Genomic DNA was extracted using a standard cetyltrimethylammonium bromide (CTAB)-based protocol [22] and the DNA concentration was determined both spectrophometrically (Bio-Rad Smartspec) and fluorimetrically by PicoGreen (Bio-Rad Versafluor).

Standard and real-time Q-PCR was carried out with a Bio-Rad iCycler using Amplitaq Gold (Applied Biosystems) and the buffers supplied and using GMOdetect master mix (Diagenode) or supermix (Bio-Rad), respectively.

Calf thymus DNA (Invitrogen) was used as carrier DNA for plasmid dilution. pUC18 plasmid was purchased from Invitrogen. All primers were purchased from Eurogentec (Oligold oligonucleotides).

DNase- and RNase-free water was purchased from both Agros Organics and Fluka. Calf intestine alkaline phosphatase was purchased from Sigma. EZ Load HT molecular weight markers (100–2,000 bp) were purchased from Bio-Rad. CTAB was obtained from Fluka.

Plants from isogenic wild-type and GTS-40-3-2 soybean seeds (also indicated as Roundup Ready soybean) were cultivated in a growth chamber under standard conditions (25 °C, 16 h/8 h  day/night regime, 80% humidity, 20,000 lux).

Leaf tissue from both GTS-40-3-2 and non-transgenic plants was subjected to CTAB extraction to obtain genomic DNA [22]. Preparation of plasmid DNA was performed using a Bio-Rad Quantum miniprep kit and purified plasmid DNA was quantified spectrophotometrically (high concentration, i.e. falling outside the PicoGreen working range) or by PicoGreen (low concentration). The genomic DNA stock concentration was adjusted to 20 ng µl^-1 with Milli-Q water and the solution was stored at -20 °C.

The dual-target plasmid, pJANUS™-02-001, was constructed in house as a pUC18 derivate into which two PCR target sites, separated by a short linker, were inserted at position 399 (HindIII site), bringing the total size of the plasmid to 3,191 bp. One target site consists of a 360-bp fragment of the p35S/plant border of Roundup Ready soybean event GTS-40-3-2; the other target sequence consists of 125 bp of the soybean lectin gene (Le1) (accession numbers are given in “Appendix”). The Roundup Ready GTS-40-3-2 soybean junction amplicon matches the junction fragment described by Windels et al. [20]. The DNA sequence representing the soybean lectin gene target was chosen in house. The plasmid was propagated in TOP10F' cells (Invitrogen) and grown in liquid Luria–Bertani medium (Miller, Merck) containing 100 µg ml^-1 ampicillin (Sigma). The plasmid was extracted using a Quantum midiprep kit according to the manufacturer's protocol (Bio-Rad) and was stored at -20 °C.

During the first steps, purified plasmid DNA was diluted in RNase- and DNase-free water in the absence of carrier DNA: the midiprep plasmid DNA (referred to as superior master stock) was quantified spectrophotometrically (Bio-Rad Smartspec) prior to dilution and stored at -80 °C. The superior master stock was diluted firstly to a concentration of 50 pg µl^-1 in Milli-Q water (master stock) in two steps (intermediate dilution to 5 ng µl^-1). The master stock was requantified by PicoGreen and the latter measured quantity of DNA applied as the starting concentration for the preparation of the working stock. The working stock was prepared by a two-step dilution of the master stock in Milli-Q water containing 4 ng µl^-1 calf thymus DNA (final plasmid concentration at 5 fg µl^-1 in a final working stock volume of 10 ml) (intermediate dilution to 5 pg µl^-1). The master stock and the working stock were prepared freshly and stored at 6 °C. The stability of the plasmid stock solutions was verified by Q-PCR analysis and was found to be constant under the above-mentioned conditions for at least 2 weeks (standard deviation of the Ct value less then 0.25 Ct for 10,000 plasmid copies).

The standard with the highest plasmid copy number concentration (S1) was prepared by dilution of the working stock in RNase- and DNase-free water (containing 4 ng µl^-1 of the carrier DNA). All subsequent standards (S2–S6) were prepared by serial dilution of standard S1 in RNase- and DNase-free water (containing 4 ng µl^-1 of the carrier DNA).

The genomic GTS-40-3-2 soybean DNA standards were prepared in a similar fashion: the highest concentration standard (S1) was prepared directly from the genomic DNA stock, whereas subsequent standards (S2–S6) were prepared by serial dilution of standard S1. No carrier DNA was used for preparation of the genomic standards.

