Zinc oxide nanoparticles were synthesized by Beijing DK Nano Technology Co. Ltd. (Beijing, China) as reported previously (Liu et al., 2016, 2017; Zhao et al., 2016a,b; Ge et al., 2017). The characteristics of ZnO NPs (morphology, size, agglomeration, etc.) were determined by transmission electron microscopy (TEM; JEM-2100F, JEOL Inc., Japan) and dynamic light scattering (DLS) particle size analyzer (Nano-Zetasizer-HT, Malvern Instruments, Malvern, United Kingdom).

This investigation was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocol (protocol number: QAU20161142) was approved by the Committee on the Ethics of Animal Experiments of Qingdao Agricultural University Institutional Animal Care and Use Committee (IACUC) (Zhao et al., 2016a,b; Liu et al., 2017). All hens (Jinghong-1 strain) were housed in a ventilated and conventional caged commercial poultry house with a lighting program of 16:8 light/dark and ad lib food and water. The formulation of the basal diet (corn–soybean base) has been previously reported (Supplementary Table S1) (Zhao et al., 2016a,b). There were seven treatments: (Taylor et al., 2010) Control treatment (no Zn added); (Ryan and Robards, 2006) ZnSO₄-25 mg/kg; (Gioria et al., 2016) ZnSO₄-50 mg/kg; (Garcia-Contreras et al., 2015) ZnSO₄-100 mg/kg; (Bundy et al., 2009) ZnO-NP-25 mg/kg; (Boyles et al., 2016) ZnO-NP-50 mg/kg; and (Carrola et al., 2016) ZnO-NP-100 mg/kg. The concentrations of ZnO NPs or ZnSO₄ used in our studies were based on the diet. If the concentration of 100 mg/kg of diet was calculated based on animal body weight (BW), it was calculated to be around 10 mg/kg BW. Therefore, the current concentrations were lower than those used in other studies (100–1000 mg/kg BW) (Yan et al., 2012; Hong et al., 2014a,b). A total of 420 hens were randomly assigned into the seven treatments, with three replicates per treatment and 20 hens per replicate. Experimental feeding started at 6 weeks (wks) of age. After 4 or 24 wks of exposure, 12 hens from each treatment were humanely slaughtered and blood (plasma) and tissue/organ samples were collected and stored at -80°C.

Sample preparation procedures for detecting NPs have been reported in our recent publication (Zhao et al., 2016a,b; Ge et al., 2017; Liu et al., 2017). Briefly, tissue samples were collected and fixed for 2 h in 2% glutaraldehyde made in sodium phosphate buffer (pH 7.2). Specimens were then washed extensively to remove the excess fixative and subsequently post-fixed in 1% OsO₄ for 1 h in the dark. Specimens were then dehydrated in an increasingly graded series of ethanol and infiltrated with increased concentrations of Spur’s embedding medium in propylene epoxide. Subsequently, the specimens were polymerized in embedding medium for 12 h at 37°C, 12 h at 45°C, and 48 h at 60°C. Fifty nanometer sections were cut on a Leica Ultracut E microtome equipped with a diamond knife (Diatome, Hatfield, PA, United States), and collected on form var-coated, carbon-stabilized Mo grids. The section containing grids were stained with uranyl acetate, air dried overnight, and imaged on a JEM-2010F TEM (JEOL Ltd., Japan). The presence of ZnO NPs in the tissues was confirmed by using X-Max^N 80 TLE EDS (Oxford Instruments, United Kingdom).

