A Meta-analysis of Ecotoxicological Hazard Data for Nanoplastics in Marine and Freshwater Systems.

There is emerging concern about the potential health and environmental impacts of nanoplastics in the environment. Information on exposure has been lacking, but a growing amount of ecotoxicological hazard data is now available, allowing a hazard assessment to be conducted for nanoplastics in freshwater and marine systems. Based on a critical evaluation of published studies and the construction of probabilistic species sensitivity distributions (PSSDs), we present a comprehensive, state-of-the-art understanding of nanoplastic ecotoxicity. Different freshwater and marine datasets were constructed based on different data quality levels, and for each of the datasets, PSSDs were built for both mass- and particle number-based concentrations. Predicted no-effect concentrations (PNECs) were then extracted from the PSSDs. We report PNECs at 99 and 72 μg L-1 , respectively, for the freshwater and marine dataset after the removal of data measured in the presence of sodium azide (NaN3 ), which is considered to be a major interfering factor in the ecotoxicity testing of nanoplastics. By comparing the PNECs, we found that nanoplastics are less toxic than microplastics and many engineered nanomaterials. In addition, the effects of size and polymer type on toxicity were also statistically tested. We observed no significant difference in ecotoxicity for nanoplastics of different sizes, whereas polystyrene nanoplastics were significantly more toxic than all other tested nanoplastics. In conclusion, the results we present provide a comprehensive description of nanoplastic ecotoxicity based on current knowledge. The results constitute a fundamental step toward an environmental risk assessment for nanoplastics in freshwater and marine systems. Environ Toxicol Chem 2020;39:2588-2598.

INTRODUCTION

Since the 1930s, the growth rate of production and use of plastics has been increasing rapidly and has surpassed all other man‐made materials (Brydson 1999). Due to their durability and versatility, plastics are widely used in construction, packaging, textiles, and many other sectors. The yearly global plastic production increased from 2 million tons in 1950 to 381 million tons in 2015, 42% of which were used as single‐use packaging (Geyer et al. 2017). Large amounts of plastics flow into our environment due to mismanagement of waste including disposal in dumps or in open, uncontrolled landfills. The discarded plastics can then enter air, water, and soil in various ways. These plastics can then further fragment into smaller and smaller pieces. Plastic debris smaller than 5 mm was defined as microplastics (Thompson et al. 2004), and this definition is now widely accepted (Joint Group of Experts on the Scientific Aspects of Marine Environmental Protection 2015). Microplastics can further fragment into so‐called nanoplastics (Koelmans et al. 2015). The definition of nanoplastics is still under debate, and the dispute is whether the upper size limit should be at either 1000 nm or 100 nm (Hartmann et al. 2019). There are increasing numbers of laboratory studies available that shed light on nanoplastic formation mechanisms such as photo‐degradation, thermal degradation, mechanical degradation, and bio‐degradation (Lambert et al. 2013; Magrì et al. 2018; Ekvall et al. 2019; González‐Fernández et al. 2019).

Increasing numbers of ecotoxicological tests have recently been carried out to study the adverse effects of nanoplastics on species living in freshwater (Cui et al. 2017; Pitt et al. 2018; Yuan et al. 2019), seawater (Gambardella et al. 2018; Rist et al. 2019; Zhao et al. 2019), and soil (Jiang et al. 2020; Lian et al. 2020). Similar to microplastics, nanoplastics can affect the metabolism, fertility, and mortality of aquatic organisms. However, there is no significant difference between nanoplastics and microplastics with regard to the general effects (Chae and An 2017). Moreover, most laboratory studies were conducted at unrealistically high micro‐/nanoplastic concentrations, and fish and small crustaceans are over‐represented in those effect studies (de Sá et al. 2018).

For microplastics, a handful of complete risk assessment studies have been published. A risk assessment of microplastics in freshwater, which combined available hazard and exposure data using species sensitivity distributions (SSDs) and a probabilistic exposure assessment implied that there is currently negligible risk for freshwater species except for a very small probability in Asia (Adam et al. 2019). Besseling et al. (2019) constructed an SSD combining marine, estuarine, and freshwater lowest‐observed‐effect concentrations (LOECs). Both of these microplastic risk assessments included the nanoplastic size range in their hazard assessment. For nanoplastics alone, Venâncio et al. (2019) analyzed the available studies reporting the median lethal and effect concentration (LC50/EC50) values for marine algae and built an SSD with the data. These authors considered 3 types of nanoplastic: pristine polystyrene, cationic pristine polystyrene (PS‐NH2), and poly(methylmethacrylate) (PMMA). So far, no further analysis of different factors affecting the toxicity has been conducted because there were not enough data available for one single species. For microplastics, Adam et al. (2019) tested the effect of polymer type, but no significant difference in toxicity was observed for different polymers. So far, no complete risk assessment for nanoplastics has been performed due to missing information on both exposure and hazard assessment.