Three unknowns at 0.1, 1 and 5% GTS-40-3-2 soybean (percent haploid genome equivalents) were prepared by admixing GTS-40-3-2 soybean genomic DNA with genomic DNA from non-transgenic soybean plants. The unknowns were prepared as a single large batch, mixed overnight at 4 °C under constant agitation. Thirty aliquots of 200 µl of each unknown were prepared as test samples for the interlaboratory trial. The remainder of each unknown was stored separately and used to determine the “true values” of each unknown. Throughout the validation of the different materials received (master mix, primers, etc.) several quantifications of the unknowns were performed (Table 1) and the GMO value for each unknown (assigned value) was determined as the average of the different measurements (within run and between runs). Table 1

Overview of datasets in the “true value” calculation for samples U1, U2 and U3 (expressed as % Haploid Genome Equivalents)

Twelve laboratories were provided with the test materials, as described above. A set of genomic DNA samples with unknown GMO content (U1–U3) were included. In addition, each laboratory received a 7-ml vial of 2× universal Taqman PCR master mix (Diagenode), nuclease-free water and the GMO-specific primers and probe (targeting the 35S/plant border region) and the soybean-specific primers and probe (targeting the lectin Le1 gene). Primers and probes were provided lyophilized. All primers and probes were manufactured by Eurogentec (Oligold primer/probe quality) from a single batch and subsequently split into 12 vials. All materials were shipped on dry ice (DHL, same-day shipping). An Excel-based data report format was provided, allowing retrieval of all analytical data from all analyses.

All experimental analyses were performed between October 2007 and January 2008. All laboratories (but one) sent back their results including the raw data.

The GMO content of a sample was determined by the participating laboratories using real-time PCR relative to the endogenous reference. As a consequence, two calibration curves, transgenic and endogenous, were set up for each DNA calibrator of this relative quantification. The slope and the efficiency (ε) of the PCR are related to each other by the equation based on conventional real-time PCR theory [23]: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \varepsilon = \left( {{{10}^{ - \frac{1}{\text{slope}}}} - 1} \right) \times 100. $$\end{document}

The similarity of the transgenic and the endogenous calibration curves obtained with each type of calibrator was investigated by comparing (means and variances) the efficiencies of the PCR and the linearity (R²) of the two calibration curves. F ratio and Student t tests were performed to investigate if variances and means between groups were similar. When the F test revealed significant differences (p < 0.05), a Student t test assuming unequal variances was performed, using the Welch–Satterthwaite equation [24, 25] to calculate an approximation to the effective degrees of freedom. Otherwise, two-sided Student t tests assuming equal variances were used to investigate if the differences between groups were significant.

Percent genetic modification data of unknown samples were processed using procedures in SAS (version 9, SAS Institute, Cary, NC, USA). PROC UNIVARIATE provided an array of tests (Shapiro–Wilk, Kolmogorov–Smirnov, Cramer–von Mises and Anderson–Darling) for departure of the data distribution from normality, and graphic tools (box plot, stem-and-leaf representation) to detect outlying data. An ISO procedure [26] based on use of the Grubbs test and the Cochran test was run by the software program AMPE [27] to identify and discard outlying data prior to running analysis of variance (ANOVA). ANOVA is a technique for analysing experimental data in which one or more response (or dependent) variables are measured under various conditions identified by one or more classification factors. The variation in the response is separated into variation attributable to differences between the classification factors and variation attributable to random errors. An ANOVA constructs tests to determine the significance of the classification effects. A typical goal is to compare means of the response variables for various combinations of the classification factors. In this study, two-factor ANOVA was undertaken in SAS (PROC GLM) to investigate the effect on GMO values (response variable) of two classification factors—laboratory (1–12), source of the calibrator DNA (genomic, plasmid)—and their interaction. The two factors were arranged in a fully crossed experimental design. The Levene test was used to assess departure from homogeneous variances.

Three ISO validation measures [26] were computed: average bias (B, %), relative standard deviation of repeatability (RSD_r, %) and relative standard deviation of reproducibility (RSD_R, %). The larger these values, the less well the method performs.

For assessing method performance, (1) efficiency measures from the PCR were considered (E₁, average PCR efficiency for the endogenous target; E₂, average PCR efficiency for the transgenic target) and (2) the number of reagents used to run the PCR (indicated as “score”) was used as a measure of the practicability of the methods.

Both validation and PCR measures were combined into groups of similar measures (accuracy, efficiency and practicability) and then aggregated to an indicator of method validity, based on an implementation of the fuzzy-logic-based principle for method validation (software program AMPE [27]) and using the design and settings described in Bellocchi et al. [28]. Accuracy derives from combining the three validation measures; efficiency is the combination of the two efficiency measures; practicability includes the score. The three groups of measures are combined into the final indicator. Both fuzzy-logic-based groups and indicators range between 0 (best performance) and 1 (worst performance).

