Establishing structure-property-hazard relationships for multi-walled carbon nanotubes: the role of aggregation, surface charge, and oxidative stress on embryonic zebrafish mortality

Abstract

Increasing use of carbon nanotubes (CNTs) in consumer and industrials goods increases their potential release, and subsequent risks to environmental and human health. Therefore, it is becoming ever more important that CNTs are designed to reduce or eliminate hazards and that hazard assessment methodologies are robust. Here, oxygen-functionalized multi-walled CNTs (O-MWCNTs), modified under varying redox conditions, were assessed for toxic potential using the zebrafish (Danio rerio) embryo model. Multiple physicochemical properties (e.g., MWCNT aggregate size, morphology, and rate; surface charge and oxygen concentration; and reactive oxygen species (ROS) generation) were characterized and related to zebrafish embryo mortality through the use of multivariate statistical methods. Of these properties, surface charge and aggregate morphology emerged as the greatest predictors of embryo mortality. Interestingly, ROS generation was not significantly correlated to observed mortality, contrary to prior predictions by nanotoxicology researchers. This suggests that the mechanism of MWCNT-induced mortality of embryonic zebrafish is physical, driven by electrostatic and shape effects, both of which are related to nanomaterial aggregation. This raises the importance of rigorously considering aggregation during aqueous-based nanotoxicology assays as nanomaterial aggregation can affect perceived nanomaterial toxicity. As such, future nanotoxicity studies relying on aqueous media must sufficiently consider nanomaterial aggregation.

Graphical Abstract

1. Introduction

Carbonaceous engineered nanomaterials have emerged as some of the most promising nanomaterials across a wide range of applications due to their electrical properties, excellent strength, and varied applications (e.g., water treatment and medical devices)[1-4]. Of these materials, multi-walled carbon nanotubes (MWCNTs) have become one of the most widely-used due to their relatively low cost compared to their single-walled counterparts; their tunability in terms of length, diameter, and surface group functionality; and the large multitude of synthesis and post-synthesis treatment techniques[5]. However, while MWCNTs have tremendous potential in a variety of applications, their increasing ubiquity, and therefore increasing potential for release and exposure[6], suggests an increasing potential risk to human health and the environment. Their inherent hazard has been the focus of an increasing number of studies on MWCNT toxicity released to the environment and in the workplace[7-10]. For MWCNTs, and nanomaterials more generally, to realize their commercial potential, it is essential to fully understand what structures and properties influence the desired functional performance and undesired negative implications of nanomaterials[11-13]. In doing so, the nanotechnology community can make informed decisions to avoid the costly impacts of regrettable substitution[14], adverse impacts on public perception of emerging nanotechnologies[15], and misalignment with current and future regulatory activity in the US and globally[16].

Ecotoxicity and potential human toxicity of carbon nanotubes are explored using a wide variety of animal and cell models, both in vivo and in vitro[17-20]. Of the assays, embryonic zebrafish (Danio rerio)[21-24], commonly used in chemical toxicity studies due to its low cost, medium-to-high throughput, and analogies to the human genome[25], has been gaining popularity in nanotoxicology[21, 24]. While this assay has been used over the past decade to evaluate the toxicity of carbonaceous nanomaterials, and nanotubes in particular[26-31], a few concerns have emerged. First, there is no consensus on MWCNT toxicity mechanism(s) toward zebrafish and other aquatic species, which could lead to the unintentional (re)design of MWCNTs that may be more hazardous than anticipated. While oxidative stress induced by carbonaceous nanomaterials has been hypothesized as the main mechanism driving toxicity towards aquatic species[32-34], physicochemical properties such as length[35], surface functionality[36, 37], or the surface charge of the MWCNTs[38] have also been related to toxic responses. While these have, in most cases, been individually considered, few studies have looked at these properties simultaneously and comprehensively. A second concern is the appropriateness of aqueous-based toxicological assays, including zebrafish assays and common mammalian culture platforms, for specific nanomaterials. These assays are highly reliant on well-dispersed materials in solution to ensure homogenous exposure. However, it is well-established that nanomaterials, especially in (biological) media containing salts, can readily aggregate[39, 40]. Specifically, it has been shown that the various physicochemical properties of carbon nanotubes can impact their colloidal behavior and dispersibility in aqueous media containing salts. For example, an increase in oxygen moieties, such as hydroxyl groups, on the surface of both single- and multi-walled carbon nanotubes can lead to an increasingly negative surface charge[38]. It has been shown that there is a linear correlation between the total oxygen concentration, surface charge, and critical coagulation coefficient (CCC) of MWCNTs[41]. That is, as oxygen concentration increases, total surface charge decreases, and the CCC increases, due to stronger electrostatic repulsion between particles. Therefore, in biological media containing salts, there can be variable levels of carbon nanotube aggregation, based on the CNT properties. Further, aggregation, and subsequently point of zero charge (PZC), have been linked to embryonic mortality[38]. Yet, most aquatic toxicity studies of nanomaterials do not robustly consider aggregation, and an inaccurate stability assumption can impact conclusions regarding nanomaterial safety[42]. Those that do consider aggregation compensate for it through the addition of surfactants[43, 44] or via high levels of surface functionalization[38, 45]. The addition of surfactants may directly impact the embryos, which in turn may affect the biological response to nanomaterials and confound the experimental results[46, 47]. High concentrations of surface functionalization are generally implemented to disperse MWCNTs in aqueous solutions and ensure maximum contact with the aquatic species; however, the assumption of MWCNT stability is not explicitly validated. As such, these studies may not be representative of the actual inherent hazard of the materials. If the MWCNTs are not well-dispersed, this can significantly impact the subsequent assumption of homogeneous exposure. A third concern involves the role of the chorion, a protective outer membrane surrounding the zebrafish embryo. While some embryonic zebrafish nanotoxicity studies are performed with the chorion intact[48-50], others (especially those studying carbonaceous nanomaterials) remove the chorion to maximize the sensitivity of the assay; that is, the worst-case scenario[38, 43]. In many cases, the chorion is assumed to be acting as a physical barrier to contact, but in the case of MWCNTs, many studies have used materials with a negative charge, which may be electrostatically repulsed by the negatively charged chorion surface[51]. Additionally, with a chorion pore size of 0.5-0.7 μm, many studied nanomaterials and nanomaterial aggregates are not limited in their movement through the chorion pores by size constraints[52]. This may also explain why positively-charged nanomaterials and ions produce toxic effects when the chorion is intact[50] [53].

