml of sodium pentobarbital (64.8 mg/ml, Hospira, NDC 67384-501-52).

The lungs were lavaged 5 times with 5 ml sterile saline (Sodium Chloride injection, UPS 0.9% sterile from Abbott Labs, North Chicago, IL 60064) each time. The 1^st and 2^nd lavage fluid was put into a 15 ml polypropylene centrifuge tube. The 3^rd – 5^th lavage fluid was placed in a 50 ml polypropylene tube. The supernatant from lavages 3–5 was discarded while that from lavages 1–2 was saved (aliquots of 100 μl, −20 °C) for measuring protein level and same day analysis of LDH and β-glucuronidase (400 μl, 4 °C). The pellets from both tubes were resuspended in 0.9% NaCl and combined in a final volume of 10 ml. The suspension was used for differential cell counts (PMNs and other cells) and viability measurements.

The β-glucuronidase activity in BALF was determined by measuring its activity in the reaction 4 – nitrophenyl – glucuronide ↔ p – nitrophenol + monosaccharide (Stahl and Fishman, 1984). LDH was measured according the method described in Section 2.6. Protein levels were measured using the BCA method described in Section 2.8.

2.10. Dose- and response-metrics

The issue of response metric is related to the shape of a dose-response curve. At the point with the steepest slope (Fig. 1a), a slight increase in dose leads to a remarkable increase in response. In this work, “steepest” slope and “maximum” slope are used interchangeably. The word “maximum” is often used in mathematics while “steepest” seems to be more instinctive when using the graphical method as described below.

Fig. 1.

Calculation of the steepest slope using the mathematical method. (a). Fit a dose-response curve and get y=f(x). (b). Take the first derivative y′ = f′(x) and solve the equation. (c). Obtain the second derivative y″ = f″(x), solve the equation, let it be zero, and get x, which is actually D_MS, the dose where the maximum slope is attained. Put D_MS back into y′ and obtain the steepest slope.

To determine the steepest slope, two methods were used: a graphical method and a mathematical method. In the graphical method, the response data are plotted against the dose, the slope for each pair of adjacent dose groups is calculated by dividing the response difference by the dose difference, and the largest slope is selected. The response difference is the difference between the two average responses of the two adjacent dose groups. To get a confidence interval for the steepest slope, all the individual data points in the two dose groups are used for a linear regression analysis.

In the mathematical method (Fig. 1), the first step is to fit a sigmoidal curve for dose-response data where appropriate, from which an equation is obtained:

$$y=b+\frac{t-b}{1+{10}^{h(log{ED}_{50}-logx)}},$$ (Equation 1)

where y is response; b is bottom, i.e., the background level of response; t is Top, meaning the maximum response that can be asymptotically obtained as dose increases indefinitely; h is the Hill slope, the steepness factor of the curve when taking logarithm on dose x; and ED₅₀ is the dose to attain half of the maximum response after adjusting for background response.

Subsequently, the first derivative of Equation 1 with respect to dose x is calculated to get an equation corresponding to the so-called “slope curve”:

$$y^{\prime }=h(t-b)\frac{{{ED}_{50}}^h{x}^{h-1}}{{(x^h+{{ED}_{50}}^h)}^2}.$$ (Equation 2)

To get the maximum value of the “slope curve”, derivative is taken on Equation 2 (i.e., to get the second derivative of Equation 1, y″; y″ is still a function of the dose x, the complex formula for y″ is not shown since it is not used for routine calculation). By solving y″=0, the dose where the steepest slope gets across (D_MS, Dose for maximum slope) is obtained (Equation 3). D_MS exists only when the Hill slope in Equation 1 is larger than one (i.e., h>1), which is evidenced by its relationship with ED₅₀:

$$D_{MS}={\left(\frac{h-1}{h+1}\right)}^{1/h}{ED}_{50}.$$ (Equation 3)

Finally, the maximum slope (MS) is obtained by putting D_MS as a value of x into Equation 2. The value of y′ obtained is MS (Equation 4).

$$MS=\frac{(t-b){(h-1)}^{1-1/h}{(h+1)}^{1+1/h}}{4{\mathit{hED}}_{50}}$$ (Equation 4)

Note that though multiple equations were given to show the process, only Equation 4 is needed for using the mathematical method. MS will be obtained as long as t, b, h, and ED₅₀ are obtained via curve fitting of a sigmoidal curve.

