Testing the interaction between analytical modules: an example with Roundup Ready soybean line GTS 40-3-2.

BACKGROUND: The modular approach to analysis of genetically modified organisms (GMOs) relies on the independence of the modules combined (i.e. DNA extraction and GM quantification). The validity of this assumption has to be proved on the basis of specific performance criteria.
RESULTS: An experiment was conducted using, as a reference, the validated quantitative real-time polymerase chain reaction (PCR) module for detection of glyphosate-tolerant Roundup Ready(R) GM soybean (RRS). Different DNA extraction modules (CTAB, Wizard and Dellaporta), were used to extract DNA from different food/feed matrices (feed, biscuit and certified reference material [CRM 1%]) containing the target of the real-time PCR module used for validation. Purity and structural integrity (absence of inhibition) were used as basic criteria that a DNA extraction module must satisfy in order to provide suitable template DNA for quantitative real-time (RT) PCR-based GMO analysis. When performance criteria were applied (removal of non-compliant DNA extracts), the independence of GMO quantification from the extraction method and matrix was statistically proved, except in the case of Wizard applied to biscuit. A fuzzy logic-based procedure also confirmed the relatively poor performance of the Wizard/biscuit combination.
CONCLUSIONS: For RRS, this study recognises that modularity can be generally accepted, with the limitation of avoiding combining highly processed material (i.e. biscuit) with a magnetic-beads system (i.e. Wizard).

Background

To comply with the European regulation concerning the labelling and traceability of genetically modified organisms (GMOs) and GM products [1] and offer freedom of choice to consumers, development of reliable, sensitive and accurate methods for GMO detection and quantification in food, feed and raw materials is essential [2]. Measurement of deoxyribonucleic acid (DNA) by polymerase chain reaction (PCR) has been widely used in different fields of food analysis and real-time PCR is the established preferred method for quantification of GMOs [3-5]. This is reflected in the EU legislation establishing the European Union Reference Laboratory [1], hereinafter referred to as the EURL for Genetically Modified Food and Feed (EURL-GMFF), whose main task is to validate methods for detection of GMOs in order to ensure full traceability along the food and feed chain [6] and that methods cover all the steps needed, including DNA extraction and subsequent quantification by PCR [7]. The whole analytical procedure for GMO quantification in food and feed consists of several, add-on sequential steps, from sample preparation to DNA extraction, purification and real-time PCR measurement. The result is provided in the form of a ratio between the GM- and the species-specific target sequences, preferably expressed in terms of haploid genomes [8,9].

The modular approach looks at an analytical method as a combination of different procedural steps in the analytical chain (the 'modules') and is suitable to rationalise both validation and application of methods for GMO detection [10]. Modularity implies independence of modules, therefore it allows for flexibility to combine modules on the one hand and for harmonisation on the other hand. If modular validation is to be applied, fit-for-purpose procedures and general acceptance of minimum requirements for each module are needed in order to evaluate the uncertainties associated with each module.

Measurement of the uncertainty is relevant to assess the quality of the analytical result. The uncertainty components of various modules are currently being evaluated at different levels by several groups (the European Network of GMO Laboratories, the Consultative Committee for Amount of Substances Metrology in Chemistry, Codex Alimentarius, the European Committee for Standardisation [CEN], and the International Organisation for Standardisation [ISO]). Nevertheless, despite these efforts on international standardisation, the minimum quality requirements that extracted DNA should meet in order to be fit for the analytical module consisting of quantitative real-time PCR measurement (qPCR) should be clearly defined and experimentally corroborated.

Previous investigations suggested that different extraction methods could influence DNA quantification in food products by qPCR [11,12]. Because PCR requires a high-quality DNA template in terms of DNA integrity and purity, the method used to extract the DNA from the starting material is critical, as pointed out by other Authors [13,14]. Other studies have demonstrated that quantification of GMOs was affected by the degree of processing of the matrix from which genomic DNA (gDNA) was extracted [15-17]. Similar influences on GMO quantification have been reported for processed corn [18].

