Introduction

Carotenoids including β-carotene have been intensively studied as target metabolites to develop biofortified crops using genetically modified (GM) technology because of their functions as precursors of vitamin A, which is essential to human health (Farré et al. 2011). When vitamin A is in short supply for infants and pregnant women, severe clinical symptoms associated with night blindness, xerophthalmia, and breakdown of the human immune system can result (Baisakh et al. 2006). Carotenoids are also beneficial to reduce the incidence of several cancers, cardiovascular diseases, age-related macular degeneration, and other degenerative diseases (Landrum and Bone 2001; Perera and Yen 2007). To alleviate vitamin A deficiency (VAD) through biofortifying crops, a carotenoid-biofortified rice (PAC rice) was developed using a recombinant PAC (Psy-2A-CrtI) gene that has bicistronic expression, which is an effective alternative to simultaneously introduce two carotenogenic genes through connection with a self-cleavable 2A sequence (Ha et al. 2010; Jung et al. 2011).

As metabolomics can provide a biochemical snapshot of an organism’s phenotype, it was developed to identify unpredicted changes of metabolites, and has become an important complementary tool in safety assessment (Catchpole et al. 2005). Among these analyses, the profile of low-molecular-weight molecules is closely related to an organism’s phenotype and can highlight important nutritional characteristics (Hoekenga 2008; Kok et al. 2008). The PAC rice was intentionally engineered to enhance levels of useful carotenoid metabolites but could contain a redistribution of secondary metabolites involved in isoprenoid metabolic pathways. However, primary metabolites may have unintended effects on the metabolism of the plant itself. Pigmented rice cultivars have been previously reported to be higher in carotenoids as well as in anthocyanin than common white rice cultivars (Kim et al. 2010). As the specific profile of metabolite molecules might be closely related to the organism’s phenotype, including important nutritional characteristics, to assess the effects of the new gene insertion on metabolite content of novel biotechnology-derived rice samples, we examined the profiles of low-molecular-weight hydrophilic metabolites in six rice cultivars: five conventional non-GM cultivars (three white and two red), the non-GM counterpart cultivars, and the PAC rice GM cultivar. This polar metabolite profiling is suggested as an efficient evaluation method to determine the degree of substantial equivalence among GM and non-GM rice cultivars.

In addition, the OECD (2004) consensus documents have identified key foods and both feed nutrients and anti-nutrients required for safety studies of new rice varieties. Guidelines published by the European Food Safety Authority (EFSA) or Codex Alimentarius Commission of FAO/WHO have requested that the size of the data package be expanded to demonstrate equivalence over two seasons, and in cases of herbicide tolerance, to include treatments with and without the recommended herbicide applications. These herbicide treatment regimens have been included to allow the effects of the herbicide on crop composition to be evaluated. Comparative compositional analyses have been reported for transgenic rice containing a bar gene (Li et al. 2008; Choi et al. 2012). However, herbicide applications to such transgenic rice have not been reported. PAC rice contained the bar gene as a selection marker to give the benefit of herbicide tolerance. Therefore, we analyzed the nutritional composition and anti-nutrient content of PAC rice grown with application of the herbicide glufosinate, and then assessed whether nutritional quality of the PAC rice is altered in generations following herbicide treatment.

Materials and methods

Rice samples

PAC rice was developed using Agrobacterium tumefaciens-mediated transformation into japonica-type Korean rice (Oryza sativa cv. Nakdongbyeo) by Ha et al. (2010). Field trials used for production of material for the comparative assessment should be performed in order to assess differences and equivalences between three test materials: the GM plant, its counterpart, and non-GM reference varieties (EFSA 2011). For equivalence comparisons between GM and non-GM rice cultivars, a non-transgenic parent cultivar, Nakdongbyeo (NDB), was used as the counterpart. Both cultivars were grown in the same paddy field at the Rural Development Administration (RDA), Suwon, Korea, in 2010. A randomized complete plot design was established at 10 sites. Plot size was 4 m2. Other common rice cultivars, including Hwasungbyeo (HSB) and Ilpumbyeo (IPB), were grown under the same environmental conditions. A further two red pigmented cultivars, Hongjinjubyeo (HJB) and Jeogjinjubyeo (JJB), were provided by the National Institute of Crop Science, RDA, Suwon, Korea, and were also grown under the same environmental conditions. Two months after transplantation, glufosinate ammonium herbicide (Basta; Bayer Crop Science, Germany) was applied at 108 g active ingredient/ha as a single application. After harvesting, the whole grain (rough rice) samples were dried to a final moisture content of 10 %. Rice grain samples were manually hulled and ground to obtain a fine powder using a cyclone mixer mill (HMF-590; Hanil, Seoul, Korea) and a mortar and pestle. The powder was stored at −80°C prior to analysis.

