Low concentration effects and different outcome in repeated reproduction tests with silver nanoparticles, silver nitrate and Folsomia candida (Collembola)

Toxicity of silver nanoparticles (AgNP) has been studied frequently due to a rise in application in various products. Various studies on AgNP toxicity with terrestrial and aquatic organisms confirmed their negative effects. In our previous experiments, strong variability was observed in the reproduction of Collembola (Folsomia candida) in different repeats. To investigate the effects of silver on the reproduction of Folsomia candida, they were exposed in laboratory-controlled conditions to AgNP and silver nitrate (AgNO3) at a concentration of 30 mg/kg dry soil for 28 days and compared to controlled individuals not exposed to silver. We repeated reproduction tests on the toxicity of silver to Folsomia candida four times throughout one year (April, July, October and January) in order to explore the temporal variability of their outcome. While adult survival was similar in all treatments and seasons, reproduction in the control increased from April to October. Significantly lower reproduction was found in January with only 385–424 juveniles per vessel, compared to 504–633 individuals in other months. Strong toxic effects of both silver treatments were observed in July, April and October. However, AgNP showed no toxic effects on the reproduction of F. candida in January. The relative toxicity of both substances varied between single experiments: AgNP were more toxic than AgNO3 in April and July, and less toxic in October and January. These findings indicate that the reproduction of F. candida in the control had a significant effect on the results of the toxicology experiments. Moreover, we demonstrated the reproductive toxicity of AgNP in soil at a much lower concentration than reported thus far. Therefore, to guarantee reliability and reproducibility, we recommend to disregard any test results where the reproduction rate of F. candida in the control is significantly different from the average in the respective laboratory, even if the validity criteria of the test are met.


Background
The application of silver nanoparticles (AgNP) is strongly increasing in several areas such as electrical, medicine, food and textile products. AgNP can be released from these products during washing [1], disposal, and via industry wastewater [2,3] to the environment. Up to 90% of Ag remains in sewage sludge in which the estimated Ag annual increase is 1.6 µg/kg [4,5]. There are concerns about unintended exposure of humans and the environment to AgNP [6], resulting in a large research effort into the hazards and behavior of AgNP in the environment [7]. Numerous aquatic toxicity Zhang and Filser Environ Sci Eur (2020) 32:136 tests with AgNP and silver ions using a wide range of species have resulted in classifying AgNP as "particularly toxic" [8][9][10]. Nano-sized particles may pass through cell membranes, and the accumulation of intracellular nanoparticles can lead to cell malfunction, and the toxicity is presumed to be size and shape dependent [11]. A combined effect of Ag + and nano-size was illustrated for the discrepancy in toxicity pattern of AgNP compared to AgNO 3 [12][13][14]. Research focusing on the terrestrial matrix has shown that soil pH, texture, organic matter, and ionic composition can affect the toxicity of AgNP to soil invertebrates [15][16][17]. Negative effects of AgNP were shown for reproduction, survival and growth of nematodes [18][19][20], earthworms [20][21][22][23][24] and enchytraeids [24]. The toxicity of AgNP for Collembola was first reported by Waalewijn-Kool et al. [25], who observed no effect on survival and reproduction for Folsomia candida (F. candida) exposed to AgNP (3-8 nm coated with paraffin) at a measured concentration of 673 mg Ag/kg dry soil. Mendes et al. [26] reported a negative effect of AgNP (NM-300 K) on the reproduction of F. candida, with EC 20 and EC 50 values of 173 and 540 mg Ag/kg, respectively.
F. candida Willem 1902 is a common Collembola species that is present in soils all over the world and has been used for ecotoxicological testing in numerous publications [27][28][29][30][31][32][33][34]. Series of tests are conducted to examine mortality, reproduction, bioaccumulation, and effects on the behavior of F. candida to evaluate the toxicity of organic and inorganic contaminants. Filser et al. [35] introduced a miniaturized reproduction test of F. candida according to the OECD 232 (2009) [36] standard reproduction test in which 4 adults and 10-g soil were used instead of 10 adults and 30-g soil.
Based on our years of experience with F. candida, we have found that its reproduction decreases significantly at certain times of the year, particularly in winter. Does this difference in reproduction affect the results of the toxicity test? To test the accuracy and reproducibility of the test, we evaluated the toxicity of AgNP and AgNO 3 on the reproduction of F. candida throughout one year. We hypothesized that (a) the reproduction of F. candida is different between equally repeated reproduction experiments and (b) the toxic effects of AgNP and AgNO 3 vary between repeats.
NM-300 K has been used in a variety of studies and projects, and was included in the OECD Working Party on Manufactured Nanomaterials sponsorship program. As a dispersion in stabilizing agents, NM-300K contains 4% w/w each of polyoxyethylene glycerol trioleate and polyoxyethylene (20) sorbitan monolaurate (Tween 20) with a silver content of 10.16% by weight [37]. It was distributed by the Fraunhofer Institute for Molecular Biology and Applied Ecology (IME) and provided by Joint Research Centre of the European Commission as a part of the UMSICHT project (BMBF 0340091A).