To probe the significance of each of these three concerns in evaluating the toxicity of MWCNTs using a zebrafish embryo assay, a representative aqueous-based toxicity assay, a total of 20 samples of oxygen-functionalized MWCNTs (O-MWCNTs) were produced to: 1) validate previous findings relating certain MWCNT physicochemical properties to toxicity; 2) elucidate the mechanism of MWCNT-induced nanotoxicity toward zebrafish embryos; 3) determine the role of aggregation (and PZC) on zebrafish embryo assays; and 4) probe the relationship between dechorionation, embryo mortality, and MWCNT exposure. This study aims to inform and encourage safe and sustainable MWCNT design that considers potential toxic effects early in the design process. By understanding the cause of MWCNT-induced toxicity, designers can develop materials with reduced or eliminated adverse eco- and/or human health impacts. At the same time, by understanding the role of aggregation and the chorion in embryonic zebrafish studies, toxicologists can also pursue the appropriate experimental protocols to most effectively evaluate the toxicity of and to inform the design of safer nanomaterials.

2. Experimental

2.1. MWCNT Preparation

Pristine MWCNT with outer diameters of 10-20 nm, inner diameter of 3-5 nm and a purity of >95% were purchased from CheapTubes (Cambridgeport, VT, USA). The as-received material was refluxed in nitric acid (HNO₃, 70%) for 1, 2, 4, 6, or 8 hours, which leads to the production of 5 samples that have increasing surface oxygen concentrations and decreasing average lengths, as the reflux time increases[54]. The MWCNTs were then repeatedly rinsed with DI water to remove remaining HNO₃ and dried at 100°C for 24-48 hours. These samples either underwent high-temperature annealing in an inert He atmosphere at maximum temperatures of 400°C, 600°C, or 900°C for one hour or remained unannealed (UA). This annealing occurred in a Thermo Scientific Lindberg/Blue M Tube Furnace under a constant flow of He at a rate or 400-500 sccm. The temperature of the oven was ramped up at a rate of 10°C/min until it reached its final set-point, at which it remained for 1 hour. Samples remained in the thermal annealer under inert He gas until the temperature of the oven was below 80°C, at which point the sample was removed and stored. The annealing step of this process has the effect of reducing surface oxygen groups added during the acid reflux step, where, generally, as annealing temperature increases, total surface oxygen concentration decreases[55]. Specifically, at lower annealing temperatures, carboxyl surface moieties have been shown to decompose, while at higher temperatures, hydroxyl and carbonyl-type groups begin to decompose[56]. As a result, these two steps yielded a total of 20 oxygen-functionalized MWCNTs (O-MWCNTs), featuring 5 groups of samples with variable lengths, and 4 samples of varying surface oxygen concentration within each of those groups.

2.2. MWCNT Characterization

The structure and properties of the O-MWCNTs prepared for this study were characterized for elemental composition by x-ray photoelectron spectroscopy (XPS), dispersed aggregate morphology by static light scattering (SLS), O-MWCNT surface charge as related to PZC, dispersed aggregate size by dynamic light scattering (DLS),and average particle length by scanning electron microscopy (SEM). Elemental composition via XPS was measured using a Physical Electronics PHI VersaProbe II Scanning XPS Microprobe (Chanhassen, MN, USA) and is given as the relative atomic concentration of carbon and oxygen for each O-MWCNT sample. The relative O-MWCNT surface charge was evaluated by a mass titration technique that provides the PZC of each O-MWCNT sample, as described by Lee et al.[57]. Briefly, O-MWCNTs were added to DI water at roughly 10 mg increments and stirred. After 10-15 minutes of stirring, a pH measurement was taken. These two steps were repeated until there was a plateau in the pH vs. mass plot, yielding the final PZC value. Generally, as the PZC of a sample deviates from the solution pH, there is a larger absolute surface charge, where a particle with PZC below the solution pH will have a negative surface charge in solution, and a material with a PZC above the pH of the solution will have a positive surface charge.

For SLS and DLS, a 25 mg/L solution of each O-MWCNT sample in the stock solution (0.5% DMSO, 62.5 μM CaCl₂, pH 7.1 in DI water) was prepared and sonicated for 30 minutes (22 Hz, 26 W) and left to sit for 10 minutes before each run, to best replicate the time needed in between sonication and O-MWCNT to zebrafish embryo exposure. This solution was chosen to be representative of the concentrations of the embryonic zebrafish media while reducing multiple scattering and adsorption of the incident laser light. The aggregate morphology, aggregate size, and aggregate growth rate were measured using a multi-detector light scattering unit (ALV-GmbH, Langen, Germany) with a Nd:vanadate laser (Verdi V2, Coherent, Inc., Santa Clara, CA, USA) operating at a wavelength of 532 nm. Aggregate morphology is given as the fractal dimension (D_f), where D_f=1 represents a more rod-like and less compact aggregate, and D_f=3 represents a more compact, spherical aggregate[58, 59]. Scattering intensity was collected at angles of 30° to 150° at 2° increments to calculate D_f for each sample. Aggregate size and size distribution are estimated as the hydrodynamic radius (in nm) of the aggregates in solution based on their Brownian motion. The scattered light from the dispersed samples was collected at a fixed angle (90°) for 20 measurements (30 s each). Any samples that showed noticeable growth in size over the first 10 minutes is noted in Table 1 with an asterisk. Overall, this technique has been shown to be good for estimating the relative hydrodynamic radius of fairly stable carbon nanotubes and aggregates in solutions, regardless of their non-spherical shape[60], even if it not appropriate as a direct measure of MWCNT length.

Table 1:

O-MWCNT sample name, preparation steps, relative surface oxygen concentration (%), the average hydrodynamic radius in solution (nm), the fractal dimension (D_f), the point of zero charge (PZC), and the average particle length as measured with SEM (nm).