The modular approach was proposed to facilitate validation of the procedure for GMO measurement [10]. This pragmatic approach implies that each step (module) in the analytical procedure, i.e. DNA extraction and determination of the DNA concentration, can be decoupled either from the previous or from the next one, provided that each step fulfils certain quality criteria. The advantage of such an approach lies in the fact that separately validated modules, for example for DNA extraction or for real-time PCR, could then be combined. The same approach has also been taken in ISO standards on GMO analysis [19,20]. Moreover, if the real-time PCR measurement is not influenced by the type of extraction method applied it can be validated with DNA extracted by any method and from any type of matrix, although scientific evidence for this has yet to be produced.

In this pilot study, the EURL-GMFF's validated method for detection of the world's leading GM soybean crop [21] - glyphosate-tolerant Roundup Ready soybean (unique identifier MON-Ø4Ø32-6) - was chosen as a model. The study is intended as a proof of concept for testing whether the DNA extraction methods applied to a given matrix prior to qPCR analysis can affect quantification of GMO in food or feed and to investigate which, if any, criteria should be applied in order to make the qPCR module independent from the DNA extraction module.

Methods

Sample material

Powdered certified reference material (CRM) for genetically modified soybean line 40-3-2 (1% RRS ERM-BF410d) was provided by the European Commission's Institute for Reference Materials and Measurements (IRMM, Geel, Belgium) and is commercialised by Fluka http://www.sigmaaldrich.com/analytical-chromatography/fluka-analytical/about-fluka-and-riedel.html. Feed meal (ingredients: soybean flour, wheat bran, dried beet pulp, beans, sugar beet molasses, calcium carbonate, magnesium oxide, sodium hydrogen carbonate and palm oil) was purchased in a local store.

Biscuit samples, kindly provided by the University of Parma (UPAR), Italy were prepared as follows: 100% RRS flour was mixed with non-GM RRS flour (provided by Progeo Molini S.p.A, Italy: http://www.progeomolini.it) to obtain a 1% GM mixture. Subsequently, the mixture was used for biscuit production with the following ingredients: 200 g of wheat flour, 100 g of mixed RRS flour, 50 g of sugar, 7 g of yeast and 200 g of water, stirred just until the ingredients were moistened and the dough formed a soft ball. The dough was kneaded until smooth. Biscuits were cut out, each weighing 15 grams. The biscuits were baked in an oven at 180 °C for 10 minutes.

Data generation

An experimental plan was designed by the ISS (Rome, Italy) and the EURL-GMFF (Joint Research Centre, European Commission, Ispra, Italy) to evaluate the effect of alternative DNA extraction methods applied to matrices (and possible interaction effects) on determination of the content of Roundup Ready soybean (RRS, line GTS-40-3-2) relative to total soybean DNA (see Figure 1).

Figure 1

Experiment design followed to test for module independence.

Three matrices containing Roundup Ready soybean were selected. Feed and biscuit contained the GM ingredient at unknown DNA concentrations. The third matrix was certified reference material containing 10 g kg-1 of Roundup Ready soybean (CRM 1%, ERM-BF410d). The samples tested were selected to reproduce a range of real cases from unprocessed grain/flour (CRM 1%) to feed meal generally present on the market in complex feed formulations and a heavily consumed bakery product for humans. Total DNA was extracted at ISS (six extractions from each matrix in the same day) from 200 mg of each of the above-mentioned matrices by three well-known and widely used extraction methods, based on different purification and extraction mechanisms:

- the CTAB-based protocol ('CTAB') a cellular lysis buffer, and selective precipitation with cetylammonium bromide [20];

- the Dellaporta-derived method ('Dellaporta') uses first a lyses step (thermal lyses in the presence of Tris HCl, EDTA, CTAB and β-mercaptoethanol) followed by deprotenisation and removal of contaminants by phenol-chloroform precipitation [22]. The method has been validated by EURL-GMFF in the context of method validation for specific quantification of genetically modified soybean line Roundup Ready®. The protocol requires extraction of 6 g of sample, but in the case of the CRM 1%, given its homogeneity and stability, the quantity was reduced to 200 mg, with all the reagents scaled down accordingly. No changes were made to the protocol to extract DNA from the other matrices http://gmo-crl.jrc.ec.europa.eu/summaries/A2704-12_soybean_DNAExtr_report.pdf;

- and the Promega (Promega Italia s.r.l., Milan, Italy) Wizard™ ('Wizard') magnetic DNA purification for food kit [23]. The method uses a lyses step (thermal lyses in presence of Tris HCl, EDTA and SDS) and applies the Promega's MagneSil magnetic bead technology for the DNA isolation based on silica as affinity matrix in a mobile solid phase.