Test soil
RefeSol soils were selected as reference soils by the German Federal Environment Agency, and they matched the properties stated in various OECD terrestrial ecotoxicological guidelines. In this study, we used RefeSol 01-A (provided by the Fraunhofer IME, Schmallenberg, Germany), a loamy sand soil with a pH of 5.67, 0.93% organic carbon, 71% sand, 24% silt, and 5% clay.

Soil preparation and reproduction test
The reproduction test was carried out following a miniaturized version of OECD 232 [35]. F. candida was taken from our lab culture, originally obtained from the working group of Professor Achazi at Freie Universität Berlin in the early 1990s. As the initial difference in age and size may affect the observed difference in reproduction, controlling the age of the individuals is important [38,39]. To synchronize F. candida, adults were placed in a breeding container for three days to lay eggs and then were removed. After hatching, four 9-12-day-old juveniles of similar size were placed randomly in each vessel with 10 g test soil. The vessels were incubated in a climate chamber (Sanyo MLR-350H) at 20 °C with a 12-hour light/12-h dark cycle with 80% humidity and 500 Lux illumination. During the test, 5 pieces of dried baker's yeast (Dr. Oetker) were added to the animals twice a week, and the old food bunches were removed. The test vessels were aerated twice a week, and moisture content of the soil was kept constant at 50% of the maximum water holding capacity by replenishing the water loss once a week. After 28 days of exposure, 100-mL deionized water was added to each test container, and the soil was transferred to a plastic container. F. candida floating on the surface of the dispersion were visible after adding two drops of ink to the water. A picture was taken of each container, to count the juveniles and adults using Image J 1.46r software package.
Stock solutions of AgNO 3 and AgNP were prepared by diluting both substances with deionized water. The flasks with stock dispersions of AgNP were placed in an ultrasonic bath (Bandelin sonorex RK 100H with an output of 160 W, 35 kHz) and sonicated for 20 min before use. After the dispersions and solutions were prepared, they were added to the soil to obtain a concentration of 30 mg Ag/kg dry soil. To obtain a homogeneous distribution, the test substance solution was first added to a small portion of the soil (20 g), which then was mixed to the final test soil thoroughly with a spoon. All soil samples were adjusted to 50% of the maximum water holding capacity and thoroughly mixed. Additionally, a control without any silver chemicals was included in each study. For each treatment and control, 6 replicate glass vessels (30 mL) were filled with 10 g prepared soil 1 day before the test began. For each test treatment and control, three replicate soil samples were analyzed for pH (Jürgens, WTW, Weilheim, Germany) at the end of the reproduction test. To test the accuracy and reproducibility of the experiment, the reproduction test was repeated four times: in the first week of April, the second week of July, the first week of October and the second week of January of the following year.