Gas chromatography time-of-flight mass spectrometry analysis of polar metabolites

Extraction of polar metabolites was performed in accordance with a previously described procedure (Kim et al. 2007). A sample of 100 mg of ground sample was extracted with 1 mL of methanol/water/chloroform (2.5:1:1 by vol). A total of 60 μL of ribitol solution (0.2 mg/mL) was added as an internal standard (IS). Extraction was performed at 37 °C with a mixing frequency of 1,200 rpm using a thermomixer compact (Eppendorf, Hamburg, Germany). The solutions were then centrifuged at 16,000g for 3 min. The polar phase (0.8 mL) was transferred into a new tube and 0.4 mL of water was added. The well-mixed content of the tube was centrifuged at 16,000g for 3 min. The methanol/water phase was dried in a centrifugal concentrator (CVE-2000; Eyela, Tokyo, Japan) for 2 h, followed by a drying in a freeze-dryer for 16 h. Methoxime (MO) derivatization was carried out by adding 80 μL of methoxyamine hydrochloride (20 mg/mL) in pyridine and shaking at 30 °C for 90 min. Trimethylsilyl (TMS) etherification was carried out by adding 80 μL N-methyl-N-trimethylsilyltrifluoroacetamide (MSTFA) at 37 °C for 30 min. Gas chromatography time-of-flight mass spectrometry (GC–TOFMS) was performed using a gas chromatograph (model 7890A; Agilent, Atlanta, GA, USA), which was coupled to a Pegasus HT TOF mass spectrometer (LECO, St. Joseph, MI, USA). The derivatized sample (1 μL) was separated on a 30 m × 0.25-mm I.D. fused-silica capillary column coated with 0.25 μm CP-SIL 8 CB low bleed (Varian, Palo Alto, CA, USA). The split ratio was set at 1:25. The injector temperature was 230 °C. The helium gas flow rate through the column was 1.0 mL/min. The temperature program used an initial temperature of 80 °C, which was maintained for 2 min, followed by an increase to 320 °C at 15 °C/min, and finally a 10-min hold at 320 °C. The transfer line and ion-source temperatures were 250 and 200 °C, respectively. The scanned mass range was 85–600 m/z, and the detector voltage was set at 1,700 V.

Compositional analysis

Crude protein content was estimated by determining the total nitrogen content using the Kjeldahl method (AOAC 2005a), and crude fat was analyzed by the Soxhlet extraction method (AOAC 2005b). Ash content was determined by gravimetrically measuring sample residue after ignition in an oven at 600 °C to constant weight (AOAC 2005c). Total dietary fiber was determined on duplicate samples of dried and defatted material according to AOAC method 991.43 (AOAC 2005d).

For the amino acid analysis, the sulfur-containing amino acids cysteine and methionine were oxidized by performic acid before hydrolysis with hydrochloric acid, and the remaining 15 amino acids were analyzed with an automatic amino acid analyzer (L-8500-A; Hitachi, Tokyo, Japan) directly after protein hydrolysis with hydrochloric acid (AOAC 2005e).

Fatty acid content was determined by lipid extraction and saponification with 0.5 N sodium hydroxide in methanol. The saponification mixture was methylated with 14 % boron trifluoride/methanol, and the resulting methyl esters were extracted with pentane. The methyl esters of the fatty acids were analyzed by GC (model 5890A; Hewlett Packard, Avondale, PA, USA) (AOCS 1997a). Pentadecanoic acid was used as the IS.