The extraction methods were selected according to their wide application in GMO analyses, the difference on lyses buffer and on DNA isolation technique.

The concentration of the extracted DNA was determined at the EURL-GMFF, by fluorescence detection using PicoGreen dye for dsDNA quantification with a BioRad VersaFluor fluorometer (BioRad, Milan, Italy), and at the ISS, spectrophotometrically by NanoDropTM ND-1000 (Celbio s.p.a., Milan, Italy). Spectrophotometric measurements were performed at wavelengths of 230, 260 and 280 nm. Absorbance ratios at 260:280 nm and 260:230 nm were used to assess the detection of protein impurities or other contaminants that absorb strongly at or near 280 nm (with a 260:280 ratio < 1.8 and > 2.0) and impurities of phenol or other organic compounds (with a 260:230 ratio < 2.0) co-extracted with the nucleic acids from the specimens [24].

Acceptance criteria for DNA quality

DNA extracts were analysed by electrophoresis of 1% agarose gel pre-stained with ethidium bromide. Two microlitres of each extract were mixed with tracer dye - 6X DNA loading dye (Fermentas Cat. #R0611) - and an appropriate amount of distilled water, loaded in the gel and run at 70V. Lambda DNA digested with EcoRI and Hind III was used as the molecular size marker (Promega Cat. #G1731). A gel camera (Bio-Rad Gel Doc 2000) provided the digital photograph recording system.

Three performance criteria need to be met before DNA extracts can be accepted: DNA concentration needs to meet the requirement of the validated method; the DNA is of sufficient length to be amplified by PCR; and it is adequately free from PCR inhibitors. In the case of the latter, the DNA quality was demonstrated by analysing in duplicate four four-fold serial dilutions of each DNA extract (inhibition runs) using the validated soybean-specific reference system (lectin gene: http://gmo-crl.jrc.ec.europa.eu/summaries/40-3-2_validated_Method.pdf). First, a four-fold dilution series is prepared (from 1:4 to 1:256). To assess the presence of inhibitors, the Ct values of the diluted samples are plotted against the logarithm of the dilution factor and an equation is calculated by linear regression. Moreover, the Ct value of the 'undiluted' sample extrapolated from the linear regression is compared with the Ct measured from the same sample. The conditions to be met before DNA extracts can be accepted are: the slope of the regression line must be between -3.6 and -3.1; the linearity (R) must be equal to or above 0.98; and the difference between the measured Ct and the extrapolated Ct value (ΔCt) must be within 0.5. This approach has been endorsed by the European Network of GMO Laboratories (ENGL: http://engl.jrc.ec.europa.eu) and validated by the EURL-GMFF [25].

The content of Roundup Ready® soybean relative to the total DNA content in the DNA extracts was quantified using a validated Roundup Ready® soybean line-specific method http://gmo-crl.jrc.ec.europa.eu/summaries/40-3-2_validated_Method.pdf. To determine the % RRS soybean content in samples, four microlitres of DNA extracts were loaded per reaction, up to a maximum amount of 200 ng of DNA based on estimates from fluorimetric means). Each DNA extract was analysed in triplicate and the relative GM content was calculated by generating two standard curves (one for the reference sample, the other for the RRS system) plotting the Ct values measured for the calibration points. Calibration curves were built by DNA copy number, starting from a 10% GM content in the first sample and then serially diluting to encompass more than two orders of magnitude.

To determine the amount RRS DNA in the DNA extract, the RRS copy number was divided by the copy number of the soybean reference gene (lec) and multiplied by 100 to obtain the percentage value (GM% = RRS/lec * 100).