It is crucial to note that most of the published nanoplastic ecotoxicity studies were conducted using pristine polystyrene nanoparticles because they are commercially available as nanoplastic suspension with well‐characterized particle sizes (Pikuda et al. 2019). However, 2 independent studies have shown that the presence of biocides in these suspensions is seriously affecting the interpretation of the toxicity studies (Pikuda et al. 2019; Heinlaan et al. 2020). It was shown that after dialysis of the suspension and consequently removal of the sodium azide (NaN3) stabilizer, the pristine polystyrene nanoplastics were no longer toxic at the levels reported before (Pikuda et al. 2019; Heinlaan et al. 2020). Therefore, the results of published studies with pristine polystyrene nanoplastics could be seriously biased by the added stabilizer.

The aim of the present study was to perform a hazard assessment for nanoplastics in marine and freshwater systems based on a critical evaluation of published studies and the construction of probabilistic species sensitivity distributions (PSSDs) following Gottschalk and Nowack (2013). A statistical analysis of the effects of nanoplastic size and polymer type on their ecotoxicity was conducted. The initial complete dataset was refined by removing both data obtained in the presence of NaN3 and those studies that only reported “highest‐observed‐no‐effect concentrations” (HONECs).

MATERIALS AND METHODS

Data collection and extraction

A literature search was carried out with the keywords “microplastic” and “nanoplastic” covering the timeframe until April 2020. The present study only included peer‐reviewed effect studies that used species living in freshwater and marine waters. Studies in which the material size was larger than 1 μm, as well as those with no size information, were excluded. Target endpoints were population growth, mortality, reproduction (embryo development included), and photosynthesis. When both acute and chronic data were reported in the same study, we only considered the chronic data because they are more relevant for the hazard assessment within the risk assessment framework under the European Union Registration, Evaluation, Authorisation and Restriction of Chemicals (REACH) guidelines (European Chemicals Agency 2008).

Data harmonization

We extracted toxicity data reported in a variety of different dose descriptors: half‐effect or inhibition concentration (E/IC50), LC50, LOEC, HONEC, and no‐observed‐effect concentration (NOEC). The HONECs were extracted from studies in which even at the highest tested concentration, no effect was observed. However, a HONEC was considered not to be of high reliability and therefore it was only obtained from a study when no other dose descriptor was reported and the highest tested concentration was larger than 1 mg L–1. For building the PSSD, we harmonized the various descriptors into chronic NOECs by applying 2 uncertainty factors, as described in Wigger (2020). The uncertainty factor UFdescriptor (UFd) transforms all dose descriptors into NOEC values (Supplemental Data, Table S1). Acute data were converted into chronic data by applying an uncertainty factor UFtime (UFt; Supplemental Data, Table S2). We also transformed all mass concentrations into particle‐based number concentrations and vice versa. The transformation was performed using the given particle diameter and the known densities of the used materials.

Data subgrouping

The ecotoxicity data were grouped into 2 datasets for marine and freshwaters. Those 2 datasets are hereafter termed the “full dataset.” This dataset serves as a reference point to which published data can be compared but is not intended to be used for any hazard assessment. As a next step, we removed data obtained from materials containing NaN3 as long as it was not removed from the suspension by dialysis. We found the information on the presence of the biocide either in the published articles or in the material safety data sheets provided on the websites of the nanoplastic suppliers. Nanoplastics synthesized by the authors of a study, with no description of adding any stabilizer in the article Methods section, were considered as NaN3‐free. If no material safety data sheet was available on the websites of the suppliers, the companies were contacted for further information. Using this approach, the presence or absence of biocides could be confirmed for all published studies. Detailed information for all studies is listed in the Supplemental Data, Table S6. After this procedure, we obtained 2 datasets for freshwater and marine systems without NaN3 data. In a next step, we further removed the HONEC values and obtained 2 datasets without NaN3 and HONEC data. Because pristine polystyrene nanoplastics are the most tested nanoplastics in ecotoxicity studies, the dataset is highly influenced by this material. To obtain a dataset for other plastics, we removed all marine pristine polystyrene data and got a dataset that was still sufficient for making a PSSD. This final PSSD was not performed for the freshwater dataset because only 2 species would have been included.

PSSD+

The SSDs are a widely acknowledged method used in environmental protection, assessment, and management practices (Posthuma et al. 2019). The present study performed the hazard assessment based on the PSSD method developed by Gottschalk and Nowack (2013), which was further refined by Wigger et al. (2020) to include uncertainties and interlaboratory variation (PSSD+ method as described in the Data harmonization section). The PSSD+ method first calculates one probability distribution for each species for all available data points, and then all probability distributions are integrated into a single PSSD by a Monte‐Carlo routine (Wigger et al. 2020). For each dataset, we built PSSDs for both mass and number concentrations.