O- MWCNT Sample	Acid Treatment	Maximum Annealing Temperature	Relative Surface Oxygen (%)	Aggregate Radius (nm)	Fractal Dimension, Df	Point of Zero Charge, PZC	Average MWCNT Length (nm)
1-UA	1 hr reflux in 70% nitric acid	Unannealed	2.6	65.7	1.399	4.51	683
2-UA	2 hr reflux in 70% nitric acid	Unannealed	5.7	51.5	1.000	3.62	531.3
4-UA	4 hr reflux in 70% nitric acid	Unannealed	7.5	59.8	1.354	3.27	430.7
6-UA	6 hr reflux in 70% nitric acid	Unannealed	8.5	48.5	1.168	3.24	429.1
8-UA	8 hr reflux in 70% nitric acid	Unannealed	8.9	47.7	1.370	3.14	374
1-400	1 hr reflux in 70% nitric acid	400°C	1.7	289.0*	1.284*	6.74	683
2-400	2 hr reflux in 70% nitric acid	400°C	3.1	53.2	1.466	4.19	531.3
4-400	4 hr reflux in 70% nitric acid	400°C	4.1	63.0	1.391	3.75	430.7
6-400	6 hr reflux in 70% nitric acid	400°C	5.0	58.7	1.666	3.60	429.1
8-400	8 hr reflux in 70% nitric acid	400°C	5.3	57.8	1.641	3.47	374
1-600	1 hr reflux in 70% nitric acid	600°C	1.3	201.6*	1.357*	7.60	683
2-600	2 hr reflux in 70% nitric acid	600°C	2.3	66.4	1.569	6.18	531.3
4-600	4 hr reflux in 70% nitric acid	600°C	2.8	75.4	1.625	5.93	430.7
6-600	6 hr reflux in 70% nitric acid	600°C	3.9	64.7	1.793	4.50	429.1
8-600	8 hr reflux in 70% nitric acid	600°C	3.9	67.3	1.732	4.24	374
1-900	1 hr reflux in 70% nitric acid	900°C	0.5	333.4*	1.452*	8.45	683
2-900	2 hr reflux in 70% nitric acid	900°C	1.5	216.7*	1.525*	7.11	531.3
4-900	4 hr reflux in 70% nitric acid	900°C	2.0	170.9*	1.666*	6.75	430.7
6-900	6 hr reflux in 70% nitric acid	900°C	1.8	67.8	1.800	6.85	429.1
8-900	8 hr reflux in 70% nitric acid	900°C	2.2	84.3	1.936	5.53	374

^*These values are not representative of the hydrodynamic radius or fractal dimension of a single primary particle in solution due to rapid aggregation early in the measurement process.

For SEM measurements evaluating length as a result of oxidative cutting, unannealed (1-UA, 2-UA, 4-UA, 6-UA, 8-UA) O-MWCNT samples were dispersed in ethanol at concentrations below 1 mg/L. Annealed samples were not analyzed, as previous work by Yamamoto et al.[61] showed no significant change in length as a result of thermal annealing treatments. The ethanol-MWCNT solution was dropped onto a silicon wafer and left to dry under vacuum. Images of 100-200 individual MWCNTs were captured for each sample. ImageJ (NIH) was then used to measure each individual particle and determine average length.

2.3. Aggregation Testing

O-MWCNT aggregation was evaluated using time resolved-dynamic light scattering (TR-DLS) over a 3-hour time period. O-MWCNT solutions for TR-DLS measurements were made in the same concentrations and prepared in the same way as those used for SLS and DLS readings. TR-DLS measurements were taken at 60 s time intervals for a total of 180 measurements, leading to the collection of 3 hours of size information. This number of measurements and amount of time to evaluate aggregate growth is well-established in the literature [40, 62, 63], and is enough to establish an estimated linear aggregate growth rate, and ensure noise or slight variations were not misconstrued as a real size change. The size change of aggregating O-MWCNTs was derived from experimentally measured particle size data, where the average hydrodynamic radius over each time interval was recorded. The estimated primary particle or agglomerate size was plotted over time to evaluate aggregation kinetics of the O-MWCNTs in stock solution during the first 3 hours of exposure to zebrafish embryos.

2.4. Toxicity Testing

Toxicity testing was performed on tropical 5D wild-type zebrafish embryos with and without enzymatically removed chorions, following established procedures[64]. To control for non-contact effects, a smaller subset of chorionated embryos was screened alongside the dechorionated embryos. Before exposure to the O-MWCNTs in the stock solution, dechorionated embryos were acclimated from fish water to CaCl₂ through partial media changes to limit background mortality caused by physiological stress related to media change. Embryos with their chorion intact did not go through the acclimation process because the chorion makes them less susceptible to the stress of media change. Toxicity testing on the chorionated and dechorionated embryonic zebrafish was performed under direct exposure of chorionated and dechorionated zebrafish embryos to O-MWCNTs.

Dechorionated (N = 16) and chorionated (N = 8) embryos were placed in 96-well plates. O-MWCNTs were probe sonicated for 30 minutes prior to exposures to embryos. Embryos were dechorionated at 4 hours post fertilization (or left intact), then exposed to O-MWCNTs between 9-10.5 hours post fertilization (hpf). At time of exposure, 50 μL of O-MWCNTs was added to each well, such that embryos were exposed to three final concentrations of O-MWCNTs: 12.5, 25, or 50 mg/L, all solubilized in 0.5% DMSO and a CaCl₂ concentration of 62.5 μM. A stock solution containing DMSO and CaCl₂ was used for the toxicity studies to limit MWCNT aggregation as much as possible, while also limiting media-related mortality to zebrafish embryos. Control embryos were exposed to 0.5% DMSO and 62.5 μM concentration of CaCl₂. Embryos were statically exposed until 120 hpf in the dark. At 24 hpf, embryos were assessed for the presence or absence of 4 developmental toxicity endpoints for each zebrafish embryo (MO24: mortality at 24 hpf; DP: developmental progression; SM: spontaneous movement; and NC: notochord distortion)[65]. At 120 hpf, 18 developmental endpoints were assessed, yielding a total of 22 assessed endpoints (Table S1). The zebrafish acquisition and analysis program (ZAAP), a custom program designed to inventory, acquire, and manage zebrafish data, was used to collect developmental endpoints as either present or absent (i.e., binary responses were recorded).

All statistical analysis of the morphology endpoints was performed using code developed in R[66]. The data used were binary incidences recorded for each endpoint from ZAAP (as described above) to compute the lowest effect level (LEL) for each endpoint. To determine statistical significance for each O-MWCNT:endpoint, the respective control incidence rate was compared to the incidence rate observed at each concentration. Because the endpoints are binary and replicates are measured in separate wells, the 0/1 response for each material endpoint-concentration-replicate combination translates to a series (N = 16 dechorionated, N = 8 chorionated) of Bernoulli trials, or “coin-flips.” Therefore, the LEL significance threshold was estimated using a binomial test, which provided a straightforward method to adjust for plate and/or chemical effects and the pooling/separation of controls[67]. The recorded LEL was the lowest concentration at which the observed incidence exceeded the significance threshold (p ≤ 0.05).