Experiment design and data processing

Both the DNA yield and the % GM DNA quantification were processed using SAS procedures (version 9, SAS Inst. Inc., Cary, NC, http://www.sas.com). PROC UNIVARIATE provided an array of tests (Shapiro-Wilk, Kolmogorov-Smirnov, Cramer-von Mises and Anderson-Darling) for departure of data distribution from normal plus graphic tools (box plot and stem-and-leaf representation) to detect outlying data. Then analyses of variance (PROC GLM) were performed by SAS to construct tests to determine the significance of classification effects. One typical goal was to compare the means of the response variables (DNA yield and % GM DNA) for various combinations of the classification variables. In particular, two-way analysis of variance was applied to investigate the effect on the % RRS values of the extraction method, the matrix and the interaction between them. For DNA measurements, a three-way analysis was performed to include the effect of the method for determining the DNA concentration (fluorometric or spectrophotometric). The basic factors were arranged in a fully crossed experiment design (see Table 1). The Levene test was used to assess departures from homogeneous variances. One-factor conventional and Welch's analyses of variance were also run. Welch's non-parametric approach is robust to departures from non-homogeneity of variance and is assumed to be more reliable than conventional analysis of variance under such conditions. The Ryan-Einot-Gabriel-Welch multiple range test (REGWQ) was used for post-hoc comparisons for overlapping groups of means.

Table 1

DNA measurements and GMO values (% RRS) for all DNA extraction method/matrix combinations.

Extraction method	Matrix	260:280 nm ratio	260:230 nm ratio	Spectrophotometric DNA yield(μg g-1)	Inhibition run	% RRS	Fluorescence DNA yield(μg g-1)
CTAB	Feed	1.89	2.42	114.0	Ok	0.90	84.5
	Feed	1.87	2.39	133.5	Ok	0.88	101.0
	Feed	1.88	2.34	127.0	Ok	0.78	99.0
	Feed	1.86	2.42	121.0	Ok	0.63	94.0
	Feed	1.87	2.35	144.0	Out of AC*	0.82	114.0
	Feed	1.89	2.42	103.5	Ok	1.11	80.0
	Biscuit	1.6	1.64	28.5	Ok	1.74	7.5
	Biscuit	1.38	1.32	12.5	Ok	1.55	3.0
	Biscuit	1.41	1.17	10.0	Out of AC	2.04	3.0
	Biscuit	1.63	1.63	25.5	Ok	1.19	6.5
	Biscuit	1.58	1.71	24.5	Ok	1.40	7.5
	Biscuit	1.3	1.10	10.0	Ok	1.32	2.5
	CRM 1%	1.67	1.87	69.5	Ok	1.31	56.0
	CRM 1%	1.70	1.90	45.5	Ok	1.34	38.5
	CRM 1%	1.73	2.15	65.0	Ok	1.01	59.0
	CRM 1%	1.69	1.92	43.0	Ok	1.14	36.0
	CRM 1%	1.69	1.95	41.0	Ok	1.45	47.5
	CRM 1%	1.73	2.05	67.0	Ok	1.09	81.0
Wizard	Feed	1.70	0.87	83.0	Out of AC	1.31	32.5
	Feed	1.77	0.67	105.5	Out of AC	0.96	38.0
	Feed	1.66	0.92	118.0	Out of AC	1.50	33.5
	Feed	1.86	0.27	70.0	Ok	1.07	37.5
	Feed	1.79	1.05	49.0	Out of AC	1.35	24.8
	Feed	1.72	1.10	68.0	Out of AC	1.35	28.3
	Biscuit	1.77	0.39	92.0	Ok	3.29	4.1
	Biscuit	1.84	0.13	11.5	Ok	3.06	1.7
	Biscuit	1.86	0.18	30.5	Ok	4.90	2.8
	Biscuit	1.83	0.34	64.0	Out of AC	3.15	2.6
	Biscuit	1.8	0.31	84.0	Ok	2.03	3.2
	Biscuit	1.86	0.17	62.5	Ok	2.11	3.4
	CRM 1%	1.59	0.23	62.0	Out of AC	1.04	3.7
	CRM 1%	2.02	0.21	46.5	Ok	1.45	10.9
	CRM 1%	1.98	0.17	56.5	Ok	0.58	4.1
	CRM 1%	2.09	0.04	14.5	Ok	1.08	5.4
	CRM 1%	1.87	0.27	86.5	Out of AC	1.34	9.6
	CRM 1%	1.91	0.08	29.0	Ok	1.31	12.2
Dellaporta	Feed	2.19	0.18	72.0	Out of AC	0.99	10.3
	Feed	2.05	0.27	111.0	Ok	0.90	24.8
	Feed	2.07	0.27	107.5	Ok	0.94	22.8
	Feed	1.94	0.61	294.0	Out of AC	0.80	61.0
	Feed	2.17	0.2	78.5	Ok	0.86	14.8
	Feed	2.75	0.09	34.5	Ok	0.71	7.8
	Biscuit	2.27	0.11	62.5	Ok	1.15	4.5
	Biscuit	2.87	0.05	29.0	Ok	1.13	4.2
	Biscuit	2.39	0.08	43.5	Ok	1.26	3.0
	Biscuit	2.83	0.05	29.5	Ok	1.40	4.0
	Biscuit	2.58	0.06	32.0	Ok	0.96	3.0
	Biscuit	2.58	0.06	34.5	Ok	1.01	4.0
	CRM 1%	1.90	2.11	74.0	Ok	1.03	42.5
	CRM 1%	1.91	2.25	78.5	Ok	1.14	33.0
	CRM 1%	1.71	1.89	21.5	Out of AC	0.93	15.0
	CRM 1%	1.58	0.73	40.5	Out of AC	1.06	17.7
	CRM 1%	1.78	4.25	45.0	Ok	1.25	22.2
	CRM 1%	1.84	1.94	20.0	Out of AC	1.25	2.8