The fifth percentile of the SSD/PSSD (hazardous concentration for 5% of species [HC5]) was extracted as the concentration below which 95% of the species are protected from the substance, as described in the REACH guidance (European Chemicals Agency 2008). The predicted no‐effect concentration (PNEC) was derived from the HC5 by using an assessment factor of 1. Because we used a probabilistic method, we obtained a probability distribution of the PNEC instead of a single value.

Statistical tests

Correlation tests were performed on the ecotoxicity data to determine whether there was any statistical difference between the NOEC values obtained from different nanoplastic sizes. In addition, a nonparametric statistical hypothesis test was used to test the effect of polymer types on NOEC values. Because toxicity is highly dependent on the tested species, it is necessary to perform the correlation test on a single species. Daphnia magna was chosen because it included the highest number of data points. Pearson correlation tests were performed on the relationship between size and NOEC as a test for a linear relationship. Mann–Whitney tests were performed on the full dataset, the dataset without NaN3, and the dataset without NaN3 and HONECs for the effects of different polymer on NOECs. The Mann–Whitney test was applied to samples that did not follow a normal distribution. The null hypothesis was that the 2 sample means are equal; it was rejected if the p value obtained was <0.05. All statistical tests were performed in R (R Development Core Team 2013).

RESULTS

Hazard data in freshwater

In total, 47 ecotoxicity values were collected for the hazard assessment in the freshwater compartment, covering 15 species (Table 1; full dataset in the Supplemental Data, Table S4). Algae were the most tested species with 7 different species. Two fish, 1 higher plant, and 4 invertebrates were also included in the dataset. Daphnia magna was the most represented species (21 data points, which is 44% of all data points collected). The full dataset met the REACH criteria for assessing environmental risks using the SSD method (European Chemicals Agency 2008) as listed in the Supplemental Data, Table S3. After the removal of data points in which the test particle suspension contained NaN3 as a stabilizer, the number of data points was reduced to 22, with only 6 species represented. This refined database failed to meet 3 criteria in the European Chemicals Agency guideline: occurrence of a second family in the phylum Chordata; occurrence of an insect species; and occurrence of a species of higher plants. After we removed the HONEC values, we obtained a dataset with only 13 data points from 4 species.

Table 1

Summary of ecotoxicological datasets for effects of nanoplastics in freshwaters and marine systems

	Freshwater			Marine
Characteristic of dataset	Full dataset	NaN3 removed	NaN3 and HONECs removed	Full dataset	NaN3 removed	NaN3 and HONECs removed	NaN3, HONECs, and PS data removed
No. of data points	47	22	13	39	25	19	8
No. of species	15	6	4	19	16	14	8
Proportion of chronic data points	47%	73%	46%	56%	64%	68%	100%
Proportion of HONEC	40%	41%	0	29%	24%	0	0
Fish species included	Yes	Yes	No	No	No	No	No
Algae species included	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Crustacean species included	Yes	Yes	Yes	Yes	Yes	Yes	No
Insects species included	Yes	No	No	No	No	No	No
No. of polymer types	5	5	5	5	5	4	3

HONEC = highest‐observed‐no‐effect concentration; PS = polystyrene.

Hazard data in marine water

We collected 39 data points covering 19 species in marine systems (Table 1). The full dataset is available in the Supplemental Data, Table S5. Compared with the primary freshwater dataset, the marine dataset consisted of fewer HONEC values and covered fewer trophic levels. Higher plants, fish, and insect species were missing. When the data points obtained in the presence of NaN3 and all HONECs were removed, the refined datasets still covered 16 and 14 data points, respectively. A dataset is usually considered good in biodiversity coverage when 10 or more taxa are covered (Posthuma et al. 2019). Therefore, relatively high‐quality PSSDs can be built using these 2 datasets as well. There was also the possibility of building a PSSD without pristine polystyrene nanoplastics because 8 data points from 8 different species were still available in this dataset.

Effect of size on NOEC values

In both the freshwater and the marine dataset, all the test nanoplastics were present as spheres. We used the data points of the species D. magna to test the correlation between ecotoxicity (NOEC values) and nanoplastic size. The p values from the Pearson correlation test are listed in Table 2. For the mass‐based NOEC, none of the datasets exhibited any correlation between size and NOEC. However, for the particle‐number–based NOEC, 2 p values were significant (<0.05) for the full dataset and for the data points in which NaN3 was removed.

Table 2

Correlation between the size of nanoplastics and their effects on organisms based on no‐observed‐effect concentration values

Form of NOEC	Dataset description	Group size	Correlation coefficient	p value
Mass‐based NOECs (μg L–1)	Full dataset	21	–0.11	0.63
	NaN3 removed	14	–0.17	0.57
	NaN3 and HONECs removed	9	–0.23	0.54
Particle number‐based NOECs (part/m3)	Full dataset	21	–0.45	0.04*
	NaN3 removed	14	–0.56	0.04*
	NaN3 and HONECs removed	9	–0.47	0.25

The corresponding correlation graphs are shown in the Supplemental Data, Figure S5.

Statistically significant at p < 0.05.

