4. Discussion
In the present study, gold nanospheres and nanostars that have shown to be promising for biomedical applications [
1,
5] were synthetized and characterized. The characterization data agreed well with the literature [
20,
22,
29]. Star-shaped AuNPs show better optical properties than nanospheres of similar diameters [
1,
5], but the question remains whether these improved optical properties also implicate significant differences in the way star-shaped nanoparticles interact with biological systems. Even if AuNPs of approximately 50 nm diameter were described as noncytotoxic in phagocytic and nonphagocytic hepatic cells, a shape-dependent toxic effect was described in other cells [
10,
30]. In fact, several in vitro studies have shown differences in toxicity and uptake, but the conclusions were contradictory [
7,
9]. For example, Favi et al. reports that 61.46 nm gold nanospheres are more toxic than 33.69 nm gold nanostars on rat fat endothelial cells, while in Sultana et al. study, 50 nm gold nanostars are more cytotoxic than 50 nm gold nanospheres to human endothelial cells [
7,
9]. An additional concern with these in vitro studies is the lack of in vivo complexity and specially the relevance of the low but biologically relevant in vivo doses [
8,
31,
32]. In addition, the use of gold nanoparticles with similar physicochemical properties is very important in comparison studies, as the simultaneous variation of different properties (e.g., size, shape, or coating) can easily confound data interpretation [
7,
8,
31]. In the present study, in order to facilitate interpretation, we used appropriate synthesis methods [
20,
22] to guarantee a similar diameter, and used the same coating agent.
A likely confounding factor that explains the contradictory results often found in comparative studies is the way the dose is calculated and expressed in the experimental study design. The vast majority of studies use Au mass per animal weight as dose, which represents the gold mass concentration as determined by GFAAS or inductively coupled plasma mass spectrometry (ICP-MS) [
11,
33,
34]. But, as previously demonstrated, the same Au mass concentration does not necessarily translate into the same concentration of gold nanoparticles, leading to inaccurate comparative studies [
10,
35,
36]. This is especially relevant when comparing nanoparticles of different sizes, for which Au mass concentration versus AuNPs concentration varies the most [
10,
36]. In the present study, we were able to synthesize nanospheres and nanostars of very similar sizes (~40 nm versus ~47 nm, respectively), but still decided to calculate the administrated dose in nanoparticle concentration to avoid misleading results. As expected, the same concentration administrated in AuNPs (1.33 × 10
11 AuNPs/kg) did not correspond to the same concentration in Au mass (146 µg/kg for nanospheres and 120 µg/kg for nanostars). We therefore believe that, with this study design, we can accurately compare two types of nanoparticles, at the same tested dose, and with all relevant physical chemical properties kept constant except for the shape.
The biodistribution of the tested nanoparticles was shape-dependent, and the liver and the spleen were, as expected, the main site for accumulation for both types of AuNPs [
12,
34]. A higher gold content was found in HepG2 cells exposed to nanospheres as compared to nanostars [
37], which agrees well with our in vivo data. The uptake of AuNPs seems to be, not only shape-dependent, but also dependent on the cell/organ type, as distinct biodistribution profiles were determined for the liver, spleen, and lung. In vitro studies have previously shown that, at same tested Au concentration, gold nanospheres are internalized to a higher extent than nanostars in particular cell lines [
32,
37], while for other cell types the opposite occurs [
9,
31]. A content in Au lower than the LOQ found in the other organs, tissues and fluids was probably due to the lower dose of injected AuNPs compared with other in vivo studies [
11,
12]. Nevertheless, a lower content in Au for these organs, tissues and fluids compared with liver, spleen or lung was previously reported for similar experimental conditions [
11,
12]. The lack of Au content in brain was expected since in a previously published paper of our group, ~50 nm MUA-coated nanospheres and nanostars were unable to cross the in vitro hCMEC/D3 human blood brain barrier model [
10].