* 'Out of AC' stands for data outside the acceptance criteria for DNA quality from inhibition runs

For % GM DNA quantification, statistical results were generated from the analyses described above by either including or not including the data not complying with the DNA quality criteria recommended by the EURL-GMFF.

For feed and biscuit, the true % GM content was unknown and therefore estimated based on the log-normal distribution for the variation associated with PCR-based measurements [e.g. [26]]. For each matrix and DNA extraction method, two ISO validation metrics (average percent bias - B - and percent repeatability standard deviation - RSD) were calculated and, together with PCR amplification efficiency rates (E- reference gene efficiency and E- target gene efficiency), combined into a modular, synthetic fuzzy-based indicator. In particular, two aggregation modules were defined - accuracy and efficiency - combining validation metrics and efficiency rates respectively. In both cases, the basic measures were given equal importance (expert weight equal to 0.50) in terms of affecting the relevant module. The two modules were then aggregated into an indicator giving greater importance to accuracy (0.75) than efficiency (0.25). This approach relies on the rationale and expert settings outlined by [27] and use of AMPE software for the calculations [28].

Results

A programme of experiments was designed by the ISS (Rome, Italy) and EURL-GMFF (Joint Research Centre, European Commission, Ispra, Italy) to evaluate the effect of alternative DNA extraction methods applied to matrices on determination of the content of Roundup Ready® soybean (RRS, line GTS-40-3-2) relative to total soybean DNA (see Figure 1).

DNA extracted from the three matrices (biscuit, feed and CRM 1%) according to three mentioned DNA extraction methods were first analysed by agarose-gel electrophoresis, where two microlitres of the samples were loaded and analysed by comparison with molecular size markers (see Figure 2). This way, a comparison of the nucleic acids extracted by the same extraction procedure from the different matrices, i.e. biscuit, feed and CRM 1%, is readily available. DNA from complex matrices such as biscuit and feed show signs of extensive degradation with all the procedures. Commonly used DNA isolation methods such as 'CTAB' and 'Della Porta' provide higher DNA yield when applied to feed material and, in decreasing order, to CRM 1% and biscuit.

Figure 2

Agarose-gel electrophoresis of DNA extracts by matrix and DNA extraction method. M, λ DNA/EcoRI + HindIII.

Roundup Ready® soybean amplicon has a length of 84-bp, whilst the soybean-specific reference amplicon - based on a lectin gene - has a length of only 74-bp http://gmo-crl.jrc.ec.europa.eu/summaries/40-3-2_validated_Method.pdf, thus in the present study average length of DNA fragments are well above the target amplicon size, though less visible for Wizard extracts from CRM 1% matrix; this ensures amplifiability of target analytes as demonstrated by amplification data from inhibition runs (outcomes shown) and quantification (% RRS) of the target analyte in qPCR (Table 1). Data on the DNA concentration (by spectrophotometer and PicoGreen) and GMO quantification (determination of % RRS) for three matrices (feed, biscuit and CRM 1%) for three DNA extraction methods (CTAB, Wizard and Dellaporta) are summarised in Table 1, which also shows the absorbance ratios (260:280 and 260:230) and outcomes from inhibition runs. DNA yields are given as weight/weight ratios between DNA and the matrix.