A dual-target plasmid was constructed wherein a part of the endogenous lectin gene and a region spanning one of the insert junction regions of GTS-40-3-2 soybean are integrated (see “Materials and methods”). An in-house Q-PCR efficiency estimation with the GTS-40-3-2 detection system described by Terry et al. [21] was performed on a linear dilution series spanning from about one to 100,000 plasmid copies (Fig. 1). The results show an acceptable linear correlation between 10 and 100,000 copies (endogene R² = 98.8; transgene R² = 99.3). The PCR efficiencies determined were within the limits as proposed by the European Network of GMO Laboratories (ENGL) guidelines (see Table 2). It was concluded that the pJANUS™-02-001 plasmid performs adequately for use as a calibrator template DNA in combination with the GTS-40-3-2 Q-PCR system of Terry et al. [21]. Note that this pJANUS™-02-001 is a versatile tool, allowing the use of other Q-PCR methods [29–31] for event-specific and endogene detection. Fig. 1

The dynamic range of quantitative polymerase chain reaction (PCR) system of Terry et al. [21] with pJANUS™-02-001 plasmid as template DNA. Ct results from a dilution series spanning four log values are plotted against the number of copies involved in the reactions. The dashed line represents the endogene no template control (NTC) reaction, whereas the dotted line represents the transgene NTC reaction

In total, 12 laboratories participated in the Roundup Ready soybean trial. Each laboratory ran two 96-well plate analyses with both transgene and endogene reactions; in one analysis 100% Roundup Ready soybean isogenic genomic DNA was used for construction of a calibration curve, whereas in the other analysis the pJANUS™-02-001 plasmid DNA was used in a carrier DNA background. Standard curves consisted of six-point dilution series (non-linear, spanning from 15,600 copies down to about eight). Both analyses contained the same set of three unknown samples (at 4.5, 1.94 and 0.29% measured true values, see “Materials and methods”).

When genomic DNA was used as the calibrator, the estimated PCR efficiencies for both endogenous and transgenic targets were relatively low (Table 3). However, compared with the genomic DNA calibrator, the plasmid calibrator did not yield significantly higher efficiencies [p(t) > 0.05] for both the endogenous and the transgenic targets. With the endogenous target, PCR efficiencies from the genomic DNA showed a significantly larger variance [p(F) < 0.05] than those obtained by using the plasmid DNA. Table 3

Comparison of the PCR efficiencies and linearity of the calibration curves of the genomic and plasmid DNA calibrators by means of a two-sided Student t test (comparison of means) and the F ratio test (comparison of variances) for endogenous and transgenic targets

The coefficient of determination (R²) provides information about the fitting of data to a linear calibration curve. Comparison of R² values of the plasmid and genomic DNA calibration curves did not show significant differences for both endogenous and transgenic targets. Also in this case, significantly larger variability [p(F) < 0.05] was observed with the genomic calibrator for endogenous DNA.

Outliers were identified at four data sets and were discarded. The remaining GMO values obtained from the unknown samples were in general normally distributed, although in two cases a small deviation from normality (p < 0.05) was observed. Violations of the assumption of equal variances were not recognized. For this reason, a conventional parametric approach was applied in the fully crossed two-factor ANOVA. Table 4 shows the breakdown of the total variance into contributing components to perform statistical inferences for experimental factors (laboratory and DNA source) and their interactions. On the basis of the data in Table 4, there are only random differences (p > 0.05) associated with either the laboratory or the DNA source. Also, non-significant interactive effects between the laboratory and the DNA source (p > 0.05 in Table 4) prove the stability of the use of both genomic DNA and plasmid DNA as a calibrator under different laboratory conditions. Table 4

Genetically modified organism (GMO) content from unknown samples: two-factor PROC GLM analysis of variance table

A set of performance measures are reported in Table 5 as computed from the interlaboratory trial using estimated assigned values as true GMO contents. The precision measures are similar for genomic and plasmid DNA, indicating quite large variability (RSD_r >25% and RSD_R >35%, indicating precision measures exceeding the minimum performance requirements set by the ENGL [7]). Table 5

Basic and fuzzy-logic-based aggregated performance metrics for genomic and plasmid calibrator DNA, evaluated at three GMO levels

Whether the average bias tends to be higher when the plasmid DNA is used as a calibrator (especially for higher GMO contents), the values for the accuracy are similar. Higher PCR efficiencies yielded by the plasmid DNA are reflected in better values for the efficiency (0.141 against 0.465). However, for the target gene, the efficiencies are less than the 0.98 limit as set by the ENGL as the acceptance criterion.

Overall, the final values of the aggregated indicator are more than 0.500, with the plasmid calibrator performing only marginally better (0.523–0.547) than the genomic DNA (0.569–0.575).