At the tested Au concentration, the animals did not show any behavioral change upon exposure to various treatment groups versus control. Nevertheless, severe changes in behavior, hypopnea, tremor and arching of the back and even death were already reported in animals treated with other types of AuNPs, such as rods and nanocages at high doses, and with 40 nm gold nanospheres. For the 40 nm gold nanospheres, the effects were scarcer than for gold nanorods, suggesting a shape-dependent effect, and it was only observed at the highest dose of ~10
12 AuNPs/g animal [
38]. So, the effect was also concentration-dependent and the lack of any behavioral change from our study can be easily explained by the lower concentration in AuNPs administrated to the rats. In fact, the concentration of AuNPs used in our study was lower than in many vivo studies [
11,
12,
38], so no significant differences are expected in the conventional toxicological assays. As metabolomics showed the ability to detect modifications in various biological pathways even at low subtoxic concentration of AuNPs [
39], further discussion will mainly focus on this approach.
Variables associated to metabolic data due to collection, storage and preparation of the samples either by incorrect data processing or equipment operation can affect the quality of the analysis [
40]. In order to eliminate this variability, quality control samples (QCs) corresponding to a pool of all samples included in the study were used [
40]. The relative standard deviation (RSD) of each variable across the QCs was used as an indicator of the reproducibility and repeatability of the analysis [
40]. The unsupervised multivariate analysis method, represented by a principal component analysis (
Figure 2) showed that QCs cluster in the score plot separates from the treated group, suggesting a high data quality.
For the unsupervised multivariate analysis of the metabolomic fingerprinting, PCA is the most widely used method [
25,
41]. Normally, is the first line analysis to formulate initial biological hypothesis that are further confirmed using a supervised method, such as orthogonal partial least squares discriminant analysis (OPLS-DA) [
25,
42]. An OPLS-DA classification guided by well-separated PCA scores has a greater likelihood of producing biologically relevant results. In the current study, PCA showed a tendency to separate among all groups. OPLS-DA analysis, recommended for two class separation, was further applied to maximize this separation tendency and to obtain a clearer and straightforward interpretation [
42]. By using OPLS-DA analysis, the variation within group (for example animal variation) is usually eliminated, resulting in much better scores in which classes are separated. One of the challenges of using OPLS-DA analysis is the possibility of presenting class separation even for a completely random variable [
25], so further validation is necessary. The current study obtained good class separation for the liver analysis (
Figure 3) of Au nanospheres versus Au nanostars. In order to validate the results obtained by OPLS-DA in the liver, among many cross-validation parameters (
Table 3), the quality assessment statistic (Q
2) that indicates how well the model predicts new data is often recommended [
40]. Values higher than 0.4 obtained for the liver samples are considered acceptable [
25]. Another validation was made using CV-ANOVA analysis, and the significant results for the liver with
p < 0.05 reinforce the quality of the model [
41] (
Table 3). The sum of squares of uncorrelated (R
2X) and predictive (R
2Y) components representing the variation explained by the model are also used for validation [
25,
40]. High values of R
2 and Q
2 were described as good indicators for the power of models in selecting significant data [
25,
40]. As suggested, OPLS-DA can lead to a good separation even when data are randomly distributed, as the permutation tests were used to randomly assign class labels to different individuals [
43]. In order to evaluate whether the specific classification of the individuals in the two designed groups (OPLS-DA) is significantly better than any other random classification in two arbitrary groups, the classification models of the permutation test are further calculated and the obtained values (R
2 and Q
2) should be lower than the initial OPLS-DA analysis [
40,
43]. The performed permutation tests for the liver show lower values than in the original OPLS-DA analysis, as required for validation (
Table 3).
Finally, discriminant metabolites obtained from the GC-MS analysis should be associated to biochemical processes to get the big picture of all metabolic changes that characterize a specific treatment [
40]. Many metabolites can be common to specific biological pathways and software such as the MetaboAnalyst can be used to integrate all metabolites in specific biochemical pathways [
17]. In fact, biological pathways as the metabolism of glutathione, ascorbate, fatty acids, aminoacids, and purine and pyrimidine, were found altered in both nanospheres- and nanostars-treated animals compared to control and were considered also important in the comparison between gold nanostars and nanospheres (
Figure 6). These metabolic pathways were previously related to the toxicological effect of AuNPs [
44].