Levels of calcium, potassium, sodium, magnesium, zinc, iron, and copper were determined by inductively coupled plasma optical emission spectrometry (Integra XL inductively coupled plasma optical emission spectrometer; GBC, Melbourne, Australia), according to AOAC method 999.11 (AOAC 2000).

Vitamins B1 and B2 were extracted according to a slight modification of the methods of Sims and Shoemaker (1993) and Esteve et al. (2001), respectively. The vitamins were determined using a HPLC method with fluorometric detection (Shimadzu, Kyoto, Japan). Niacin and vitamin E were detected as described by our group (Kim et al. 2011, 2012). Phytic acid was analyzed by ion exchange chromatography following Latta and Eskin (1980). Trypsin inhibitor activity was determined in alkali solvent-extracted rice samples using AOCS method Ba 12-75 (AOCS 1997b).

Statistical analysis

In unpolished rice seeds, 52 metabolites were identified by GC–TOFMS, and the profiled metabolite data were analyzed using principal components analysis (PCA). The PCA (SIMCA-P v.12.0; Umetrics, Umeå, Sweden) output consisted of score plots to visualize the contrasts between different samples and loading plots to explain the cluster separation.

Equivalence tests were used to determine whether treatment differences exceeded the range of normal variation of the comparator. We used the two one-sided test (TOST) procedures for equivalence testing. The statistical analysis was conducted with the SAS 9.2 software package (SAS Institute, Cary, NC, USA). In the TOST procedure, the null hypothesis tested was “treatment 1 is not equivalent to treatment 2” versus the alternative hypothesis “treatment 1 is equivalent to treatment 2”. To test for equivalence, the 90 % confidence intervals for the difference between the two treatments were determined. The confidence intervals, 100(1 − 2 × α)% (=90 %), where α = 0.05, were calculated in pairs for the treatment differences. Thus, the 90 % confidence interval method was equivalent to conducting TOST at the 5 % significance level. Equivalence boundaries were set to ±20 % of the means. Experimental data were also analyzed by the analysis of variance (ANOVA) combined with Duncan’s multiple-range test using the SAS 9.2 software.

Results

Metabolic profiles among GM and conventional rice cultivars

To evaluate substantial equivalence between GM and non-GM rice cultivars through profiling of unintended polar metabolites, GC–TOFMS was used to identify low-molecular-weight metabolites in carotenoid-biofortified rice and five conventional rice cultivars: three white, including their counterparts (NDB, HSB, and IPB), and two red (HJB and JJB). The ChromaTOF software was used to assist with peak location. Peak identification was performed by comparing the reference compounds and using an in-house library. In addition, identification of several metabolites was performed by direct comparison of the sample mass chromatogram with those of commercially available standard compounds, which were obtained by a similar MO/TMS derivatization and GC–TOFMS analysis. In total, 52 metabolites (i.e., 20 organic acids, 19 amino acids, 9 sugars, 3 sugar alcohols, and 1 amine) were detected in unpolished rice grains (Fig. 1). The corresponding retention times and their fragment patterns are given in Table 1. Quantification was performed using selected ions. The quantitative calculations of all analytes were based on the peak area ratios relative to that of the IS. The data for the 52 metabolites were subjected to PCA to assess the overall experimental variation and to examine differences in metabolite profiles among cultivars (Fig. 2). We are interested in data similarity rather than the ability to discriminate classes. Thus, we reason that, if PCA clusters metabolome samples close together, then they can be objectively considered to be similar. PCA revealed that the two highest-ranking principal components accounted for 61.2 % of the total variance within the dataset. The PCA results clearly demonstrated the absence of marked variances among samples of the same cultivar. Although the first principal component, accounting for 43.9 % of total variance, resolved the measured metabolites profiles of pigmented and non-pigmented rice, PCA could not distinguish between transgenic rice (PAC) and its non-transgenic counterpart rice (NDB). To further investigate the contributors to the principal components, the metabolic loadings in principal component 1 (PC1) and principal component 2 (PC2) were compared. In PC1, the corresponding loading was positive for all sugars except trehalose and raffinose. The loading indicated that sugars were higher in red-pigmented rice than in non-pigmented white rice. Collectively, transgenic rice (PAC) could not be distinguished from its non-transgenic counterpart (NDB), supporting their substantial equivalence from profiling of unintended polar metabolites. PAC and NDB showed closer co-separation than other white (HSB and IPB) and red (HJB and JJB) cultivars.