With the steepest slopes obtained, a toxicity comparison can now be conducted by comparing these slopes, with a higher value meaning higher toxicity. Note that no matter which criterion is used for ranking toxicity (benchmark dose, ED₅₀, or slope), there are always exceptions. There is no guarantee that ranking based on one criterion is completely consistent with that of another.

2.11. Correlation between in vitro and in vivo data

The correlation coefficient between two endpoints can be calculated using the slope data when data of the two endpoints are available for a sufficient number of NPs. Since no evidence of normal distribution is present, the slope data was analyzed using a non-parametric method (Spearman correlation)

2.12. Statistical analysis

Dose-response modeling and correlation analyses were performed using the software GraphPad Prism 5 (GraphPad Software, La Jolla, CA, USA) while other calculations were completed in Microsoft Excel spreadsheet. When using the mathematical method for obtaining the steepest slopes for protein carbonylation data, raw data were first normalized by dividing each data point by the average of the control group so that a fixed “bottom” value of 1 for fitting. This was not done for LDH data because LDH data have a meaningful unit while protein carbonylation readings do not. A statistical test was considered significant when p < 0.05. Confidence intervals (CIs) were calculated for the steepest slopes in different ways for the graphical and mathematical methods. For ROS data, the slopes were obtained with CIs via linear regression. In the graphical method for the in vivo data, CIs were also obtained via linear regression for the data of the two dose groups that the steepest slope was based on. In the mathematical method for the LDH and protein oxidation data, each of the CIs for the slopes was obtained by manually resampling the dose-response data using Jackknife technique (randomly deleting one data point each time to get a sample for slope calculation).

3. Results

3.1. NP characterization

Hydrodynamic sizes were found to be much larger than the particle sizes in dry state (Table 1). The size measurements in PBS and in RPMI 1640 cell culture medium were similar with the latter slightly larger. As expected, the measured sizes of the NPs were remarkably smaller in RPMI medium with presence of 1% FBS than without due to coating of NPs (about 70% reduction in size).

Table 1.

Physicochemical characterization of the anatase TiO₂ NPs. Primary particle sizes and size distributions of the NPs were measured using transmission electron microscopy (TEM). Hydrodynamic sizes and size fluctuations over time were measured using dynamic light scattering (DLS) technology. The Data are based on size distribution by Intensity (of the scattered light measured). Specific surface area (SSA) was measured using the Brunauer-Emmet-Teller (BET) method. Particle sizes were also calculated based on SSA. These anatase NPs were prepared with titanium tetra-isopropoxide (TTIP, 97%, Aldrich) using a diffusion flame or premixed flame method. In the diffusion method, the oxidizer (O₂) combines with the fuel (methane, CH₄) by diffusion. A premixed flame is a flame in which the oxidizer is mixed with the fuel before it reaches the flame front.

Preparation method (Type of flame)	Calculated Size (nm) based on SSA	TEM Size (nm, Mean±SD)	Hydrodynamic Size in medium (nm, Mean±SD)				SSA (m2/g)
			Saline (pH 5.9)	PBS (pH 7.4)	RPMI 1640	RPMI with 1% FBS
Diffusion	10	11±3	778±11	704±15	1008±17	186±6	157.10
Diffusion	16	17±5	906±14	705±16	643±12	183±6	100.00
Diffusion	30	29±10	702±10	541±9	590±14	191±8	52.56
Diffusion	50	47±13	631±8	537±14	592±9	237±7	31.52
Premixed	3	5±1	879±13	887±21	903±20	211±6	426.1
Premixed	7	7±3	869±15	825±11	822±13	189±10	209.6
Premixed	16	16±6	844±12	704±17	813±15	196±7	95.8
Premixed	30	24±9	841±13	650±10	783±14	188±7	53.10
Premixed	53	49±21	763±11	530±9	586±19	212±9	29.68

3.2. Cell-free ROS data

All anatase TiO₂ NPs showed a dose-dependent linear increase in generating ROS, a typical example is shown for a 10 nm TiO₂ NP prepared using the diffusion flame method (Fig. 2). Therefore, the slopes were obtained via linear regression. The slopes increased with increasing NP sizes based on NP surface area as a dose metric while there was no clear trend when the slopes are based on mass (Fig. 3).

Fig. 2.