Nuclear magnetic resonance (NMR) analyses were performed as previously described with slight modifications (Soininen et al., 2009; Lee et al., 2016). Before the NMR spectroscopy, 200 μl plasma was mixed with 80 μl D₂O solution containing sodium phosphate buffer (0.1 M, pH 7.4) and sodium 3-trimethylsilyl-2,2, 3,3-d4-propionate (TSP) as an internal standard (δ = 0 ppm). The ¹H NMR spectra was acquired using a 600.13 MHz Bruker AV600 spectrometer (Bruker, Rheinstetten, Germany) with a 5-mm CryoProbe at 300 K. NOESY and a zg pulse sequence of ¹H NMR spectra and zggpr pulse sequence of J-resolved (JRES) NMR spectra were used to acquire the NMR information. In plasma NMR analyses, molecular identification and quantification are hampered by the complexity of plasma samples that contain a wide variety of molecules. To resolve this issue, we adopted an approach based on two molecular models, the LMWM model and the LIPO model, as previously described (Soininen et al., 2009). The LIPO model provides information on lipoprotein lipids and subclasses which are acquired through water-suppressed ¹H NMR spectrum of serum. Alternately, the LMWM model is acquired by suppression of most of the broad macromolecules and lipoprotein lipid signals; this improves the sensitivity of low-molecular-weight metabolites (Mäkinen et al., 2008; Soininen et al., 2009; Yan et al., 2012; Wan et al., 2016). The LIPO window showing broad overlapping ¹H NMR resonances coming mainly from lipid molecules in various lipoprotein particles were recorded with 80 k data points after four dummy scans using eight transients acquired with an automatically calibrated 90° pulse and applying a Bruker NOESY presat pulse sequence with mixing time of 10 ms and irradiation field of 25 Hz to suppress the water peak. The acquisition time was 2.7 s and the relaxation delay 3.0 s. The 90° pulse was calibrated automatically for each sample. A constant receiver gain setting was applied for all the samples. The LMWM data were acquired using a T₂-relaxation-filtered pulse sequence which suppressed most of the broad macromolecule and lipoprotein lipid signals and enhanced the detection smaller molecules. The LMWM data were recorded with 64 k data points using 24 (or 16) transients acquired after four steady-state scans with a Bruker 1D CPMG pulse sequence with water peak suppression and a 78 ms T₂-filter with a fixed echo delay of 403 ms to minimize diffusion and J-modulation effects. The acquisition time was 3.3 s and the relaxation delay 3.0 s. Both LIPO and LMWM data were processed and phase corrected in an automated fashion. Prior to Fourier transformations to spectra, the measured free induction decays for both LIPO and LMWM windows were zero-filled to 128 k data points and then multiplied with an exponential window function with a 1.0 Hz line broadening.

The ¹H NMR spectra were processed by MestRe-C2.3 software (Yan et al., 2012). The spectra were binned with a unit of 0.005 ppm between 0.2 and 10.0 ppm, and then integrated spectral intensity for each bin. The regions of internal standard and water resonance were excluded before being normalized by the total spectral area. The binned data were adjusted by generalized log transformation and mean-centered before multivariate analysis.

The processed NMR datasets were examined by using principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA) by SIMCA-P10.0 software package (Version 10, Umetrics AB, and Umea, Sweden). PCA was used to reduce the complexity of the metabolomics data matrix without additional information and provides the visual performance of the original cluster for each group. PLS-DA connected the classified information and NMR dataset to determine the variance between the different treatment groups. Two-dimensional score plots were used to visualize the separation of the samples and the corresponding loading plots were applied to identify the spectral variable contribution to the position of the spectra that were altered by different treatments (Yan et al., 2012).

Total RNA was isolated as described previously (Zhao et al., 2016a). RNA concentration was determined by Nanodrop 3300 (ThermoScientific, Wilmington, DE, United States). Two micrograms of total RNA was used to make the first-strand complementary DNA (cDNA; in 20 μl) using RT2 First Strand Kit (Cat. No: AT311-03, Transgen Biotech, China) following the manufacturer’s instructions. The generated first-strand cDNAs (20 μl) was diluted to 150 μl with double-deionized water (ddH₂O). Then, 1 μl was used for one PCR reaction (in a 96-well plate). Each PCR reaction (12 μl) contained 6 μl of qPCR Master Mix (Roche, German), 1 μl of diluted first-stand cDNA, 0.6 μl primers (10 mM), and 4.4 μl of ddH₂O. The primers for qPCR analysis were synthesized by Invitrogen and present in Supplementary Table S2. The qPCR was conducted by the Roche LightCycler^® 480 (Roche, German) with the following program – step 1: 95°C, 10 min; step 2: 40 cycles of 95°C, 15 s; 60°C, 1 min; step 3: dissociation curve, step 4: cool down. Three or more independent experiment samples were analyzed (Zhao et al., 2016a,b).