Fig. 1
figure 1

Selected ion chromatograms of metabolites extracted from non-transgenic rice (cv. NDB) (a) and transgenic rice (PAC) (b) as MO/TMS derivatives separated on a 30 m × 0.25-mm I.D. fused-silica capillary column coated with 0.25-μm CP-SIL 8 CB low bleed. Peak identification: 1 pyruvic acid, 2 lactic acid, 3 valine, 4 alanine, 5 oxalic acid, 6 glycolic acid, 3′ valine, 7 serine, 8 ethanolamine, 9 glycerol, 10 leucine, 11 isoleucine, 12 proline, 13 nicotinic acid, 14 glycine, 15 succinic acid, 16 glyceric acid, 17 fumaric acid, 7′ serine, 18 threonine, 19 β-alanine, 20 malic acid, 21 salicylic acid, 22 aspartic acid, 23 methionine, 24 pyroglutamic acid, 25 4-aminobutyric acid, 26 threonic acid, 27 arginine, 28 glutamic acid, 29 phenylalanine, 30 p-hydroxybenzoic acid, 31 xylose, 32 asparagine, 33 vanillic acid, 34 glutamine, 35 shikimic acid, 36 citric acid, 37 quinic acid, 38 fructose, 38′ fructose, 39 galactose, 40 glucose, 41 syringic acid, 42 mannose, 43 mannitol, 44 ferulic acid, 45 p-coumaric acid, 46 inositol, 44′ ferulic acid, 47 tryptophan, 48 sinapic acid, 49 sucrose, 50 cellobiose, 51 trehalose, 52 raffinose, IS internal standard (ribitol)

Full size image
Table 1 The detected chromatographic and spectrometric data of the 52 identified compounds analyzed by GC–TOFMS
Full size table
Fig. 2
figure 2

Scores (a) and loading plots (b) of principal components 1 and 2 of the PCA results obtained from polar metabolite data on transgenic rice (PAC), its non-transgenic counterpart cultivar (NDB), and four conventional rice cultivars. The dotted circle indicates that PAC rice was less separated from NDB than other white rice cultivars (HSB and IPB)

Full size image

Nutrient compositional analysis of GM rice following herbicide treatment

Following guidelines published by the EFSA and Codex Alimentarius Commission of FAO/WHO, composition studies for herbicide-tolerant crops must include transgenic entries that are treated, and not treated, with the herbicide they are engineered to tolerate. PAC rice including the bar gene as a selectable marker was grown in the same manner as is common for commercial production, using either conventional weed control practices or an application of glufosinate-ammonium herbicide, and was then compared to its respective counterpart for statistical analysis. Comparative analysis of the rice grains was completed using a statistical procedure to assess equivalence. Performing an ordinary t test and inferring equivalence from the absence of a significant difference entails an uncontrolled increase in the risk of false-positive conclusions; that is, the assumption of “equivalence”. In other words, a “non-significant difference” is different from “significant equality”. Statistical equivalence for a component was assumed if the mean values of the two treatments did not substantially differ, i.e., the difference in the mean values was within a certain interval. In equivalence studies, the choice of a 90 % confidence interval is customary as it corresponds with the customary 95 % level for statistical testing of equivalence. We set equivalence boundaries at 20 % of the means, as recommended by the Nordic Council of Ministers (Nordic Council 2000), and as applied in equivalence tests using limits of ±20 % for comparative compositional analyses of glufosinate-tolerant rice and insect-resistant rice, according to Oberdoerfer et al. (2005) and Park et al. (2012), respectively. In total, 45 different compositional analyses were conducted to evaluate the equivalence of PAC rice and conventional rice.