NOEC = no‐observed‐effect concentration; HONEC = highest‐observed‐no‐effect concentration.

Effect of polymer type on ecotoxicity

The dataset of D. magna was also selected to test for the effect of polymer type on ecotoxicity. The data were grouped into 3 subgroups: pristine polystyrene, modified pristine polystyrene (PS‐COOH and PS‐NH2), and non‐pristine polystyrene (PMMA and polyhydroxybutyrate [PHB]). Six Mann–Whitney tests were performed, and the results are shown in Table 3. There was no significant difference between pristine polystyrene and modified pristine polystyrene in both the full dataset and the dataset with NaN3 data removed, for both mass and number concentrations. However, the difference between the non‐pristine polystyrene group and the modified pristine polystyrene group was always statistically significant (p < 0.05). The pristine polystyrene and non‐pristine polystyrene groups were significantly different for the mass‐based NOECs in both datasets but not the number‐based one.

Table 3

Effect of polymer type on no‐observed‐effect concentration values for nanoplastics using a Mann–Whitney test

Dataset (sample size)	Subsamples	NOEC form	p value
Full dataset (21)	PS vs non‐PS	Mass	0.007*
		Particle number	0.083
	PS vs modified PS	Mass	0.203
		Particle number	0.397
	Non‐PS vs modified PS	Mass	0.009*
		Particle number	0.009*
NaN3 removed (14)	PS vs non‐PS	Mass	0.035*
		Particle number	0.403
	PS vs modified PS	Mass	0.389
		Particle number	0.270
	Non‐PS vs modified PS	Mass	0.037*
		Particle number	0.020*

Statistically significant at p < 0.05.

NOEC = no‐observed‐effect concentration; PS = polystyrene.

Probabilistic hazard assessment

We built 14 PSSDs based on chronic NOECs for 7 different datasets in both mass and number concentrations. Four of the SSDs are shown in Figures 1 and 2, and the others are given in the Supplemental Data, Figures S1 to S5. All data points from the same species were plotted, and the geometric means of the NOECs from each species were used to sort the species to form the PSSD. It is important to note that 10 000 species distributions were calculated based on Monte‐Carlo simulations. The lines in Figures 1 and 2 show the mean PSSD, and the different color shadings indicate the probability distributions of the PSSD. Shown are the Q5 (5th quantiles of species probability distribution), Q25, Q75, and Q95. Different polymers are represented by different colors.

Figure 2

The probability species sensitivity distributions (PSSDs) for nanoplastics in freshwater built on the dataset without NaN3. (A) Mass concentrations; (B) number concentrations. NOEC = no‐observed‐effect concentration; Q = quantile (e.g., Q25 = 25th quantile). The full genus‐level names are listed in the Supplemental Data, Table S7.

PSSDs for freshwater species

For freshwater species, PSSDs were built for both mass concentrations (Figure 1A) and number concentrations (Figure 1B) based on the full dataset. Five different types of polymers were found in both PSSDs: pristine polystyrene, PS‐COOH, PS‐NH2, PMMA, and PHB. In total, 16 species were covered in both PSSDs. The lowest NOEC in freshwater was observed in tests with the higher plant species Ceratopteris pteridoides (geometric mean NOEC = 80 μg L–1) in the mass‐based PSSD (Figure 1A), and the algae Chlorella pyrenoidosa (geometric mean NOEC = 4.6 × 1013 part m–3) in the number‐based PSSD (Figure 1B). The highest NOEC was observed in tests with Heterocypris incongruens both for the mass‐based (geometric mean NOEC = 1.0 × 106 μg L–1) and for the number based SSD (geometric mean NOEC = 1.4 × 1018 part m–3). A wide range of NOEC values was collected for D. magna, with a minimum of 1.0 × 103 μg L–1 and a maximum of 1.0 × 106 μg L–1 or from 1.8 × 1015 part m–3 to 2.3 × 1019 part m–3 (Supplemental Data, Tables S4 and S5).

Figure 1

The probability species sensitivity distributions (PSSDs) for nanoplastics in freshwater built on the full dataset. (A) Mass concentrations; (B) number of concentrations. NOEC = no‐observed‐effect concentration; Q = quantile (e.g., Q25 = 25th quantile); PMMA = poly(methylmethacrylate); PHB = polyhydroxybutyrate; PS = pristine polystyrene; PS‐COOH = modified pristine polystyrene; PS‐NH2 = cationic pristine polystyrene. The full genus‐level names are listed in the Supplemental Data, Table S7.