Liver samples were lysed in RIPA buffer containing a protease inhibitor cocktail from Sangon Biotech, Ltd. (Shanghai, China). Protein concentration was determined using a BCA kit (Beyotime Institute of Biotechnology, Shanghai, China) (Liu et al., 2017). The information for the primary antibodies (Abs) is present in Supplementary Table S2. GAPDH and Actin were used as loading controls. Secondary donkey anti-goat Ab (Cat no. A0181) was purchased from Beyotime Institute of Biotechnology, and goat anti-rabbit (Cat no.: A24531) Abs were bought from Novex^® by Life Technologies (United States). Fifty micrograms of total protein per sample was loaded onto 10% SDS polyacrylamide electrophoresis gels. The gels were transferred to a polyvinylidene fluoride (PVDF) membrane at 300 mA for 2.5 h at 4°C. Subsequently, the membranes were blocked with 5% bovine serum albumin (BSA) for 1 h at room temperature (RT), followed by three washes with 0.1% Tween-20 in TBS (TBST). The membranes were incubated with primary Abs (Supplementary Table S3) diluted at 1:500 in TBST with 1% BSA overnight at 4°C. After three washes with TBST, the blots were incubated with the HRP-labeled secondary goat anti-rabbit or donkey anti-goat Ab, respectively, for 1 h at RT. After three washes, the blots were imaged. The images were quantified by Image J.

The qRT-PCR data were statistically analyzed based on ΔΔC_t using proprietary software from SABiosciences online support¹. Other data were statistically analyzed with SPSS statistics software (IBM Co., New York, NY, United States) using ANOVA. Comparisons between groups were tested by one-way ANOVA analysis and LSD tests. All groups were compared with each other for every parameter (mean ± SD). Differences were considered significant at p < 0.05.

The ultra-structure of ZnO NPs used in this investigation has been published in our previous articles (Supplementary Figure S1) (Zhao et al., 2016a,b; Ge et al., 2017; Liu et al., 2017). The particles (∼30 nm) were almost spherical with a milk-white color, a surface area of 50 m²/g, and a density of 5.606 g/cm³. Intact NPs were identified in the liver (Supplementary Figure S1) by TEM and confirmed by energy dispersive spectroscopy (EDS) with Zn (Supplementary Figure S1). Three standard Zn peaks were noted.

The LMWM ¹H NMR spectra, displaying metabolic fingerprints of small molecules from plasma metabolites of ZnO NP or ZnSO₄-treated animals, are presented in Figure 1A; while the LIPO ¹H NMR spectra, displaying metabolic fingerprints of large molecules from plasma metabolites of ZnO NPs or ZnSO₄-treated animals, are presented in Figure 2A. Peaks were assigned to specific metabolites based on chemical shift and peak multiplicity according to previous literature (Mäkinen et al., 2008; Soininen et al., 2009; Yan et al., 2012; Wan et al., 2016). ZnSO₄ was used in this investigation to compare the effects of either Zn²⁺ or intact particles because ZnSO₄ produces a sole Zn²⁺ effect (Supplementary Figure S2).

Partial least squares discriminant analysis was used to uncover latent biochemical information from the ¹H NMR spectra. In the LMWM model for small molecules, the score plots showed a clear separation between ZnO NPs and ZnSO₄ at 25, 50, and 100 mg/kg after 4 wks of exposure (Figures 1B,C). Furthermore, there was a clear separation between ZnO NP treatment and the control group, but not between the ZnSO₄ treatment and the control. After 24 wks of exposure, there was no separation between ZnO NPs, ZnSO₄, or the control (Figures 1B,C). In the LIPO model for large molecules, the score plots showed an unclear separation between ZnO NPs, ZnSO₄, and the control after 4 or 24 wks of exposure (Figures 2B,C).