Dose-response curve of ROS data for a 10-nm anatase TiO₂ NP in a cell-free system showing the calculation of slope via linear regression. The slope obtained was 0.051 μM H₂O₂ equivalent per μg/ml. The ROS were measured in a cell-free system (sodium phosphate buffer) using the dye 2′,7′-dichlorofluorescein diacetate (DCFH-DA). The fluorescence was measured using a spectrofluorometer with excitation/emission wavelengths of 485 nm/535 nm. The assay was performed in the dark to avoid any interference from photocatalytic activities of the NPs.

Fig. 3.

Slopes of cell-free ROS data versus particle size (nm) for the TiO₂ NPs. Each of these slopes was obtained from a dose-response dataset via linear regression with an example shown in Fig. 2. The slopes have a unit of μM H₂O₂ equivalent per μg/ml based on mass or μM H₂O₂ equivalent per cm²/ml based on surface area. While there is no clear trend for slope expressed in mass (a), surface area based slope increases with increasing size (b). These NPs were either prepared by a diffusion flame or a premixed flame method.

3.3. LDH data

The 24-hr LDH dose-response data for all the anatase TiO₂ NP showed sigmoidal curves with a 30-nm anatase NP as an example (Fig. 4). The mathematical method was used for calculating the steepest slopes and the slopes were plotted against NP size (Fig. 5). As can be seen, the slope increased as a function of NP size when based on surface area but not on mass.

Fig. 4.

LDH dose-response data (mass dose) for an antase TiO₂ NP (30-nm, diffusion flame) fitted using a sigmoidal equation in GraphPad Prism. The 24-hr LDH data (cultured R3/1 cells) were plotted against the logarithm of dose (μg/ml). The steepest slope was obtained using the mathematical method described in Fig. 1.

Fig. 5.

Slopes of the in vitro LDH data from exposure study in R3/1 cells for the anatase TiO₂ NPs (mathematical method). While there is a decreasing trend with increasing NP size for slopes expressed in mass (unit: μM/min LDH activity per μg/ml mass concentration) (a), surface area (unit: μM/min LDH activity per cm²/ml mass concentration) based slopes increase with increasing NP size (b). These NPs were prepared by a diffusion or a premixed flame method.

3.4. Protein carbonylation data

An example for the carbonylation data was presented using a 30-nm anatase NP (Fig. 6). The steepest slopes were obtained by the mathematical method and plotted against NP size either based on mass or surface area (Fig. 7), showing again that the slope increased as a function of NP size when based on surface area but not mass.

Fig. 6.

Dose-response data (mass dose) of protein carbonylation following exposure of R3/1 cells. The protein carbonyl level increases in a dose-dependent manner after 24-hr exposure. Carbonyl values were normalized with respect to the average value of the control group (Average of control was set as 1). This example is about a 30-nm anatase TiO₂ NP (prepared by a diffusion flame method).

Fig. 7.

Slopes of the in vitro protein carbonylation data from the anatase TiO₂ exposure study in R3/1 cells (mathematical method). Each of these slopes was obtained from a dose-response dataset with an example dataset shown in Fig. 6. While there is no clear trend for slope expressed in mass (Unit: fold increase of carbonyl per μg/ml) (a), surface area based slopes (Unit: fold increase of carbonyl per cm²/ml) increase with increasing NP size (nm) (b). These NPs were prepared using diffusion (marked with “D”) or premixed flame method (marked with “P”).

3.5. In vivo study

Rats were exposed to the anatase TiO₂ NPs for 24-hr at three different doses via intratracheal instillation and several endpoints were measured 24 hr later. Compared to other endpoints (data in supplementary information, Figures S-2 to S-13), the number of PMNs in bronchoalveolar lavage fluid (BALF) was the most sensitive indicator of pulmonary inflammation and showed clear dose-response relationships (Fig. 8). The PMN dose-response curves had a general trend of decreasing steepness of slope with increasing NP size when based on mass as a dose metric, while the trend was generally reversed based on surface area. Using the graphical method, the slopes for all these NPs were obtained and plotted against NP size (Fig. 9). These slopes increase as a function of NP size based on surface area (maximum response per cm² surface area of NP) but decrease based on mass (maximum response per μg of NP).

Fig. 8.

PMN number in rat bronchoalveolar lavage fluid 24-hr after intratracheal instillation of the anatase TiO₂ NPs (Dose unit: μg or cm² per rat). There is a clear response for all NPs and many dose-response curves level off at high-dose end, suggesting that PMN number is a sensitive marker for acute lung inflammation and it can serve as the in vivo endpoint for correlating with in vitro data. The NPs were prepared using diffusion (marked with “D”) or premixed flame method (marked with “P”).