Statistics
Statistical analyses were performed with SPSS 23.0. Normality of the data was analyzed untransformed according to Shapiro-Wilk test (p > 0.05), albeit a minor deviation of normality in the controls in April and July (p = 0.048 and p = 0.032, respectively). Variance homogeneity was analyzed according to Levene's test (p = 0.164). A general linear model (Two-way ANOVA) was used to analyze the main and interaction effects of treatment and month as influencing factors; pairwise comparison (Bonferroni) was used to show significant differences between single months or treatments, respectively. All tests were run with and without outliers; the outliers did not change the results of significance. Therefore, outliers were included in all final results.

Reproduction in different treatments during four repeats
The mean soil pH CaCl 2 was 5.59 (SD = 0.39) in the control, 5.57 (SD = 0.45) in soil spiked with AgNP and 5.44 in soil spiked with AgNO 3 (SD = 0.46). No significant differences were detected during the four experiments between treatments and control, nor between the single experiments.
All reproduction tests met the validity criteria according to OECD guideline 232. The mean of adults' mortality in control did not exceed 20%, and did not differ between the treatments (p = 0.628). The mean number of juveniles per vessel in control was not lower than 100 for each replicate control (n = 6), and the coefficient of variation of reproduction was less than 30% (Additional file 1: Table S1).
Collembola reproduction significantly differed between repeats. Figure 1 shows the number of F. candida juveniles in the AgNP treatment, AgNO 3 and control in four test repeats. Only the reproduction with AgNO 3 did not differ between repeats (Additional file 1: Table S2 in Appendix). The reproduction in the control was lowest in January and highest in October. In the AgNP treatment, the highest reproduction was detected in October and the lowest in July (Additional file 1: Table S1 in Appendix). The comparison between treatments in single repeats is shown in Fig. 2.

Toxicity of AgNP and AgNO 3 during four repeats
The effect of treatment and month as impact factors is significant, and they have an interaction effect (Additional file 1: Table S4 in the Appendix). Figure 2 displays toxic effects of AgNP and AgNO 3 compared to the control. Neither the retrieved adults nor the adult's size differs significantly between treatments and repeats (p > 0.05, Additional file 1: Table sS5, S6 in Appendix). Compared to the control, the reproduction of F. candida was significantly reduced in April, July and October in the AgNP treatment (p < 0.05), but not in January (p = 0.736), and during all tests in the AgNO 3 treatment (p < 0.05, Fig. 2, Additional file 1: Table S2 in the Appendix). AgNP were (1) less toxic for the reproduction of F. candida than AgNO 3 in October (p < 0.05), (2) but the same as AgNO 3 in toxicity in other months (p < 0.05, Fig. 2, Additional file 1: Table S3 in Appendix).