In Figure 1A and B, the data obtained in the presence of NaN3 are represented by triangles. Those data points occupied not only the lowest part of the SSD curve but also the highest part. We then built PSSDs without these data points (Figure 2A and B). There were still 5 types of polymers present in both PSSDs, with pristine polystyrene being the dominant type, accounting for 50% of all data. The number of species covered by the PSSD was reduced from 15 to 6, and the order of sensitive species changed accordingly. Danio rerio, a fish, became the most sensitive species (geometric mean NOEC = 320 μg L–1). Chlorella vulgaris, an alga, had the lowest NOEC of all in number‐based metrics (geometric mean NOEC = 3.1 × 1014 part m−3). In tests with Scenedesmus obliquus the highest NOECs were observed in both mass‐based (geometric mean NOEC = 2.2 × 104 μg L–1) and number‐based units (geometric mean NOEC = 6.9 × 1014 part m–3).

We built further refined the PSSDs by removing all the HONEC data (Supplemental Data, Figure S1A and B). Only 4 species were covered by each of the 2 PSSDs, and D. magna and C. vulgaris became the most sensitive species in the mass and number‐based PSSDs, respectively.

PSSD for marine species

Mass‐based (Figure 3A) and number‐based (Figure 3B) PSSDs were built for species living in marine waters, based on the data of the full marine dataset. Five different polymers were identified: pristine polystyrene, PS‐COOH, PS‐NH2, PMMA, and polyvinyl chloride (PVC). Nineteen different species were covered in both PSSDs. The lowest NOEC was observed with a mussel species, Mytilus galloprovincialis (geometric mean NOEC = 1.4 μg L–1). Another mussel, Meretrix meretrix (geometric mean NOEC = 1.3 × 1012 part m−3), moved from the second most sensitive species (Figure 2A) to the most sensitive in the number‐based PSSD (Figure 3B). The species with the highest NOECs for mass‐ and number‐based PSSDs were Karenia mikimotoi (geometric mean NOEC = 2.5 ×104 μg L–1) and Tetraselmis chuii (geometric mean NOEC = 3.4 × 1017 part m–3), respectively. The widest range of single NOECs was observed for Dunaliella tertiolecta, a green algae, ranging from 0.5 to 5 × 104 μg L–1 and from 9 × 1012 to 1.4 × 1018 part m–3. Brachionus plicatilis, a rotifer, with 7 reported ecotoxicity values, had the highest number of NOEC values.

Figure 3

The probability species sensitivity distributions (PSSDs) for nanoplastics in marine system built on the full dataset. (A) Mass concentration; (B) number concentration. NOEC = no‐observed‐effect concentration; Q = quantile (e.g., Q25 = 25th quantile). PMMA = poly(methylmethacrylate); PVC = polyvinyl chloride; PS = pristine polystyrene; PS‐COOH = modified pristine polystyrene; PS‐NH2 = cationic pristine polystyrene.

When we removed the data in the presence of NaN3 (Figure 4A and B), the number of species was reduced from 19 to 16. Artemia salina (geometric mean NOEC = 65 μg L–1) became the species with the lowest NOEC, and D. tertiolecta (geometric mean NOEC = 5 × 104 μg L–1) was the species with highest NOEC of the PSSD (Figure 2A). In the number‐based PSSD (Figure 4B), Skeletonema costatum (geometric mean NOEC = 6.9 × 1011 part m–3) was the most sensitive species, but the geometric mean NOEC decreased compared with the full dataset. This was because the highest NOEC data points for S. costatum were removed because they contained NaN3. Dunaliella tertiolecta was still the least sensitive species in the refined PSSD.

Figure 4

The probability species sensitivity distributions (PSSDs) for nanoplastics in marine system built on the dataset without NaN3. (A) mass concentration; (B) number concentration. NOEC = no‐observed‐effect concentration; Q = quantile (e.g., Q25 = 25th quantile). PMMA = poly(methylmethacrylate); PVC = polyvinyl chloride; PS = pristine polystyrene; PS‐COOH = modified pristine polystyrene; PS‐NH2 = cationic pristine polystyrene. The full genus‐level names are listed in the Supplemental Data, Table S7.

Similar to what we had done for the freshwater dataset, we were able to build a further refined PSSD in which, in addition to the NaN3 data, all HONECs were removed (Supplemental Data, Figure S2A and B). Fourteen marine species were covered in both PSSDs, only 2 species less than in the refined PSSD. Artemia salina and S. costatum remained the most sensitive species for mass‐ and number‐based PSSDs, whereas K. mikimotoi and T. chuii became the least sensitive species for mass‐ and number‐based PSSDs, respectively. The change was small compared with the NaN3‐free PSSDs because not many HONECs were included in that marine dataset.

We built a final PSSD on a dataset in which the pristine polystyrene data were removed from the former dataset. The most and least sensitive species in both mass‐ and number‐based PSSDs remained the same (Supplemental Data, Figure S3A and B), but the number of species was reduced to 8.