2.5. Reactive Oxygen Species and Oxidative Stress Assays

To understand if oxidative stress plays a role in O-MWCNT mediated toxicity, O-MWCNTs were evaluated based on their ability to oxidize glutathione (GSH), a biologically-relevant probe for the production of reactive oxygen species (ROS). Since GSH serves as a cellular antioxidant, this assay is best positioned to act as an indicator of a material’s potential to induce oxidative stress[68]. This procedure has been previously described[55], and is described in further detail in the Supporting Information. Briefly, 100 mg/L O-MWCNTs were placed in a solution containing 333 μM GSH, and the concentration of GSH is monitored at 1, 3, and 6 hours using Ellman’s reagent and a spectrophotometric technique. The percent conversion of GSH to glutathione disulfide (GSSG) was calculated compared to a control that contained no O-MWCNTs. Averages and standard deviations were calculated from triplicate experiments. The calculated percent loss of GSH was considered to be correlated to oxidative stress potential, and therefore, a higher percent loss indicated an O-MWCNT sample that was more likely to induce oxidative stress[69].While this assay is relevant to understanding oxidative stress, the conversion of GSH to GSSG is usually caused by reactions with peroxide groups or as a result of electron transfer to oxygen catalyzed by the material in the suspension[70]. As such, ROS that can cause cellular damage, such as superoxides (O₂^•−), hydroxyl radicals (•OH), and singlet oxygen (¹O₂), are not well-quantified by this assay, and it is unclear whether the conversion of GSH to GSSG is mediated by the production of hydrogen peroxide (H₂O₂)[71]. To quantify the relative production of O₂^•−, •OH, and ¹O₂, probe molecules that are selective to each ROS were placed in O-MWCNT suspensions for a period of 3 hours. This procedure has been described previously[70]. Briefly, the generation of O₂^•− was indicated by the reduction of 2,3-Bis-(2-methoxy-4-nitro-5-sulfophenyl)-2H-tetrazolium-5-carboxanilide (XTT), the generation of •OH was indicated by the conversion of terephthalic acid (TPA) to hydroxyterephthalate (hTPA), the generation of ¹O₂ was indicated by the degradation of the furfuryl alcohol (FFA) probe, and the generation of H₂O₂ was indicated using the Amplex Red assay. The H₂O₂ assay was performed in the presence of 100 mg/L of each O-MWCNT sample, to directly compare the results of the Amplex Red assay with that of the GSH assay. Further details on these methods can be found in the Supporting Information.

2.6. Statistical Analysis

The logistic relationship between the properties and toxic outcomes was explored, to further understand the relationship between O-MWCNT physicochemical properties and O-MWCNT-induced embryonic zebrafish toxicity. The logistic binomial model was chosen due to its commonality among studies determining binomial toxicity outcomes[72], as well as its ease of interpretation. Logistic regression is used to predict the probability of a binomial response (dead vs. alive) as a result of continuous predictor values. More justification about the choice of a logistic binomial regression model for this system can be found in the work by Gilbertson et al.[38] All statistical analyses and model development was performed using code developed in R[66].

Five total O-MWCNT properties (hydrodynamic radius, surface oxygen concentration, PZC, aggregate morphology, and the potential for oxidative stress/ROS production) were included as potential predictors of embryo toxicity. According to the work of Peduzzi et al., the use of a logitistic regression model is justified, considering the number of exposed embryos (160 chorionated embryos and 320 dechorionated embryos per exposure concentration) and the number of properties evaluated (5 properties)[73]. Two models were developed: one considered the impacts of all O-MWCNT samples on toxicity, while the other only considered the impacts of well-dispersed O-MWCNT samples. The embryonic zebrafish toxicity related to each studied O-MWCNT property can be modeled according to Equation (1), as previously described[38]:

$$p_i=\frac{e^{({\beta }_0+{\beta }_1{x}_1+{\beta }_2{x}_2+\cdots +{\beta }_k{x}_k)}}{1+e^{({\beta }_0+{\beta }_1{x}_1+{\beta }_2{x}_2+\cdots +{\beta }_k{x}_k)}}$$ (1)

where p_i is the probability of MWCNT-induced embryo mortality, β₀ is an intercept coefficient, and β₁ is the parameter coefficient for the independent variable, x₁ (i.e. the O-MWCNT property of interest). To evaluate multivariate models, the parameter coefficient can be calculated out for the k^th independent variable and a multiple regression analysis can be performed on Equation (1) to determine the properties most highly correlated with toxicity.

To evaluate goodness-of-fit and which predictors most likely contribute to embryo toxicity, an iterative model selection process was used, with the goal of finding the most robust model without over-fitting the data. P-values were used to assess each property’s fit in the model, where p>0.05 indicated a poor predictor that was removed during the iterative process. Further, Akaike information criterion (AIC) and McFadden’s R² values $({\mathrm{R}}_{\mathrm{McF}}^2)$ were used to evaluate and select the best model fits, where the lowest AIC among all models is considered the best model, and a ${\mathrm{R}}_{\mathrm{McF}}^2$ between 0.2 and 0.4 indicates a very strong fit[74, 75]. A standardized coefficient (or beta coefficient) was also calculated in R for each statistically significant predictor in the final model, where the higher the absolute value of the standardized coefficient, the more influential the parameter is on the probability of embryo toxicity[76].

2.7. Physical Penetration of Phospholipid Membranes

To understand the potential physical interaction between potentially toxic nanomaterials and cellular membranes without the complexity of an entire cell, a simplified spherical lipid bilayer model has been employed[77, 78], where the release of fluorescent molecule encapsulated in a lipid vesicle was quantified. Since it has been shown that reactive oxygen species have a negligible impact on lipid vesicle membrane integrity[77], this test was primarily used to assess physical interactions between O-MWCNTs and cells/cellular models. The procedure has been previously outlined[77]. Briefly, monosaturated synthetic 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC) lipids (Avanti Polar Lipids) were dried to form a thin film. This film was rehydrated in a solution of 50 mM 5(6)-carboxyfluorescein (CF) and 50 mM NaMOPS (pH 7.5) and extruded through a 0.1 μm polycarbonate membrane (GE Whatman) a total of 21 times. After vesicle extrusion, the resulting solution was passed through a HiTrap desalting column along with biological buffer containing 50 mM MOPS and 90 mM NaCl at a ph of 7.5 to separate the lipid vesicles containing CF from the CF that was not encapsulated during the extrusion process.