As shown in Supplementary Table 1, no significant differences were found for the mean values of ash, lipid, protein, carbohydrate, and dietary fiber between the non-transgenic comparator and the transgenic rice grain produced either with conventional herbicides or following the application of glufosinate-ammonium herbicide. For equivalence analysis of amino acids and fatty acids, the percentages of particular amino acids and fatty acids in the total protein and total fatty acid, respectively, were calculated (Supplementary Tables 2 and 3). The levels of 17 amino acids and 9 fatty acids in PAC rice were also very similar to those of its non-transgenic counterpart. Statistical analysis indicated equivalence as expected. Although the consensus document published by the OECD was not available, the measured values of fatty acids fell within the ranges of values observed in brown rice of the six typical Korean cultivars as reported by Choe et al. (2002). Of the remaining 14 analytes, copper, potassium, magnesium, and niacin showed no statistical equivalence between PAC rice and its non-transgenic counterpart (Supplementary Table 4). When one or more components of the transgenic crop differed significantly from the composition of the near-isogenic comparator, the levels of the analytes from the transgenic crop were comparable to the levels of those analytes reported in other cultivars of the same crop. This approach recognizes that although no crop or food can be proven safe, transgenic crops can be compared with similar crops that have a history of safe consumption. No compositionally unsafe varieties have been identified in major crops such as maize, soybean, cotton, and rice; therefore, a new transgenic variety is considered safe if its analyte levels are similar to other varieties of the same crop (Herman et al. 2010). The copper, potassium, magnesium, and niacin levels in PAC rice fell within the ranges reported by the OECD (2004) and raised no safety concerns. Vitamin B2 levels were lower than those provided by the OECD (2004), but this was also true for the non-transgenic control, indicating that this was not a consequence of genetic modification.

Discussion

As omics strategies are applied to measure as many features of the target system as possible, they are a natural choice for evaluating substantial equivalence. Reported applications include transcriptomics (Barros et al. 2010), proteomics (Corpillo et al. 2004), and metabolomics (Catchpole et al. 2005; Kusano et al. 2011). Of these, metabolomics is of particular interest because the composition of low-molecular-weight molecules includes important nutritional and toxicological characteristics. Thus, the metabolomics approach has been developed and applied to identify unintended effects on a new gene insertion (Catchpole et al. 2005; Kusano et al. 2011). PCA at the metabolome level has been used widely in assessing differences between GM crops and their non-GM counterparts (Catchpole et al. 2005; Kusano et al. 2011). The goal of substantial equivalence is not to draw a conclusion about a novel organism’s safety status because that would require the testing of all compounds, which is not possible. Instead, by examining a broad set of traits, substantial equivalence evaluations aim at obtaining a picture of the magnitude and nature of incurred changes to use as a screen for potentially problematic changes and a starting point for further investigation (Kuiper et al. 2001). Metabolic profiling using GC–TOFMS is a useful tool for investigating compositional similarities between GM and conventional crops (Catchpole et al. 2005; Kusano et al. 2011). This tool also provides a potential method for assessing undesirable changes in GM crops that enhance a nutrient or create novel nutrients through the intentional modification of metabolic pathways. The analysis report here demonstrated that the metabolite profile of PAC rice was similar to that of its parent cultivar, rather than that of other cultivars. These results agree with numerous published studies for other transgenic events, whereby input traits were found to have a negligible effect on crop composition as compared to traditional breeding methods (Oberdoerfer et al. 2005).

Analyses of nutrients and anti-nutrients were conducted to investigate the compositional equivalency between transgenic PAC rice (both treated and not treated with glufosinate-ammonium herbicide) and its non-transgenic counterpart. For copper, potassium, magnesium, and niacin, equivalence between the datasets could not be confirmed statistically, but all mean values calculated for these micronutrients in the transgenic samples were within the reference range reported by the OECD (2004). The levels of other components tested in the PAC rice were statistically indistinguishable from those of the non-transgenic counterpart, which indicates that the PAC rice is substantially equivalent to its conventional counterpart.