PNEC values

The PNEC distribution for the 7 datasets was extracted as the 5th percentile of the PSSD datasets. The statistical evaluation of the PNECs is presented in Table 4 in the form of the median value followed by the range between the 25th (Q25) and the 75th quantiles (Q75). The complete PNEC distributions are shown in the Supplemental Data, Figure S4. The median PNECs for the full freshwater and marine dataset were 71 and 1.3 μg L–1 or 8.6 × 1012 part m–3 and 2.3 × 1012 part m–3. The PNEC increased to 99 μg L–1 after the NaN3 data were removed, whereas there was no change when the HONEC data were further removed from the freshwater dataset. For the marine hazard assessment, the PNEC after removal of the NaN3 data increased to 71 μg L–1. After removal of both the NaN3 and HONEC data, the PNEC remained the same. When the pristine polystyrene data were also removed, the PNEC increased to 500 μg L–1. Generally, the mass‐based PNECs increased with removal of data that were at the lower fractions of the mass PSSDs. On the other hand, the number‐based PNECs showed a different trend, with either no large change for the freshwater PSSD or a decrease for the final marine datasets. This is because some sensitive data were removed in the freshwater datasets. In Supplemental Data, Figure S4B, there was a similar shifting trend of mass PNEC distribution for the marine dataset. Due to the small proportion of HONECs in the dataset, the 2 distributions overlapped after the removal of HONEC data. However, for the number‐based PSSD in the marine system, the PNEC distributions of the refined datasets shifted to lower NOECs after NaN3 and HONEC data were removed. The 3 most sensitive data points from 3 different species in the full dataset were very close (Figure 2B). One of them (from S. costatum) remained in the final dataset and made the probability distribution narrower. In such a case, the PNEC distribution could shift to the left when some of the data were removed because not only does the most sensitive species contribute to the PNEC but also some low data points of other species.

Table 4

Statistical summary of the predicted no‐effect concentration distributions for nanoplastics in freshwater and marine systems in both mass and number concentrations

Dataset description	Mean PNEC (μg/L)	Median [Q25, Q75] PNEC (μg/L)	Mean PNEC (part/m3)	Median [Q25, Q75] PNEC (part/m3)
Freshwater
Full dataset	71	71 [54, 87]	2.9 × E + 13	8.6 × E + 12 [6.6E + 12, 1.5E + 13]
NaN3 removed	99	99 [76, 120]	6.8 × E + 13	1.7 × E + 13 [1.3E + 13, 1.3E + 14]
NaN3 and HONECs removed	2700	2700 [1200,4100]	8.8 × E + 12	8.8 × E + 12 [6.9E + 12, 1.1E + 13]
Marine
Full dataset	1.3	1.3 [1.0, 1.6]	2.5 × E + 12	2.3 × E + 12 [1.3E + 12, 3.5E + 12]
NaN3 removed	72	70 [49, 93]	6.9 × E + 11	6.9 × E + 11 [5.3E + 11, 8.5E + 11]
NaN3 and HONECs removed	72	70 [49, 94]	6.9 × E + 11	6.9 × E + 11 [5.4E + 11, 8.4E + 11]
NaN3, HONECs, and PS removed	496	500 [390, 660]	6.9 × E + 11	6.9 × E + 11 [5.3E + 11, 8.4E + 11]

PNEC = predicted no‐effect concentration; PS = polystyrene; HONEC = highest‐observed‐no‐effect concentration.

DISCUSSION

We built in total 14 PSSDs for freshwater and marine systems, based on mass and number concentrations, and thus obtained datasets of different quality levels. The PSSDs based on the full dataset covered more biodiversity than the existing nanoplastic SSDs (Besseling et al. 2019; Venâncio et al. 2019), spanning 15 species in the freshwater system and 19 species in the marine system compared with 10 and 16 in Besseling et al. (2019) and Venâncio et al. (2019), respectively. The most sensitive species in our mass‐based marine PSSD was the mussel M. galloprovincialis, which was also reported by Venâncio et al. (2019). These authors used an EC50 (0.14 mg L–1) directly for M. galloprovincialis based on the ecotoxicological study of Balbi et al. (2017). However, we applied a UFt of 10 and a UFd of 10 to transform the ecotoxicity data from the EC50 to a chronic NOEC of 1.4 μg L–1.

An important part of our data curation effort was to remove data from experiments in the presence of NaN3. The PNECs shifted from 71 and 1.3 to 99 and 72 μg L–1 for freshwater and marine PSSDs, respectively, when NaN3 data were removed. The ecotoxicity reduction effect is in line with the proven effect of the suspension stabilizer NaN3 on biasing the ecotoxicity of pristine polystyrene nanoplastics (Pikuda et al. 2019; Heinlaan et al. 2020). These resulting PSSDs better reflect the real influence of nanoplastics on freshwater and marine species. The presence of toxic components or byproducts has also been influencing the hazard assessment of engineered nanomaterials in an early phase of research, and much attention has been given to the critical evaluation and harmonization of test protocols (Spohn et al. 2009; Hartmann et al. 2015). The full dataset and the associated PSSDs and PNEC values are therefore presented in the present study only to support the importance of data curation and should not be used to form any conclusions about the hazards of nanoplastics.