For several samples the absorbance ratios fall outside the criteria adopted as an indication of DNA purity (260:280 between 1.8 and 2.0; 260:230 ≥ 2.0). CTAB/feed was the only combination in which impurities from proteins were not found. Wizard showed far less than acceptable absorbance ratios in the nucleic acid preparations, thus pointing to consistent impurities with any matrix. With Dellaporta, impurities were observed, especially when applied to feed and biscuit matrices. In general, clues to impurity from absorbance ratios tend not to be related to the acceptance criteria for inhibition runs.

Inhibitory effects were revealed in more than 40% of the cases with Wizard, in about 30% with Dellaporta and in about 10% with CTAB. Similar percentages were observed with matrices, feed being the most inhibited and biscuit the least. The worst combination was Wizard extraction from a feed matrix (five out of six samples inhibited). No inhibitory effects were observed with either CTAB on CRM 1% or with Dellaporta on biscuit.

DNA yield (μg g-1), estimated from spectrophotometric determination of the DNA concentration, showed a moderate departure from normal (P < 0.05) only with the Dellaporta/feed combination, with some bias towards high values (positive asymmetry). Estimates of RRS concentration in DNA extracts (% RRS in Table 1) were normally distributed with all the tests (P > 0.05), but some violations of the assumption of equal variance were identified (Levene test, P < 0.01).

With both DNA yield and DNA concentration, comparisons with one-way conventional analysis of variance revealed significant differences (P < 0.01) between both factors. The same responses were confirmed by non-parametric Welch's analysis of variance. Hence, in spite of heterogeneous variances, conventional one-way ANOVA and the non-parametric approach proved equally reliable. For this reason, the conventional parametric approach was applied in the fully crossed three-way (DNA yield) and two-way (% GM DNA) analyses of variance (Tables 2 and 3, respectively).

Table 2

DNA yield - three-way analysis of variance

Source of variation	Degrees of freedom	F value	Probability
Total	107
Extraction (of DNA)	2	5.7	0.0047
Matrix	2	44.2	< 0.0001
Concentration (of DNA)	1	54.3	< 0.0001
Extraction/Matrix	4	5.6	0.0005
Extraction/Concentration	2	5.1	0.0079
Matrix/Concentration	2	3.8	0.0257
Extraction/Matrix/Concentration	4	1.38	0.2484
Residual	90

Table 3

% GMO (% RRS): two-way analysis-of-variance

Source of variation	All data			Excluding 'out of AC'* data
	Degrees of freedom	F value	Probability	Degrees of freedom	F value	Probability
Total	53			38
Extraction (of DNA)	2	19.49	< 0.0001	2	4.94	0.0140
Matrix	2	28.41	< 0.0001	2	13.45	< 0.0001
Extraction/Matrix	4	11.29	< 0.0001	4	6.69	0.0006
Residual	45			31

* 'Out of AC' stands for data outside the acceptance criteria for DNA quality from inhibition runs

DNA yield

The mean range of variability in determination of the DNA yield by spectrophotometric and fluorometric measurements (see Figure 3) reflects the ability of the spectrophotometer to react with considerably higher amounts of solutes (from an average of 18.5 μg g-1 in the CTAB/biscuit combination to an average of 116.3 μg g-1 for Dellaporta/feed) than the fluorometer (from an average of 3.0 μg g-1 for Wizard/biscuit to an average of 95.4 μg g-1 for CTAB/feed).

Figure 3

Mean DNA yield (and standard error) determined for each combination of matrix (feed, biscuit or CRM 1%) and DNA extraction method (CTAB, Wizard or Dellaporta).

Based on analysis-of-variance results (see Table 2), the estimated DNA yield is strongly affected by any of the factors (DNA extraction method, matrix and method used to estimate DNA concentration) at the levels employed in the experiment (P < 0.01).

The results in Table 2 show that two-way interactions are either highly significant (P < 0.01 for extraction/matrix and extraction/method for DNA concentration) or marginally significant (0.01