To quantify the effect of the MWCNT-vesicle interactions, a dye leakage assay was performed by adding solutions of dye-encapsulated vesicles and select MWCNTs at concentrations of 100 mg/L and 5 mg/L, respectively. After exposure at 10, 20, 30, 60, 120, and 180 minutes, MWCNTs and intact vesicles were filtered out and the fluorescence of the resulting solution was monitored at an excitation wavelength of 485 nm and an emission wavelength of 528 nm. There was no statistically significant interaction between the O-MWCNT and the CF solution that may interfere with spectrofluorometric readings. To determine the maximum possible fluorescence, 0.5 wt% of Triton X-100 was added, solubilizing the vesicles and releasing all CF entrapped within. Based on this information, the normalized fraction of leaked CF could be calculated using Equation (1):

$$\frac{C}{C_{\mathit{max}}}=\frac{I-I_0}{I_{\mathit{max}}-I_0}$$ (1)

where C is the concentration of leaked CF, C_max is the maximum possible released CF dye concentration, I is the measured fluorescence intensity determined by a spectrophotometer, I₀ is the measured intensity before vesicle exposure to MWCNT, and I_max is the maximum intensity of CF dye released from vesicles, as determined by exposure to Triton X-100.

3. Results and Discussion

3.1. O-MWCNT physical and chemical properties

Physical and chemical MWCNT property changes as a result of nitric acid treatment and high-temperature annealing, including differences in O-MWCNT hydrodynamic radius, fractal dimension, surface oxygen concentration, and PZC were evaluated (Table 1). There are some readily identifiable trends within the property data. First, as the acid reflux time increases, the surface oxygen concentration of the O-MWCNTs increases while the average nanotube length decreases, as expected[79, 80]. Additionally, annealing at higher temperatures results in a decrease of the relative surface oxygen concentration, a well-established trend[55]. Of note is that the PZC is well-correlated with surface oxygen percentage. As the relative surface oxygen concentration increases, the PZC tends to decrease, likely due to the predominantly negative charge of the carboxyl groups that can be found on the surface post-oxidation[81]. These trends have a few notable exceptions. For example, sample 4-600 (4-hour acid treatment; annealed at 600°C) has a higher relative surface oxygen content than sample 1-UA (1-hour acid treatment; unannealed) but also has a higher PZC value. This is likely due to differences in the type of oxygen functional groups at the O-MWCNT surface[82]. As annealing temperature increases, negative carboxyl and hydroxyl surface functional groups decreases, leaving a relatively larger ratio of basic carbonyl-type surface functionalities, increasing net surface charge[82, 83]. A scatter matrix exploring these trends can be found in the Supporting Information (Figure S1) with an important caveat that these trends may be strongly impacted by O-MWCNTs aggregation. For example, aggregation can make average hydrodynamic radius appear larger and can change the fractal dimension of aggregates in solution.

3.2. O-MWCNT Aggregation

Analysis of primary aggregate growth through TR-DLS over 3 hours was used to determine if a sample had a tendency to aggregate in the zebrafish exposure media (62.5 μM CaCl₂, 0.5% DMSO, pH 7.1). The observed changes in aggregate sizes over this time period (as nm/minute) resulted in two distinct sample groups: well-dispersed samples (defined as “Group 1”, Figure 1a) and aggregating samples (defined as “Group 2”, Figure 1b).

Figure 1:

Aggregation profiles and average linear aggregate growth in nm/minute collected via time-resolved dynamic light scattering showing (a) the relative stability of O-MWCNTs classified as “Group 1” particles and (b) the increase in O-MWCNT aggregate hydrodynamic radius for “Group 2” particles, over a time span of 3 hours. Aggregation experiments were performed with O-MWCNT concentrations of 25 mg/L.

Out of the 20 samples that were prepared, the five samples in Group 2 (1-400, 1-600, 1-900, 2-900, 4-900) had a significantly higher aggregate radius at t=0 than those in Group 1, as determined by DLS. This is likely due to the very high level of aggregation during the ten minutes in between the sonication step and the beginning of the TR-DLS measurement due to the relatively low surface charge of Group 2 particles in the zebrafish embryo media. This growth is also clearly apparent over the next 3 hours after t=0. The Group 2 samples had a linear growth rate of at least 1 nm/minute, while Group 1 samples, with growth rates of 0.00 ± 0.02 nm/min, showed essentially no growth over the same 3-hour period. Due to the rapid aggregate growth, and the nature of DLS and SLS measurements, samples deemed to be part of Group 2 have less reliable hydrodynamic radius and fractal dimension values, as has been indicated in Table 1.

3.3. O-MWCNT Embryonic Zebrafish Toxicity

Dechorionated and chorionated zebrafish embryos were exposed to all 20 O-MWCNT samples at concentrations of 12.5, 25, and 50 mg/L, and lethal and non-lethal effects were noted at 24 and 120 hpf. The total number of embryos affected (lethally or non-lethally) at 120 hpf did not considerably change from the number affected at 24 hpf under the same exposure conditions. This indicates that the observed effects were acute. Further, for 10 of 20 samples at concentrations above 50 mg/L, there was statistically significant mortality at 24 hpf. For these reasons, total embryo mortality at an O-MWCNT concentration of 50 mg/L at 24 hpf was chosen to be the most relevant endpoint in determining which O-MWCNT physicochemical properties are statistically related to the observed toxicity outcomes (Figure 2). While this is a much higher nanomaterial concentration than zebrafish embryos would experience in natural waters, it still has value while relating toxic effects with the properties of emerging MWCNTs.

Figure 2:

Concentration-mortality curves for all 20 O-MWCNT samples for chorionated (N=8) and dechorionated (N=16) embryonic zebrafish. All samples ranged in concentration from 0-50 mg/L. Red points note the concentration of a samples that yield a statistically significant binomial response, and points with a semi-full circle indicates the dechorionated and chorionated embryos had the same response to an O-MWCNT at the given concentration. Sample data is organized from left to right in terms of increasing acid treatment time, and from top to bottom in terms of increasing annealing temperature.