We also built PSSDs for datasets in which NaN3 and HONECs were both removed. The freshwater dataset was smaller than the marine dataset when NaN3 data and HONECs were removed. The freshwater datasets with high data quality covered less biodiversity (6 and 4 species) than those of marine PSSDs (16 and 14 species). The PNECs shifted to 2700 and 72 μg L–1 for freshwater and marine PSSDs, respectively, when both NaN3 and HONEC data were removed. The PNECs for marine datasets did not change after the removal of HONECs, whereas for freshwater datasets, the PNECs rose to a 27 times higher value. This difference is partially due to a smaller portion of HONECs in the NaN3‐free marine dataset (24%) than in the freshwater dataset (41%). Moreover, some of the HONEC data in the freshwater dataset are among the most sensitive values in the SSDs. The HONECs are worst‐case values and are an indication that the toxicity of a material is actually much less than what was expected when the experiment was designed. True NOEC values could be orders of magnitude higher than HONEC values. It is nevertheless important to include HONEC values in the evaluation because, as “no‐effect data,” they provide useful input for assessing safe levels of exposure (Warheit and Donner 2015). The high proportion of HONECs in the freshwater dataset (19 of 47 compared with 11 of 39 in the marine dataset) shows a general low toxicity of nanoplastics to freshwater species compared with marine species. However, the removal of HONEC values reduces the uncertainty of the evaluation and therefore increases the credibility of the hazard assessment.

The mean PNEC of the PSSD built on the marine full dataset of 1.3 μg L–1 was smaller than the HC5 extracted in previous SSDs of 410 (Venâncio et al. 2019) and 5.4 μg L–1 (Besseling et al. 2019). This is because we included more studies reporting effects at low concentrations compared with the work of Besseling et al. (2019). As explained just above, we used uncertainty factors to derive NOEC‐based PNECs, whereas Venâncio et al. (2019) reported EC50‐based HC5 values, resulting in a difference of up to 2 orders of magnitude. However, SSDs used in a regulatory context should always be built based on NOEC values, as indicated in the guidance documents on the use of SSDs for environmental risk assessment (European Chemicals Agency 2008).

We can place the nanoplastic PNECs in context with those of other nanomaterials and microplastics derived using the same SSD method. A comparison of the mass‐based PNECs of the present study with the published PNECs for micro‐ and nanoplastics, engineered nanomaterials (ENMs), and nano‐biomaterials (NBMs) is shown in Figure 5. The PNEC of the NBM polyacrylonitrile is several orders of magnitude higher than the others (Hauser et al. 2019). Ecotoxicity data for other often used polymeric NBMs such as poly lactic‐co‐glycolic acid or poly lactic acid are not available, but generally these materials are considered safe for the environment (Hauser et al. 2019). These polymeric NBMs could be considered to be the closest analog to nanoplastics originating from synthetic polymers.

Figure 5

Comparison of the predicted no‐effect concentrations (PNECs) from the present study and PNECs from previous studies. ENM = engineered nanomaterials (Coll et al. 2016); NBMs = nano‐biomaterials (Hauser et al. 2019); Plastics‐Others = plastic PNECs from other studies (Adam et al. 2019; Besseling et al. 2019; Venâncio et al. 2019); Plastic‐Present = plastics from the present study. The PNECs highlighted with boxes are from the PSSDs built on the highest quality datasets in freshwater and marine systems. All values are PNECs extracted from SSDs except where something else is indicated. PSSD = probability species sensitivity distribution; SSD = species sensitivity distribution; HONEC = highest‐observed‐no‐effect concentration; PS = polystyrene; E/LC50 = median effect/lethal concentration; LOEC = lowest‐observed‐effect concentration; NOEC =no‐observed‐effect concentration.

The nanoplastic PNECs with the highest data quality and regulatory relevance that are labeled in Figure 5 are higher than those of different types of ENMs, suggesting a lesser ecotoxicity for nanoplastics. The order of ecotoxicity among reported ENMs is: carbon nanotubes (CNTs) < titanium dioxide (TiO2) < fullerenes < ZnO < silver (Ag) in freshwater (Coll et al. 2016). The probabilistic risk assessment for those ENMs showed that the order of the risk for freshwater organisms is CNT = fullerenes < TiO2 Ag < ZnO (Coll et al. 2016). The Ag nanoparticles are the most toxic while the environmental concentrations are low, as modeled by Sun et al. (2014), whereas ZnO exhibited high toxicity (Coll et al. 2016) but also high predicted exposure concentrations in freshwater (Sun et al. 2014).

Compared with the mean PNEC of microplastics in freshwaters of 0.07 μg L–1 and 9.5 × 105 part m–3 reported by Adam et al. (2019) and 1.7 μg L–1 (freshwater, estuarine, and marine microplastics) reported by Besseling et al. (2019), our study found PNECs of 99 μg L–1 and 6.8 × 1013 part m–3 for the dataset without NaN3, which is much higher than previous PNECs of microplastics for the mass‐based PNEC and approximately 8 magnitudes higher for the number‐based PNEC. Based on the current state of knowledge, nanoplastics seem to be less hazardous than microplastics in both marine and freshwater systems. However, so far no information on environmental exposure of nanoplastics is available, and thus we cannot make any statements about the environmental risks of nanoplastics that could still exist if the nanoplastic concentration is much larger than the microplastic concentration.