Discussion
We investigated the toxicity of AgNP and AgNO 3 to F. candida in four repeats during 1 year. For the first time, strong toxic effects of AgNP at a concentration of 30 mg Ag/kg on the reproduction of F. candida were observed. This is in contrast to Waalewijn-Kool et al. [25] who found no effect on survival and reproduction for F. candida exposed to AgNP at a measured concentration of 673 mg Ag/kg dry soil, which was more than 20 times higher than the concentration in our study. Mendes et al. [26] found that NM-300K reduced F. candida reproduction by about 50%, yet at a concentration of 640 mg Ag/ kg soil. Mainly three factors may explain the differences between these studies: (1) Waalewijn-Kool et al. [25] used paraffin-coated AgNP in a water-only dispersion, while NM-300 K are uncoated and dispersed in a suspension that contains three organic agents; (2) The size of AgNP used by Waalewijn-Kool et al. was 3-8 nm AgNP, whereas NM-300K has a diameter of 15 nm.; (3) Loamy sand soil (LUFA-Speyer 2.2, Sp 2121, Germany, 2009) with a pH CaCl 2 of 5.5 was used by [25] and [26], whereas we used RefeSol 01A, a loamy, medium-acidic, and lightly humic sand with pH CaCl 2 of 5.67.
For NM-300K, an effect of the organic dispersion can be excluded, because tests had been made in advance to ensure that the dispersion showed no toxic effect of on the reproduction of F. candida (X. Zhang, unpublished data). McKee et al. studied the dispersion of NM-300K in OECD soil pore water and found that the dispersion caused significant immobilization of F. candida at 10 mg/l, whereas no toxic effect occurred at 40 mg/l [40]. This is in line with our findings. Second (except for differences in release kinetics, see Engelke et al.), particle size can also be excluded as nanoparticle reactivity increases with decreasing size [41]. Therefore, coating and soil type might be the main reasons for the fate and toxicity of the particles found in our study. The presence of a coating is important because it can modify the particle structure, the electrostatic surface charge and, therefore, its potential toxicity over time  [42]. Nguyen et al. [43], for instance, found considerable differences in toxicity between AgNP coated with citrate and polyvinylpyrrolidone and uncoated AgNP to macrophages and epithelial cells. They reported that uncoated AgNP, at a concentration of 1 µg/ml, decreased cell viability by 20-40% and that 20-and 40-nm particles were 10% more cytotoxic than the 60-and 80-nm particles. In exposures to coated AgNPs, cell viability dropped at 25 µg Ag/ml or higher concentrations. Similar coating effects were observed with ZnO-NPs and F. candida [44] and with iron oxide nanoparticles and mouse fibroblast cells [45]. There is strong support for the assumption that the different soil types were the main reason for the large difference in toxicity between our study and the one by Mendes et al. [26], as various studies in our lab with Collembola (McKee et al. [40]) and enchytraeids (Voua Otomo et al. under revision) have rendered much stronger toxic effects of AgNP in RefeSol 01A than in Lufa 2.2 and artificial OECD soil. In fact, it is challenging to evaluate the results of different studies on F. candida if different clonal lineages were used in the different studies. A large number of studies have shown that different lineages of F. candida can exhibit very different life history traits, and that they also differ genetically in their ability to cope with environmental stress [39,[46][47][48].
In the present study, the toxicity of AgNP varied significantly between repeats. Significant toxic effects of AgNP on reproduction were observed in April, July and October, but not in January, while AgNO 3 caused toxic effects during all repeats. The reproduction of F. candida in the control in January was 19.5-35.9% lower than in July and October. On the other hand, in the soil spiked with AgNP, the reproduction of F. candida in January was even higher than in July. In the following, we discuss four possible explanations: fungi compromising Collembola reproduction by defensive strategies or being entomopathogenic, differences in dissolution kinetics of AgNP and AgNO 3 , avoidance behavior and circannual biological rhythms.
Fungi observed in our test vessels during winter might account for the reduction of the reproduction of F. candida in January. It is possible that spores from fungi emitted in autumn or winter are brought into the laboratory by the ventilation system. We speculate that these fungi might have defence properties (toxins or crystals at the hyphal surface) that inhibited reproduction of F.candida [49,50]. An alternative explanation could be Entomopathogenic Fungi (EPF), although not all of them are toxic to Collembola [51,52]. Outbreaks of infection with entomopathogens such as Entomophthora muscae tend to occur in spring and autumn, and sporulation usually takes place in cool, humid conditions [53]; it might explain why we observed fungi in our test vessels in January and the reproduction of F. candida was significantly decreased in control. Interestingly, there was no significant reduction in reproduction in January in the soil treated with AgNP (Fig. 1), most likely due to their continuous antimicrobial activity [54,55]. On the one hand, AgNP are capable of inhibiting fungi that compromise the reproduction of F. candida; on the other hand, the direct negative effects of silver on the Collembola would partly be masked by the indirect positive effect through its suppressing effect on such fungi.