A number of perturbations in endogenous metabolites were observed in the ¹H NMR spectra of plasma samples in both the LMWM and LIPO models. Figure 3 presents prominent small molecule changes in the LMWM model analysis between ZnO NP exposure and the control, or between ZnSO₄ exposure and the control. Compared to the control, 23 metabolites were altered (Figure 3A); of these, 12 amino acids: histidine, valine, isoleucine, tyrosine, glycine, alanine, arginine (lysine), methionine, phenylalanine, threonine, and glutamine were altered (Figure 3A). The remaining 11 metabolites included fumarate, citrate, succinate, 3-hydroxybutyrate, acetate, pyruvate, formate, dimethylglycine, α-glucose, and β-glucose (Figure 3A). After 4 wks of exposure, when compared to the control, ZnO NPs produced more profound metabolite changes than ZnSO₄ in a dose-dependent manner. Around half of the changed metabolites were reduced by ZnO NPs or ZnSO₄ treatments. After 24 wks of treatment, less significant changes were found for both ZnO NPs and ZnSO₄ exposure (Figure 3A). However, most of the changed metabolites were elevated compared to the control by ZnO NPs or ZnSO₄.

We also aimed to determine the effects of ZnO NPs on metabolites coming from Zn²⁺ or intact particles. Therefore, changes in metabolites were compared between ZnO NPs and ZnSO₄ at different concentrations and different time points. Nineteen metabolites were differentially altered by ZnO NPs compared to ZnSO₄ (Figure 3B). After 4 wks of exposure, about half of them showed a decrease and the other half were increased. The three most changed metabolites were isoleucine, tyrosine, and valine, which were dramatically elevated by ZnO NPs. After 24 wks of exposure, most metabolites were reduced by ZnO NPs (Figure 3B); the most changed metabolites were glycine, dimethylglycine, citrate, and glucose (all decreased). The data here suggested that ZnO NPs were different from ZnSO₄ in that the intact NPs might play an important role in altering the levels of small molecule metabolites.

Figure 4A shows the prominent large molecule changes in the LIPO model analysis between ZnO NP exposure and the control or between ZnSO₄ exposure and the control. Compared to the control, six kinds of lipid molecules (Lipid-CH₂CH=, Lipid-CH₂CH₂CO, Lipid-CH=CH-, Lipid=CHCH₂CH=, HDL LDL VLDL(-CH₂-), and -N(CH₃)₃] and one kind of protein (N-acetyl-glycoprotein) were changed by ZnO NPs or ZnSO₄ exposure (Figure 4A). After 4 wks of exposure, ZnSO₄ produced more profound effects on these large metabolites than ZnO NPs. However, after 24 wks of exposure, ZnO NPs produced more profound effects on these large metabolites than ZnSO₄. Compared to ZnSO₄, ZnO NPs differentially altered the levels of these large metabolites at the two experimental time points (Figure 4B). After 4 wks of exposure, ZnO NPs increased Lipid–CH=CH–, –N(CH₃)₃, and N-acetyl-glycoprotein. After 24 wks of exposure, the 25 mg/kg ZnO NPs treatment increased Lipid–CH=CH–, Lipid=CHCH₂CH=, Lipid–CH=CH–, and –N(CH₃)₃ compared to the 25 mg/kg ZnSO₄ exposure; however, Lipid–CH₂CH₂CO, Lipid–CH₂CH=, and HDL LDL VLDL(–CH₂–) were lower in the 100 mg/kg ZnO NP exposure than that in the 100 mg/kg ZnSO₄ exposure (Figure 4B). The data here further indicated that the effect of ZnO NPs was different from that of ZnSO₄ in that the intact NPs might play an important role in changing the levels of large molecule metabolites.