The results of the present study are therefore in contradiction to the expectation that nanoplastics are more hazardous than microplastics (Lehner et al. 2019). Indeed, some studies have shown that smaller sized nanoplastics exhibited higher toxicity than larger particles (Fadare et al. 2019; Li et al. 2020), yet some other studies have also shown no size‐dependent effect or even shown that larger particles were more toxic (Sendra et al. 2019; Choi et al. 2020). To further understand the effect of size on nanoplastic ecotoxicity, we conducted a statistical analysis based on 21 ecotoxicity data for D. magna and found no statistically significant effect of size on ecotoxicity of nanoplastics for the mass‐based NOECs. However, higher toxicity at larger sizes was identified as a significant effect in the number‐based NOECs of the full dataset and the NaN3‐free dataset. It is important to note that the aggregation of nanoplastics was not considered in the present study and that the sizes of the nanoplastics are reported as the mean size in the original product suspension. Moreover, the interaction of nanoplastics with natural organic substances could change the toxicity, for example, humic acids can reduce the toxicity of some nanoplastics by agglomeration (Fadare et al. 2019; Liu et al. 2019). Therefore, we can expect even lower ecotoxicity of nanoplastics in the environment than what we reported in the present study. The large range of NOEC values for D. magna of up to 4 orders of magnitude between the smallest and largest NOECs was also found in microplastic risk assessment (Adam et al. 2019), and especially in risk assessment of ENMs (Coll et al. 2016), where the range can even span 6 orders of magnitude. This variability can be caused by different forms of the materials or by varying exposure conditions in addition to the normal intraspecies variability. To reduce the interlaboratory variabilities, a quality assessment method for microplastic effect studies should be taken into account (de Ruijter et al. 2020).

Pristine polystyrene nanoparticles (unmodified and modified with functional groups) are widely tested in the laboratory for nanoplastic ecotoxicity, but they are not very representative of the plastics found in the environment. Koelmans et al. (2019) reviewed studies on microplastic occurrence data from river and lake water, groundwater, tap water, and bottled drinking water and found that the order of microplastic polymer type detected globally is polyethylene ≈polypropylene > pristine polystyrene > PVC > polyethylene terephthalate (PET). Elkhatib and Oyanedel‐Craver (2020) found in their review that polyester microplastic was the most reported in effluents of wastewater, followed by polyethylene and PET. In seawater, the main polymer types are polyester, polyethylene and PET in the sampled microplastics (Coyle et al. 2020), whereas pristine polystyrene does not appear on the list of identified polymers types at all. The polymer composition of nanoplastics in environmental samples is not yet available. However, we can expect that the environmental composition of nanoplastics would be similar to that of the microplastics. We can therefore expect that pristine polystyrene nanoplastics will not be a common type of nanoplastic. In the environment. The statistical analysis performed in the present study suggests that non‐pristine polystyrene nanoplastics are less toxic than pristine polystyrene nanoplastics and pristine polystyrene nanoplastics modified with function groups, which means that common environmental nanoplastics could have a smaller ecotoxicity than that based on the dataset incorporating all polymer types. A pristine polystyrene‐specific PSSD could not be built because most of the pristine polystyrene test suspensions contained NaN3.

Hazard data as evaluated in the present study constitute one part of the environmental risk assessment of nanoplastics. Because so far no information on environmental exposure is available, a complete risk assessment cannot be performed, in contrast to microplastics, for which abundant exposure information is available. The particle number‐based PSSDs could be used to directly compare the effect with measured exposure data once they are available (usually reported in the form of particle number concentrations). However, ecotoxicity data are usually more accurate in the mass concentration metric, which is what is normally reported in ecotoxicity studies.

CONCLUSIONS

The meta‐analysis of ecotoxicological hazard data for nanoplastics in marine and freshwater systems presented in our study provides a comprehensive and state‐of‐the art description of nanoplastic ecotoxicity. We stratified the ecotoxicity data into different quality levels, which allows for an analysis without the bias caused by NaN3 and HONEC data. The current state of knowledge is far from perfect, especially because there are few ecotoxicological hazard data for the nanosized form of the most common polymer types such as PET, polyethylene, and polypropylene. With the PNEC values we obtained from the PSSDs, we took a first step toward environmental risk assessment of nanoplastics. However, to reach a full environmental risk assessment of nanoplastics, we need the exposure data that are missing for the time being.

Supplemental Data

The Supplemental Data are available on the Wiley Online Library at https://doi.org/10.1002/etc.4887.

This article includes online‐only Supplemental Data.

Supporting information.

Click here for additional data file.