Embryos with an intact chorion do not show significant mortality, even at high concentrations of 50 mg/L, for any of the 20 types of O-MWCNTs, while half of the O-MWCNT samples induced significant levels of toxicity for dechorionated embryos at the same concentration. Several hypotheses for this effect have been postulated. The chorion has been hypothesized to act as a protective barrier, which can prevent physical contact between the O-MWCNTs and the embryo[29]. Further, the chorion has been shown to protect the developing embryo, if the toxicity mechanism for a nanomaterial is predicted to be a result of oxidative stress[84]. Also, the chorion is negatively charged[51], and may electrostatically repulse the oxidized MWCNTs that have a negative surface charge. Therefore, if the mechanism of toxicity is (physically or electrostatically) contact-dependent, based on oxidative stress, or a combination of the two, the noticeable difference in mortality between embryos with and without their chorions is expected, but the contribution of each mechanism to the observed toxicity requires additional exploration. The total number of counted malformations for each O-MWCNT at all concentrations can be found in the Supporting Information (Figure S2). Finally, while the chorion can be penetrated by nickel ions and other metal ions that may act as impurities in MWCNT samples, the lack of any significant toxic outcomes for chorionated embryos suggests that the presence of potential metal impurities in the MWCNT lattice has little to no impact on MWCNT-induced toxicity.

3.4. O-MWCNT Mediated Loss of Glutathione

To determine if O-MWCNT induced toxicity is driven by oxidative stress (O-MWCNT can enable the transfer of electrons required in oxidation[85]), a commonly used biologically relevant assay to detect the oxidation of GSH to GSSG was performed[86].

Within each group of samples that underwent the same amount of time in acid reflux, the conversion of GSH to GSSG 3 hours after O-MWCNT exposure was quantified (Figure 3a; GSH conversion at 1 and 6 hours is shown in the Supporting Information).

Figure 3:

(a) Percent conversion of glutathione (GSH) to glutathione disulfide (GSSG), normalized to a control with no O-MWCNTs. This was used to quantify potential O-MWCNT-induced oxidative stress in embryos. (b) Percent conversion of furfuryl alcohol (FFA), indicating the relative amount of singlet oxygen generated. (c) Generation of superoxide as indicated by the reduction of 2,3-Bis-(2-methoxy-4-nitro-5-sulfophenyl)-2H-tetrazolium-5-carboxanilide (XTT). (d) concentration of hydroxyterephthalate (hTPA) induced by the conversion of terephthaltic acid in the presence of hydroxyl radicals. (e) Hydrogen peroxide concentration in solution as determined by the Amplex Red assay.

From this, a relationship between annealing temperature and percent conversion of GSH to GSSG can be evaluated, where an increase in annealing temperature relates to an increase in percent GHS converted. This result agrees with recent work by Wang and Gilbertson[87], who found that the maximum conversion of GSH occurred for graphene oxide samples that were thermally annealed at the highest temperature in their study. The high-temperature annealing of the graphene oxide had the effect of reducing oxygen functional groups, changing the surface chemistry, which is likely the property controlling GSH conversion in that study as well as this one.

3.5. O-MWCNT Induced Reactive Oxygen Species Generation

To further the investigation into ROS generation as a result of the presence of O-MWCNTs in an aqueous solution, probe molecules were used to quantify the production of superoxides (O₂^•−), hydroxyl radicals (•OH), singlet oxygen (¹O₂), and hydrogen peroxide (H₂O₂) over a three-hour period (Figure 3b-e). Upon reaction of the O-MWCNT suspensions with XTT, there appears to be no discernable peak at 470 nm, which would indicate no superoxide-induced transformation of XTT into XTT-formazan (Figure 3b). Therefore, in the dark, none of the O-MWCNT samples produce a detectable level of O₂^•− in solution over a three-hour period of time. Further, reaction of the O-MWCNT samples with TPA led to non-detectable levels of hTPA. That is, there was no distinguishable peak at the established elution time for any samples to indicate the presence of hTPA in solution, indicating that there is also likely no significant production of hydroxyl radicals at the surface of any of the O-MWCNT samples (Figure 3c). When O-MWCNT samples were reacted with FFA in solution, the percent conversion of FFA relative to the control did not exceed 3% conversion, indicating that similar to O₂^• and hydroxyl radicals, ¹O₂ is not produced in a significant quantity (Figure 3d). This is further confirmed by ANOVA, which demonstrates that there is no statistically significant difference between the average conversion of FFA between samples or compared to the control solution, indicating that samples are likely not producing enough ¹O₂ to convert any FFA (Supporting Information). Based on the Amplex Red assay, the generation of H₂O₂ in the presence of O-MWCNTs over a three-hour period ranged from no generation to a maximum of ~1.38 μM (Figure 3e). This minor generation of H₂O₂ is too small of a concentration to oxidize the 333 μM glutathione at the amounts seen during the GSH assay. Further, there is no relationship between the H₂O₂ production and the percent conversion of GSH to GSSG at 3 hours (Supporting Information). These results support the hypothesis that even though O-MWCNTs in solution can produce small amounts of hydrogen peroxide, the O-MWCNT mediated conversion of GSH to GSSG is likely due to MWCNT-catalyzed electron transfer. There is also likely no ROS generation as a result of the metal impurities in the MWCNT samples. Inductively coupled plasma mass spectrometry measurements on select digested O-MWCNT samples showed a ratio of more than 5000:1 carbon to nickel by mass. Not only is that amount of Ni in solution unlikely to produce high levels of ROS, all experiments are performed in the dark, and Ni nanoparticles do not produce significant levels of ROS or dissolve at a high rate in the dark[88].

3.6. Relating O-MWCNT Properties to Toxicity

To further elucidate the mechanism or properties related to O-MWCNTs toxicity to embryonic zebrafish, the relationships between five physicochemical properties and the dechorionated zebrafish embryo mortality at 24 hpf and at an O-MWCNT exposure of 50 mg/L were explored in detail (Figure 4). As length was byfar the least powerful predictor based on very low McFadden R² values (0.0244 for all samples and 0.00475 for just Group 1 samples), its relationship to embryonic mortality was not considered further. Immediately observable is the role of sample aggregation; Group 2 samples (those that tend to aggregate within the first 3 hours) have relatively low levels of toxicity compared to most Group 1 samples, which could result from a few different assay anomalies. Due to the requirements of the assay, including the salt (CaCl₂) in the media, Group 2 samples begin to aggregate at a more rapid pace[40], especially when they have a PZC close to the pH of the media (pH 7.1), leading to material aggregates either floating to the top or settling to the bottom of the embryo’s well plate.

Figure 4:

Logistic relationships between physical or chemical properties of different O-MWCNT samples and their induced toxicity toward embryonic zebrafish at a concentration of 50 mg/L at 24 hpf. The evaluated properties were (a) O-MWCNT stability in embryo media, (b) O-MWCNT aggregation hydrodynamic radius in embryo media, (c) surface oxygen concentration, (d) O-MWCNT point of zero charge, (e) O-MWCNT aggregate morphology in embryo media, and (f) % loss of GSH, representing the chance of induced oxidative stress.