Some of the discriminant metabolites, such as pyrogallol and dimethylglycine, were associated to the production of reactive oxygen species (ROS). Pyrogallol can produce oxidative damage and subsequently mutagenesis, carcinogenesis, and hepatotoxicity [
45] while dimethylglycine, an intermediate of choline oxidation, is considered an effective ROS generator (of O
2•− and hydrogen peroxide in liver mitochondria, via electron transfer to Complex 1), which can further lead to lipid peroxidation and cell injury [
46]. The involvement of AuNPs in ROS production and subsequent oxidative stress has also been previously reported [
44,
47]. The cellular responses to ROS attack can manifest through (i) increasing de novo synthesis of GSH and/or (ii) GSH oxidation. In this study, no significant increase was noticed for glutamine, glycine and cysteine, precursors of GSH, for none of the tested conditions. In accordance, ATP levels were not altered by the biochemical assays, contrary to what would be expected in de novo synthesis of GSH, that requires high energy levels [
48,
49]. On the other hand, GSH oxidation normally occurs in the presence of selenium-dependent GSH peroxidase, with the production of GSSG which is further reduced back to GSH by GSSG reductase in the presence of NADPH, together with enzymes as glutathione reductase and glutathione peroxidase [
50]. Therefore, the results suggest that the responsible for the increase of GSH, as in the current study, is not de novo synthesis of GSH, but most probably the GSSG reduction.
Among the discriminant metabolites, a large group belongs to saturated (oleic, palmitic, myristic acid) and unsaturated (linoleic, palmitoleic, docosahexaenoic, arachidonic, and 5,8,11-eicosatrienoic) fatty acids. Their levels were found increased in both treatments compared with the control group and this was associated with disturbances in their biosynthesis (
Figure 7). Palmitic acid represents the first product of the cytosolic fatty acids’ biosynthesis, in the presence of fatty acid synthases and NADPH [
51]. It can undergo oxidation into palmitoleic acid, and elongation into stearic acid which can be further oxidized to oleic acid, considered an end-product of the de novo synthesis of fatty acid [
51]. 5,8,11-Eicosatrienoic acid can be synthesized from oleic acid [
52], while docosahexaenoic acid, important for brain development, from α-linoleic acid [
53]. Their synthesis requires various enzymes for desaturations, elongations and beta-oxidations [
53] and was related to the production of pro-inflammatory prostaglandins, cytokines, and reactive oxygen species, thus playing an important role in inflammatory diseases [
54]. In fact, an important mediator of inflammatory processes, arachidonic acid, was found slightly increased in the liver of nanosphere-treated animals. On the other hand, the increase was not significant in the nanostars-treated group. Its role in the inflammatory process [
55], through enzymatic oxidation with release of prostaglandins and pro-inflammatory factors was well described [
56].
Free fatty acids are precursors of various biomolecules and they are constituents of triglycerides and glycerophospholipids [
57]. Therefore their increased levels, observed in the current study, can also be a consequence of degradation or hydrolysis of these complex compounds [
58]. The degradation of glycerolipids occurs in the presence of lipases, while the phospholipase A1, A2 and B are responsible for the breakdown of glycerophospholipids [
58]. This can explain the high levels of linoleic acid, an essential fatty acid, that cannot be synthesized in animals as they lack the necessary enzyme for desaturation of oleic into linoleic acid [
59]. Other products of triglycerides degradation, such as glycerol (nanostars treatment) and glyceric acid (both treatments), were also found increased [
58]. In the nanostars-treated group, other metabolites such as serine, ethanolamine and choline, all constituents of glycerophospholipids, were also found slightly increased. As the glycerophospholipids are the main component of biological membranes, their degradation was previously related to altered cell functions, such as mitochondrial stress, apoptosis and damage of the cell membrane integrity [
58]. In accordance, another metabolite, squalene, a precursor of cholesterol (essential component of cellular membrane) was found decreased in the nanostars group.
Other metabolites that discriminate between AuNPs treatments and control were uridine, uracil and inosine which are associated to the purine and pyrimidine metabolism. Uracil was found increased while its precursor, uridine, was found decreased for both AuNPs. A severe decrease was also found for inosine in both treatments. These metabolites are major energy carriers [
60], crucial for the synthesis of nucleotides and for the interconversion of nucleobases, nucleosides and nucleotides [
61]. Disorders of the purine and pyrimidine metabolism are associated to cellular damage [
62,
63] suggesting a damaging effect of both nanospheres and nanostars in the liver.
Metabolites such as L-proline, serine, L-threonine and branched chain amino acid L-isoleucine were found slightly increased, with L-proline reaching significance (
p value < 0.05) in the liver of nanostars-treated animals after 24 h administration versus control. It was described that rats under stress conditions, have an accelerated secretion of stress hormones which mobilize amino acids from proteins [
64]. The elevated levels of the amino acids observed in this study can thus be explained by cellular responses to stress. Even more, AuNPs with anisotropic shapes were previously associated with oxidative stress and time-dependent increase in amino acids levels on tumorous A549 cells and normal 16HBE cells [
65].
