Evaluation of Unintended Effects in the Composition of Tomatoes

Jul 28, 2014 - allergens cannot be dismissed a priori and need to be investigated. This paper describes a collection of independent transgenic tomato ...
0 downloads 0 Views 3MB Size
Download PDF
Article pubs.acs.org/JAFC

Evaluation of Unintended Effects in the Composition of Tomatoes Expressing a Human Immunoglobulin A against Rotavirus Paloma Juarez, Asun Fernandez-del-Carmen, Jose L. Rambla, Silvia Presa, Amparo Mico, Antonio Granell, and Diego Orzaez* Instituto de Biología Molecular y Celular de Plantas, Consejo Superior de Investigaciones Cientı ́ficas, Universidad Politécnica de Valencia, Camino de Vera s/n, 46022 Valencia, Spain S Supporting Information *

ABSTRACT: The production of neutralizing immunoglobulin A (IgA) in edible fruits as a means of oral passive immunization is a promising strategy for the inexpensive treatment of mucosal diseases. This approach is based on the assumption that the edible status remains unaltered in the immunoglobulin-expressing fruit, and therefore extensive purification is not required for mucosal delivery. However, unintended effects associated with IgA expression such as toxic secondary metabolites and protein allergens cannot be dismissed a priori and need to be investigated. This paper describes a collection of independent transgenic tomato lines expressing a neutralizing human IgA against rotavirus, a mucosal pathogen producing severe diarrhea episodes. This collection was used to evaluate possible unintended effects associated with recombinant IgA expression. A comparative analysis of protein and secondary metabolite profiles using wild type lines and other commercial varieties failed to find unsafe features significantly associated with IgA expression. Preliminary, the data indicate that formulations derived from IgA tomatoes are as safe for consumption as equivalent formulations derived from wild type tomatoes. KEYWORDS: tomato, IgA, antibody, rotavirus, unintended effects, metabolomics, proteomics

■

INTRODUCTION Antibody-based treatments for mucosal passive immunization (MPI) have great potential in human and veterinary medicine for the prevention of infection diseases. MPI is particularly advantageous for the treatment of enteric diseases.1−3 Generally, passive antibody-based treatments require large amounts of antibodies to be delivered to the target mucosa to ensure protection.4 The high specificity of recombinant monoclonal antibodies makes them excellent candidates for MPI; however, the increasing costs of producing recombinant antibodies in the preferred mammalian cell platforms has hampered their use in MPI. As an alternative, neutralizing recombinant antibodies aimed at MPI can be inexpensively produced in edible plant organs. Many seeds, fruits, leaves, tubers, and roots are considered safe and palatable for human consumption in minimally processed formulations without heat treatment;5 therefore, it has been proposed that antibodies produced in plant organs with generally regarded as safe (GRAS) status could be delivered as dose-controlled ingredients in partially processed formulations without the need for exhaustive purification.4 This would certainly reduce the manufacturing costs, as plant platforms are reportedly easier to scale up than other platforms, for example, those based on fermentation.5 The lack of exhaustive purification can, on the other hand, become a drawback when the composition of the final product ultimately results from an event of genetic modification. It could be argued that, besides producing a recombinant antibody, transgenesis could eventually lead to unintended effects in the final fruit composition.6,7 Such unintended effects could be derived from the integration of the transgene, from biological interactions caused by transgene-encoding proteins, © 2014 American Chemical Society

or from spurious somaclonal mutations. Unintended effects have become one of the most controversial issues in biological safety debates.8 Despite the evidence accumulated during the past 20 years indicating that transgenesis could be even less disruptive of food composition than traditional breeding,8−10 the absence of unintended, deleterious effects in the composition of genetically modified edible plants organs expressing recombinant antibodies has never been assessed. Recently, our group reported a model human immunoglobulin A (IgA) for passive protection against the enteric pathogen rotavirus, produced in transgenic tomato fruits. Dry formulations suitable for oral intake and compatible with longterm storage presented a high concentration of active antibodies (≈0.7 mg/g DW), which were shown to inhibit virus infections in an in vitro virus neutralization assay. Thus, minimally processed fruit-derived formulations containing recombinant IgA were proposed as potential vehicles for lowcost MPI treatments.11 The availability of IgA-producing fruits provides an opportunity to test for the possible unintended effects that the expression of a human antibody could have on the final composition of the fruit and to infer the possible deleterious effects on human health that such unintended effects could have. Safety concerns in tomato fruit composition arise fundamentally from two sources: proteins (allergens) and secondary metabolites (toxicants). So far, only three tomato protein allergens, Lyc e 1 (profilin), Lyc e 2 (invertase), and Lyc e 3 Received: Revised: Accepted: Published: 8158

May 16, 2014 July 18, 2014 July 27, 2014 July 28, 2014 dx.doi.org/10.1021/jf502292g | J. Agric. Food Chem. 2014, 62, 8158−8168