In order to explore the underlying mechanisms of metabolomic changes caused by ZnO NPs, metabolic enzymes in the liver were investigated. Glutamate dehydrogenase (GLUD2) and aspartate aminotransferase (ASAT) are two important enzymes associated with amino acid metabolism. Compared to the control, 4 wks of 50 mg/kg ZnSO₄ exposure increased GLUD2; however, 50 and 100 mg/kg ZnO NP exposure decreased GLUD2 protein level (Figure 5). After 24 wks, both 25 mg/kg ZnSO₄ and 100 mg/kg ZnO NPs increased GLUD2. ASAT was increased by 100 mg/kg ZnO NP exposure for 24 wks. AMP deaminase (AMPD) is a vital enzyme for AMP metabolism and plays a critical role in energy metabolism. GPT2 catalyzes a reversible transamination reaction to yield glutamate and pyruvate, and participates in amino acid metabolism and gluconeogenesis. AMPD was increased by the 100 mg/kg ZnO NP treatment after 24 wks of exposure. After 4 wks, GPT2 was elevated by the 25, 50, and 100 mg/kg ZnO NPs exposure. After 24 wks, GPT2 was stimulated by the 100 mg/kg ZnO NPs exposure (Figure 5). Several other liver metabolism enzymes such as UGT, β-oxidation, CYP2A, CYP2B, and ACACA were analyzed, but they remained unaffected by either ZnSO₄ or ZnO NPs. The data here suggest that ZnO NPs differentially affected liver metabolism enzymes as compared to ZnSO₄.

After periods of 4 or 24 wks exposure, no treatments affected BW. After 24 wks, ZnO NPs dose-dependently caused liver steatosis. As shown in Figure 5A, liver histopathology in ZnSO₄ treatments was similar to that in the control group, while the grade of macrovesicular liver steatosis in the 50 and 100 mg/kg ZnO NP exposure groups was increased with an elevation of relative liver weight in these groups (Figure 6A; Hanin et al., 2017).

It was suggested that liver lipid metabolism enzymes might be disturbed by ZnO NPs since plasma lipid levels were differentially altered. Therefore, gene expression of fatty acid synthesis enzymes and lipid synthesis enzymes were investigated. It was found that, after 4 wks, the 25 mg/kg ZnO NP exposure stimulated the gene expression of ELOVL1, ELOVL5, ELOVL6, ELOVL7, CYP51A1, GPAM, DHCRT, FASN, NSDH1, and DECRL, which matched the metabolomics data (Figure 6B). However, after 24 wks, most of these genes were decreased by ZnO NP exposure compared to the control (Figure 6B). The protein levels of important lipid synthesis enzymes were also investigated and it was found that FASN and SREBF1 were altered. After 24 wks, FASN was stimulated by the 25, 50, and 100 mg/kg ZnSO₄ treatments; however, it was only increased by the 25 mg/kg ZnO NPs. After 24 wks, SREBF1 was elevated by the 50 and 100 mg/kg ZnSO₄ exposure but not by ZnO NPs (Figure 6C). Hepatic lipase (LIPC) was also explored. LIPC was elevated by the 50 and 100 mg/kg ZnO NP treatments after 4 and 24 wks of exposure. The data here suggested lipid degradation might be stimulated by ZnO NPs, and lipid synthesis might be decreased by ZnO NPs (Figure 6D). The data in this section indicated that lipid metabolites in plasma might reflect lipid metabolism in the livers under ZnO NP treatment.

Liver damage and apoptosis were investigated because liver function was altered by ZnO NP treatment. We found that the apoptosis marker caspase 8 was elevated by the 25, 50, and 100 mg/kg ZnO NP treatments after 24 wks of exposure in a dose-dependent manner. Mitochondrial damage marker TRIB3 was increased by the 50 and 100 mg/kg ZnO NP treatments after 4 and 24 wks exposure (Figure 7). Apoptosis markers caspase 3, Bcl-2, Bcl-xl, Bax, p53, and damage marker DDIT3 were also investigated; however, they remained unaltered by ZnO NPs. The data here suggest that the ZnO NPs might have caused the death of liver cells.