Overall, both nanosphere- and nanostars-treatments altered biological pathways as glutathione metabolism, protein metabolism, fatty acid metabolism, and purine and pyrimidine metabolism suggesting a similar biological response. Nevertheless, some pathways are more affected by nanostars treatment while others are more affected by nanosphere-treatment. For example, as seen in
Table 4, the upregulation of several fatty acids’ synthesis (palmitic, oleic, linoleic, 5.8,11 eicosatrienoic, docoxahexaenoic) was significantly higher for nanosphere-treated animals, while increased aminoacids levels such as, L-proline, serine, L-threonine, ethanolamine and choline were found only for the nanostars-treated animals. Differences among the two treatments were also found regarding the intensity of upregulation of uracil levels and decrease in uridine and inosine belonging to purine and pyrimidine metabolism. Differences appeared also in L-lysine, an amino acid related to the lysyl-tRNA synthetases. These enzymes belong to the class of aminoacyl-tRNA synthetase necessary for charging tRNA with their associated amino acids (aminoacyl-tRNA) [
62]. Therefore, even if the involved biological pathways seemed to be common to both treatments, response intensity was different, with nanospheres producing a more detrimental effect on the fatty acid synthesis while nanostars affecting more the protein synthesis.
Distinct metabolomic profiling among different types of AuNPs was already reported in the literature. For example, in an in vitro metabolomic study, HepG2 cells were exposed for 3 h to 18 nm gold nanospheres with different capping agents (citrate, poly-(sodiumsterene sulfunate) and poly-vinylpyrrolidone) [
14]. Although there was a general surface-dependent decrease in intracellular metabolites for all the nanospheres, the effect was more intense for the PVP-coated AuNPs. On the other hand, citrate- and poly-(sodiumsterene sulfunate) AuNPs influenced also the ATP production in addition to the metabolomic changes. Most of the metabolites belong to the amino acid metabolism. As in our study, lipid metabolism, acyl-carnitine metabolism, carbohydrate and energy metabolism, were also pathways found altered. Nevertheless, contrary to our results, their study showed a holistic depletion of most metabolites that were attributed to the ability of AuNPs to adsorb them on their surface. As the adsorption process is related to the capping of AuNPs, the use of different capping agents in their study (citrate, poly-(sodiumsterene sulfunate) and poly-vinylpyrrolidone) versus our study (MUA) could explain the differences. Another possible explanation for the general decrease in metabolites versus general increase in our study is the use of AuNPs with different sizes (18 nm versus 40 nm) which is known to be another factor that influence biological response. Another multiomics comparison approach, including proteomics and metabolomics, was applied to the study of 5 and 30 nm gold nanospheres at 300 µM after 72 h incubation on Caco-2 cells [
13]. Once again, differences were found in the metabolomic and proteomic profile between the two tested nanoparticles further associated with higher uptake of smaller nanospheres. The biological pathways were connected to amino acid and protein metabolism, glutathione metabolism, fatty acids metabolism and energy. Some of the metabolites were found increased, mainly related to the TCA cycle and energy and the vast majority, including proteins involved in DNA synthesis and repair, the synthesis of protein and the amino acid transport were found dysregulated. For 5 nm AuNPs, the biological pathways were related mainly to small molecule biochemistry, cell assembly and organization, cellular growth and proliferation, while for the 50 nm counterparts, the most affected pathways are involved in cellular compromise (degeneration) and cell morphology. Other metabolomic in vitro study using TM-4 cells exposed for 24 h to gold nanorods (10 nm width, 40 nm length) showed alteration of pathways similar to what was found in our study: the glycine, serine and threonine metabolism, cyanoamino acid metabolism, methane metabolism, aminoacyl-tRNA biosynthesis and nitrogen metabolism [
66]. This corroborates the idea of a general panel of biological pathways involved in the toxicity of AuNPs. Similar pathways, as creatine, glycine, alanine, phosphocoline, UDP-NAG, pyruvate, succinate, lactate, methionine and glutathione dysregulation were found even at nontoxic concentration of AuNPs [
39].