This aggregate floating and sinking can be seen in Figures 5a and b, which shows solutions of Group 1 and 2 O-MWCNTs after a 24-hour period in embryo media, while Figure 5c and d shows the aggregation of these materials while in embryo media with embryonic zebrafish within 30 minutes of exposure. This variation in aggregate state can influence the assay in two ways: 1) concentration-dependent response curves (similar to Figure 2, above) are no longer representative, as the effective exposure concentration is no longer equal to the original exposure concentration, and/or 2) toxic effects that rely on close proximity between the nanomaterial and embryo (contact-dependence, oxidative stress, etc.) cannot occur because the materials and the embryos are physically separated in space. Even if aggregates stay in solution and in close proximity to the embryos, available contact area decreases as aggregate size increases, so any contact-dependent mechanisms of toxicity are greatly diminished. While aggregated Group 2 samples tended to have low levels of toxicity, the more stable Group 1 samples showed variable and distinguishable levels of toxicity.

Figure 5:

Aggregation of representative samples from Group 1 and Group 2, where (a) shows the stability of Group 1 samples in solution over a 24-hour period, (b) shows the aggregation and settling of a representative Group 2 O-MWCNT sample in solution over a 24-hour period, and (c) and (d) show Group 1 and Group 2 samples within 30 minutes of exposure to 8 hpf dechorionated embryos in exposure solution.

To further explore this variability and understand the physical and chemical properties of O-MWCNT and their observed zebrafish embryo toxicity, the logistic relationships were explored between the properties and toxic effects: 1) Groups 1 and 2 (all 20 O-MWCNT samples) and 2) Group 1, only. In many cases, the logistic relationship and associated coefficient of determination are very similar when looking at all samples vs. those in Group 1 only (Figure 4). However, when Group 2 samples are removed, there is a change in the coefficients and ${\mathrm{R}}_{\mathrm{McF}}^2$ values for Equation (1) for physical properties such as hydrodynamic radius and morphology (factors that are indicators of aggregate state). This suggests that studies that include an analysis of these properties, without considering the impact of aggregation on these values, may not be accurately representing their influence on the observed behavior. It should be noted that as PZC increases, there is a general decrease in mortality, which is a relationship that has been previously reported[38]. Since the media used in this assay had a pH around 7.1, samples with a PZC far below 7.1 tend to have a negative charge, while those with a PZC approaching 7.1 also approach a more neutral charge and were subsequently also more likely to be part of Group 2 (aggregating). However, even when only Group 1 (well-dispersed) samples are considered, there is still a trend showing a decrease in toxicity with increasingly negative O-MWCNT surface charge (increasing PZC). Even though aggregating samples show significantly fewer toxic effects, there are still some well-dispersed samples that similarly show little to no toxicity. This indicates that PZC is not the only contributor to the observed toxicity and that other properties play a meaningful role. Our model indicates that variation in embryo toxicity between well-dispersed samples could be explained by differences in the size and/or shape of aggregates in solution. Specifically, particles with a larger hydrodynamic radius as well as those with a more compact and spherical morphology (larger D_f) tend to appear less toxic to dechorionated embryos.

In addition to physical nanomaterial properties, many studies have hypothesized ROS generation and oxidative stress as a mechanism of toxicity in aquatic organisms exposed to nanomaterials. This has been hypothesized or shown for non-carbonaceous[84, 89, 90] and carbonaceous[33, 34, 91] nanomaterials, with correlations between oxidative stress and zebrafish embryo toxicity well-established for the former and less so for the latter (with no studies explicitly evaluating the ROS- or oxidative stress-based toxicity of MWCNTs). In this study, the O-MWCNT mediated conversion of GSH to GSSG was negatively correlated with embryonic zebrafish mortality, where samples that have a low conversion of GSH (low levels of induced oxidative stress) tend to also have high levels of toxicity. This disagrees with the notion that an increase in oxidative stress (i.e. increased loss of GSH) would yield an increase in toxicity. Further, reactive oxygen species such as superoxide, hydroxyl radicals, and singlet oxygen were not generated by O-MWCNTs, while hydrogen peroxide was only generated in trace quantities. Therefore, since our results do not support the hypothesis of oxidative stress- or ROS-driven toxicity by O-MWCNTs, loss of GSH and the generation of each specific ROS was not considered during the regression modeling. It should be noted that even when GSH conversion is considered in a multivariate model, it still does not emerge as a significant predictor of embryo toxicity.

Given that ROS/oxidative stress is not a significant predictor of zebrafish toxicity, the question as to which O-MWCNT properties may significantly contribute to zebrafish mortality remains. Accordingly, two multiple binomial logistic regression analyses were performed considering oxygen concentration, PZC, aggregate morphology, and hydrodynamic radius – one for all of the samples (Groups 1 and 2) and one only considering well-dispersed (Group 1) O-MWCNTs. The properties of PZC and aggregate morphology have predictive power for both cases (Table 2), but their relative significance changes depending on whether both Groups 1 and 2 or only Group 1 is considered in the analysis. In both models, PZC emerges as the parameter with a stronger effect on the outcome. This is indicated by its larger absolute standardized coefficient (β_S,PZC), a statistical comparison of the relative strength of the effect of each independent variable in a regression analysis. Similarly, in both models, as the surface charge becomes more negative (lower PZC), the toxic response of the embryos increases. As we have shown, samples that aggregate are generally less toxic than most of those that are well-dispersed, indicating that PZC is may be an important indicator of toxicity because it is also a strong indicator of aggregation potential. Further, the surface charge of the particles can also influence electrostatic interaction with the embryos.