In January, the reproduction was significantly lower in the treatment of AgNO 3 , but there was no difference in the treatment of AgNP compared to control (Fig. 2). We postulate that the different performance of both Ag forms is due to their reaction kinetics. AgNO 3 dissociates readily in water, but only part of the Ag + ions are bioavailable: they will react with anions in the soil solution, forming insoluble precipitates, or complexes with organic acids. In turn, AgNP dissolve slowly, constantly releasing new Ag + ions. Therefore, over a longer period, it is likely that more Ag + is bioavailable from AgNP than from AgNO 3 .
But what are the reasons for those differences between repeats? The hypothetical model in Fig. 3 illustrates why the treatment with AgNO 3 had a negative effect on F. candida in January, not the one with AgNP: The presumed contamination with fungal spores should have been present in low numbers at the beginning of January, then might increase due to favorable conditions and decrease again in spring due to increasing temperature. The release of dissolved Ag + upon adding AgNP to moist soil provides a low, but constant supply of Ag + ions. The low Ag + concentration should be sufficient to control the small initial fungi population in winter and to prevent their further increase, thus reducing the negative effect of fungi on the reproduction of F. candida. With AgNO 3 , the sudden release of dissolved Ag + upon adding AgNO 3 to moist soil would kill most of the present fungi, but the population might quickly recover thereafter (Fig. 3).
Some studies explained the difference in toxicity between AgNP and AgNO 3 by a release of Ag + from the particles and by a slower assimilation of AgNP, which leads to lower toxic effects on soil fauna compared with AgNO 3 [56][57][58]. Such differences in toxicity were also reported in studies with earthworms [21,22]. Similar results were observed in our study during autumn and winter. Stronger toxic effects were found in the treatment with AgNO 3 than that with AgNP, which supports the ion release theory. However, the pattern was reversed in the repeats in April and July. We believe this is a combination of Ag + release kinetics (see above) and avoidance behavior. Avoidance studies in our laboratory gave hints that F. candida and enchytraeids avoid high, but not low concentration of Ag. Assuming that they sense rather the ions than the undissolved metal, it is possible that they actively avoided (e.g., by staying mostly at the uncontaminated food patch on the surface) only the AgNO 3 treatment but not the one with AgNP in our study. Thus, in the AgNP treatment, the animals were exposed to low concentrations of Ag + permanently released by the AgNP, reducing their reproduction.
Circannual biological rhythms might be another explanation for the different toxicity results. Rozen collected earthworms (Dendrobaena octaedra) from the field and cultured them in the laboratory under constant conditions. The author found that reproduction was highest in spring and summer, and dropped significantly in the winter months, which indicated that internal regulation of reproduction may exist in the earthworm D. octaedra [59]. However, the mechanisms have not yet been understood. Nevertheless, we cannot fully elucidate what exactly caused the toxicity of AgNP in the present study. The hypothesis on the interaction with fungi should be tested in further investigations, to identify the fungi species present during winter.
Krogh summarized data of reproduction tests using F. candida and Folsomia fimetaria from 1994 to 1999 and found that the variability of reproduction in the control is obvious and many factors contribute to the variability [60]. To what extent should we trust these data in the face of different experimental results? By comparing the data from our four repeated experiments, we suggest that F. candida in each laboratory should have a database of average reproduction rate in the control, and this database should also contain information on metadata such as different test soils or strains. To increase the reliability of the experimental results beyond established validity criteria, we suggest that the test results should be disregarded when the reproduction rate in control is significantly different from the average reproduction rate established in the laboratory.

Conclusions
We demonstrated for the first time that AgNP in natural soil can have strong toxic effects on the reproduction of F. candida at a concentration of only 30 mg Ag/kg, which is about 20 times lower than reported earlier. These effects can mainly be attributed to soil conditions (compared to Lufa 2.2 and artificial OECD soil). This is the first article reporting significantly different results of repeated toxicological experiments using F. candida. Although no clear explanations for the different performances during four repeats were found, a data bank of average reproduction rate of F. candida and other species is recommended to give comprehensive results in further toxicological tests. To corroborate our hypothetical model on the different outcome in January and April, studies specifically addressing ion release kinetics of AgNP and fungi identification are needed. Furthermore, more studies of F. candida's avoidance behavior should be taken into account as well. Fig. 3 Hypothetical model on the development of deleterious fungi in winter and spring in the different treatments as affected by the released Ag + ions. There is a slow and continuous ion release from AgNP, whereas AgNO 3 ions dissolve at test start. Numbers indicate different phases on fungi populations in the single treatments and tests: (1) Efficient control of the originally small population by continuous Ag + ion release; (2) High mortality and exponential recovery due to high growth rate during winter