3.6. Correlations between in vitro and in vivo data based on the steepest slope

For the anatase NPs, correlations between the in vitro data (cell-free ROS data, cellular LDH data, and cellular protein carbonylation data) and in vivo data (number of PMNs in rat BALF) were correlated based on slopes derived from the corresponding dose-response data. These slopes were either calculated on the basis of mass or surface area as dose metric.

Correlations between slopes of the ROS data (Fig. 3) and those of PMN data (Fig. 9) were plotted in Fig. 10, either based on mass or surface area. Similarly, correlations between slopes from LDH data (Fig. 5) and those from PMN data (Fig. 9) were plotted in Fig. 11. Fig. 12 shows the correlations between slopes from protein carbonylation data (Fig. 7) and slopes from PMN data (Fig. 9).

Fig. 10.

Spearman correlation between slopes from cell-free ROS data and those from PMN data. When the slopes are based on mass, the correlation is weak with no statistical significance (a); the correlation is strong and significant based on surface area (b). These NPs were prepared using diffusion (marked with “D”) or premixed flame method (marked with “P”).

Fig. 11.

Spearman correlation between slopes from LDH data and those from PMN data. When the slopes are based on mass, the correlation is weak with no statistical significance (a); the correlation is strong and significant based on surface area (b). These NPs were prepared using diffusion (marked with “D”) or premixed flame method (marked with “P”).

Fig. 12.

Spearman correlation between slopes from protein carbonylation data and those from PMN data. When the slopes are based on mass, the correlation is weak with no statistical significance (a); the correlation is strong and significant based on surface area (b). These NPs were prepared using diffusion (marked with “D”) or premixed flame method (marked with “P”).

In all three cases, positive correlations with statistical significance were found between the in vitro and in vivo data when using surface area based dose- and response- metrics (Figures 10–12). On the other hand, when examined on a mass basis, none of the correlations had statistical significance.

4. Discussion

The major focus of this work is to assess the usefulness of in vitro (cell-free and cellular) assays by examining the correlations between the in vitro and in vivo data while using either NP mass or NP surface area as a dose metric and the steepest slope of the dose-response curve (maximum response per unit dose) as a response metric. The analysis showed that only the dose metric “surface area” showed a significant correlation between in vitro and in vivo results.

In characterizing the particles, we found that the hydrodynamic sizes were much larger than particle sizes in dry state (Table 1), indicating that these NPs tend to form agglomerates in the media. Interestingly, the measured sizes of the NPs were about 70% smaller in RPMI medium when 1% FBS was present (Table 1), suggesting that some serum components could partially reverse agglomeration probably by adsorbing onto the surface of the NPs. This finding is in line with the data published by others (Bihari et al., 2008; Allouni et al., 2009; Meissner et al., 2009). In contrast to aggregation of NPs, agglomeration of NPs maintains their surface area (ISO-DIS2011), so the fact that agglomeration of NPs occurred in aqueous solutions justifies the use of surface area as dose metric. It would be desirable, though, to have a measure of biological available surface area, which still needs to be defined.

When correlating in vitro results with in vivo data, it is critical to find a consistent criterion for comparison. In other words, it is critical to translate the response seen at a certain dose in vitro to a corresponding equivalent in vivo dose and response.

We addressed this issue by defining the steepest slope of the dose-response curve for toxicity comparison. The main reason for choosing the steepest part of a dose-response curve is that response in this part of the curve is induced by intermediate doses and the response is most sensitive to changes in dose. When doses increase further, the curve becomes less steep as the signal being measured is saturated. This change is consistent with the proposed dose-dependent transitions in mechanisms of toxicity (Slikker et al., 2004). At the lower doses, the response is close to the background response, the response is not very sensitive to a dose change either, yet the response per dose change becomes maximal at the steepest slope.

By obtaining the steepest slope, a whole dose-response dataset is reduced to this single data point. Thus, correlation between two endpoints could be evaluated based on the steepest slopes for a group of NPs.

Using three in vitro and one in vivo assay, in vitro – in vivo correlations could be correlated based on the steepest slopes. Though not all in vitro assays chosen for this study have mechanistic implications, the slope data from the in vitro assay show good correlation among themselves (Supplementary information, Tables S-1 and S-2).

When using mass as a dose metric, the correlations were poor with no statistical significance; strong correlations with statistical significance were found when using surface area as the dose metric (Figures 10–12). This suggests that the in vitro assays have predictive values for in vivo outcomes, though our work by itself is not a prediction study.