Table 2:

24 hpf zebrafish mortality logistic model results

	All O-MWCNT Samples	Group 1 Samples, Only
Intercept	6.111	6.218
Intercept Std. Error	0.922	0.962
Intercept p-value	0.000	0.000
Parameter 1	Point of Zero Charge (PZC)
Parameter 1 Coefficient, βPZC	−0.827	−0.699
Parameter 1 Std. Error	0.106	0.170
Parameter 1 p-value	0.000	0.000
Parameter 1 Standardized Coefficient, βS,PZC	−2.718	−1.576
Parameter 2	Fractal Dimension (Df)
Parameter 2 Coefficient, βDf	−1.571	−1.969
Parameter 2 Std. Error	0.588	0.689
Parameter 2 p-value	0.008	0.004
Parameter 2 Standardized Coefficient, βS,Df	−0.791	−1.088
Model Summary
βS,PZC:βS,Df Ratio	3.436	1.449
McFadden's R-sq	0.271	0.19
Null Deviance	435.8 on 319 DoF	331.4 on 239 DoF
Residual Deviance	317.5 on 317 DoF	268.2 on 237 DoF
Akaike information criterion (AIC)	323.49	274.19

DoF = Degrees of freedom

When considering samples that are relatively well-dispersed, the morphology of the aggregates increases in importance as a predictor of toxicity. In this case, while PZC still emerges as the stronger predictor, based on its larger absolute β_S,Df value, the ratio of β_S,PZC: β_S,Df more than halves, indicating an increase in relative importance of aggregate morphology as a predictor when only considering well-dispersed samples. Once it is established that a sample is relatively stable in solution, it can be seen that increasing the sample’s D_f generally results in lower toxicity. This relationship follows literature precedent for nanomaterial toxicity, especially since all aggregates in Group 1 are in the “nanoscale”, with average hydrodynamic radii below 90 nm. It has been proposed that stiffer, “sharper”, more linear MWCNTs (D_f closer to 1) can rotate and penetrate a cell’s membrane to disrupt membrane integrity and allow for greater cell uptake[92, 93]. Meanwhile, shorter MWCNTs and those with high length:diameter ratios that very readily buckle (and therefore appear more spherical in solution; D_f closer to 3) do not penetrate cells as easily[94]. This increased chance for cell uptake or penetration may be driving the increased toxicity of well-dispersed samples with linear morphologies. These findings suggest that it is likely that physical contact mechanism is a major driving factor behind well-dispersed O-MWCNT zebrafish toxicity.

3.7. Evaluating Physical Contact Mechanism of Toxicity

Due to the high complexity of evaluating interactions between nanomaterials and individual cells, it is common to use lipid vesicles as a model system to represent and quantitate the interaction between these two entities[77]. It has been shown that dye leakage from CF-encapsulated vesicles in the presence of carbonaceous nanomaterials is driven by physical contact indicating an increase in contact between the MWCNT samples and the vesicles. As such, a select group of 4 O-MWCNT samples were chosen, representing variable dispersibilities and aggregate shapes: 1-400 (highly aggregating, D_f=1. 284), 2-UA (well-dispersed, D_f=1.00), 8-400 (well-dispersed, D_f=1.641), and 8-900 (well-dispersed, D_f=1.936). Each of these were exposed to vesicles with a zeta potential value of −8.2 ± 1.3 mV in the zebrafish media, indicating a slightly negative surface charge. Over a 3-hour period of exposure time, there does not seem to be a significant change in the relative concentration of penetrated vesicles, C/C_max with respect to time. (Figure 6a). However, there is clearly variability in C/C_max with regards to samples, where 2-UA appears to disrupt the vesicles the most, and 1-400 and 8-900 disrupting the vesicles the least. Further, there is a notable trend, such that as the fractal dimension of a particle increases, both the mortality at 24 hpf and the C/C_max at 3 hours decrease for well-dispersed samples (Figure 6b). The only sample that did not follow this trend was 1-400, which do not disperse well in solution, had a very low C/C_max, and a low mortality due to minimal contact with the vesicles.

Figure 6:

Interaction of select O-MWCNTs with lipid vesicles containing 5(6)-carboxyfluorescein (CF), where (a) shows the change in normalized fraction of leaked CF (C/C_max) over a time period of 3 hours, and (b) shows the relationship between the fractal dimension of each O-MWCNT and the mortality (at 24 hpf) and C/C_max (at 3 hours after exposure).

Since the surface charge of both the MWCNT samples and the vesicles were negative, electrostatic attraction was limited. As a result, this experiment specifically aims to determine if there is a link between aggregation, aggregate shape, and a physical toxicity mechanism. It is worth noting that the surface charge of these three particles, from most to least negative is 8-400 < 2-UA < 8-900 < 1-400, while the C/C_max for each sample, from lowest to highest is 8-900 < 1-400 < 8-400 < 2-UA. Since all samples were electrostatically repulsed, this suggests that the differentiating factors are shape of each aggregate and the aggregation of the particles supporting the hypothesis that O-MWCNTs physically interact with biological systems and that this physical interaction plays a role in their toxicity.

4. Conclusions

A total of 20 O-MWCNT samples with varying lengths and levels of oxygen surface group functionalization were evaluated for toxicity to chorionated and dechorionated zebrafish embryos. Our study showed that there was no significant mortality induced by O-MWCNTs when the chorion was intact, but there was significant mortality for O-MWCNT loadings of 25 mg/L and 50 mg/L in solution with dechorionated zebrafish embryos for half of the samples, especially those that tended to be more well-dispersed. Based on the two logistic regression models we ran, we found that when all 20 O-MWCNT samples were in the model, the PZC of the nanoparticle emerged as the property that had the most influence on toxicity. However, when only considering samples that were well-dispersed in the zebrafish embryo media, the morphology of the aggregates appeared to be an increasingly influential property impacting the toxicity of each O-MWCNT. This suggests that two of the main factors driving toxicity of O-MWCNTs are their dispersibility (contact) and their aggregate shape in solution. These findings, along with the results of exposing select O-MWCNT to lipid vesicles acting as a model cell, suggest that the mechanism of toxicity of these materials relies on physical interaction.

These findings demonstrate the importance of performing a complete suite of relevant physicochemical characterizations when trying to understand the toxicity of particles at the nanoscale. This includes evaluating aggregation behavior of nanomaterials in the appropriate system conditions (e.g., biological media) to enhance the utility and accuracy of future aqueous bioassays (e.g., zebrafish embryos, mammalian cells) and to reconsider previously reported findings where such things were not explicitly considered. Particle aggregation in the relevant solutions likely leads to a decrease in effective dose compared to the initial dose of the bioassay. As a result, nanomaterials may appear to have a lower toxicity concern due to a lack of exposure[95], leading researchers to believe a material is less hazardous than it may actually be. For the zebrafish embryo assay and other bioassays in aqueous media to be appropriate for toxicity studies moving forward, it is essential for nanotoxicologists to understand the aggregation behavior of their materials, to ensure their results are valid to evaluate hazard based on actual, rather than assumed, exposure.

The approach and findings presented here can aid in the safe and sustainable design of nanomaterials moving forward. It is essential to understand structure-property-function and structure-property-hazard relationships[11], to ensure that the widespread replacement of traditional materials in products with high-performing nanomaterials does not lead to a regrettable substitution[14] based on valid toxicity assessments.