The poor in vitro – in vivo correlations based on mass do not necessarily indicate a problem with this dose metric since our method should also be applicable to chemicals for which dose metric is not an issue. The poor correlation could be due to the smaller toxicity differences between the NPs when based on mass rather than surface area.

We have explored two methods for calculating the steepest slope. The graphical method is straightforward but there are several limitations when there are only three candidate slopes to choose from – as is the case in our studies – because of only three doses in our in vivo animal studies. Firstly, if the steepest slope is not in the middle of the selected dose range, it is conceivable that the slope may be steeper when adding another dose that is outside the current dose range. Secondly, statistical power is limited when using the graphical method. By including data of only two dose groups at a time for calculating slopes and excluding all the other seemingly “irrelevant” groups, the graphical method never takes the full dataset as a whole for statistical analysis. Thirdly, an additional constraint of the method related to the in vivo part is the high cost and ethical concerns when conducting animal experiments. Therefore, the mathematical method is preferred where applicable. The criterion to choose between the two methods is to plot the data and examine the shape of the curve to see if it is S-shaped. When in doubt, the mathematical method should be tried first and see if the fitting works.

In the mathematical method, data from all dose groups except the control group and those in an overdosing situation (decreasing response at the high end of the dose-response curve due to decreased viability/increased mortality) are included for statistical analysis, thus leading to increased statistical power. Furthermore, the original data points instead of group means are used for curve fitting (Fig. 4).

While ED₅₀ is the traditional means for ranking toxicity for a group of chemicals, the use of the steepest slope is equally appropriate. This is supported by the good correlations between the steepest slopes, ED₅₀, and ED₁₀ for some data where ED₅₀ and ED₁₀ can be calculated (Supplementary information, Tables S-3 and S-4).

However, we did not use ED₅₀ for two reasons. First, it is difficult to obtain an ED₅₀ for some NPs. For example, if the steepest slope is between the highest dose and the second highest dose, the maximum response cannot be estimated and thus it is impossible to get a reliable ED₅₀. In this case, the mathematical method cannot be used since ED₅₀ is needed to use the method. Using the graphical method is the best way available given such a suboptimal situation. Another reason for not using ED₅₀ is the linearity of the dose-response curve for the cell-free ROS data from which one cannot get the maximum response. Without the maximum response, one cannot get an ED₅₀.

The NPs chosen in our present study are limited to TiO₂ of different sizes ranging between 3 and 104 nm. But our method is likely to work for other NPs as well even though physico-chemical properties of different types of NPs can be quite diverse. However, when our concept of steepest slope analysis was applied to data reported by Sayes et al. (2007) who evaluated in vitro and in vivo toxicity of several particles and found that the in vitro and in vivo results showed poor correlations, our re-analysis using the steepest slope approach showed significant correlations (Rushton et al., 2010).

Furthermore, our preliminary in vitro-in vivo correlation study (Rushton et al., 2010) covered 8 different particle types (Rushton et al., 2010) using the same concept as a first proof of principle. Again, significant in vitro – in vivo correlations were found which allow developing a hazard scale for toxicity ranking of the different NPs. This is an important step in the risk assessment process.

Still, it is desirable to confirm this concept of steepest slope analysis by conducting similar studies using other NPs of various chemistry and surface modifications. In addition, all endpoints involved in this study were acute endpoints (up to 24 hours) and extrapolations to chronic effects are yet to be investigated.

The correlations established between the in vitro and in vivo assays in this work are non-mechanistic in nature. These empirical correlations are based on statistical analyses and a detailed investigation of the biological relationship between these endpoints is beyond the scope of this work. However the endpoints used in our correlation study (cell-free ROS generation, in vitro cellular LDH release, in vitro cellular protein carbonylation, and in vivo PMN response) are directly or indirectly related to oxidative stress, and therefore are very relevant.

The goal of establishing and verifying in vitro – in vivo correlations is an initial step towards predicting in vivo toxicity of NPs using validated in vitro results. No meaningful predictions can be made for an in vivo outcome based on an in vitro endpoint that is known to be poorly correlated with the in vivo response. Conversely, results of validated in vitro assays will be valuable for hazard ranking against well characterized positive and negative reference NPs. Thus, the merit of the steepest slope analysis lies in rapid in vitro toxicity screening of nanoparticles for validating in vitro assays which are often deemed to be of little relevance due to the use of high doses and other weaknesses.