Journal of Agricultural and Food Chemistry

Article

quantification were developed as previously described.11 Proteomic profiling, protein precipitation, fluorescent labeling, 2D electrophoresis, and protein identification by mass spectrometry (MALDI, LC-MS/MS) were developed as previously described.27 To identify differentially regulated proteins, each 2D gel was loaded with three samples: a transgenic line, a nontransgenic line, and an internal standard (Figure 2a, blue). The internal standard was composed of a pool of all the transgenic and nontransgenic samples used in the analysis. Each of the three samples in a gel was labeled with a different CyDye (Cy2, Cy3, and Cy5). 2D gels were scanned using a Typhoon Trio (GE Healthcare). The three different images corresponding to each fluorophore were overlapped, and the spots from the different gels were matched. Gel image analysis was performed with DeCyder 2D software v. 6.5. Principal component analysis (PCA) was performed with the software package SIMCA-P 11 (Umetrics, Sweden). Sample J was not included in the PCA plot to improve visualization. The PCA plot with the complete set of samples is shown in Supplementay Figure 1 in the Supporting Information. UPLC-QTOF and Data Processing. Ultraperformance liquid chromatography−quadrupole time of flight (UPLC-QTOF) analysis of tomato powder was performed as previously described,28 with minor modifications. Following extraction in 3 volumes of cold (−20 °C) methanol acidified with 0.125% formic acid, the supernatant was dried (in two separate 750 μL aliquots) in a Speed-Vac and stored at −80 °C. Prior to analysis, dry extracts were solubilized in 150 μL of methanol−0.1% formic acid and filtered through a 0.2 μm Anotop 10 membrane filter (Whatman, Kent, UK). Samples were analyzed by UPLC-MS using an ACQUITYUPLC-PDA system coupled to a QTof micro mass spectrometer (Waters Corp., Milford, MA, USA).28 Mass spectrometry analysis was performed by electrospray ionization (ESI) in positive and negative modes (using one of the extract aliquots for each ionization mode). A reference sample (made by pooling equal amounts of all biological samples) was injected four times (at regular intervals) in each injection series. MassLynx raw files converted to netCDF format with DataBridge (Waters Corp.) were processed with XCMS package v. 1.34.0 (http://bioc.ism.ac.jp/2.11/bioc/html/xcms. html) run under R v. 2.15.2 (http://www.r-project.org/). The centWave algorithm29 was used for feature detection; peak grouping and retention time correction were performed with the density and peakgroups algorithms,30 respectively. Mass features were grouped using the CAMERA algorithm.31 The area of each extracted mass feature was first normalized to the total chromatogram area; the reference sample was then used to normalize the data between different runs (for each mass feature, the test sample area is divided by the mean area of the reference samples included in that run). Preprocessing of the data generated two discrete data files, one for each ionization mode: ESI− (electrospray ionization in negative mode) and ESI+ (electrospray ionization in positive mode). The ESI− data set contained 386 mass variables, whereas 835 peaks were initially identified in the ESI+ mass profile. To remove low-quality data, a first filtering step was done by imposing two different conditions: (i) a mass feature should be detectable in a minimum of three biological replicates in at least one of the tomato lines, and (ii) the average-day coefficient of variation for the reference sample (computed on normalized data) should not be greater than 40%. Mass peaks that did not fulfill these requirements were considered unreliable and were excluded from the study. A second filtering step was accomplished by removing the redundant isotopic features, which were identified by the package CAMERA (a collection of algorithms for metabolite profile annotation with the primary purpose of annotation and evaluation of isotope peaks, adducts, and fragments in peak lists).31 The final ESI− and ESI+ data sets, with 107 and 246 peaks, respectively, were joined together in a single matrix for further analysis. Statistical Data Analysis. PCA and OPLS-DA of the protein and metabolite profiles of tomato fruits were performed with the software package SIMCA-P 11 (Umetrics). Variables were log-transformed, mean centered, and scaled to unit variance prior to analysis. Twosample t tests on log transform data were performed using MeV v4.9 (http://www.tm4.org/mev.html), an application of the TM4Microarray Software Suite.32 Safety thresholds in the proof-of-safety analysis

(nonspecific lipid transfer protein), are listed in the official International Union of Immunological Societies (IUIS) allergen database, although additional potential allergens have been reported.12−16 To investigate the changes in the fruit protein profile associated with the overexpression of recombinant proteins, a proteomic analysis is required. Proteomics is a powerful nontargeted tool that can be used to detect unintended differences derived from genetic manipulation. Proteomics-based approaches have been used a number of times to compare protein profiles of transgenic maize,17 tomato,18 or potato19,20 with their nontransgenic counterparts.8 A second possible source of concern is the presence of toxic metabolites. Fruits are rich in chemically diverse compounds, which can be present in a wide range of different concentrations. The most important toxicants in tomatoes are the steroidal glycoalkaloids α-tomatine and dehydrotomatine. Tomato glycoalkaloids are synthesized in tomato fruits during early development and then degraded during fruit maturation.21−24 There is no analytical method capable of extracting and detecting all metabolites; however, in the past two decades, several methods have been established for large-scale analysis and comparison of metabolites in plant extracts.8,25 Among them, ultraperformance liquid chromatography−mass spectrometry (UPLC-MS) specifically allows the detection and quantification of semipolar secondary metabolites, which is the major source of potential toxicants. The goal of this work was to examine a broad set of components in the fruit to search for unintended effects produced by the expression of a human IgA, which could eventually affect the safety of minimally processed formulations derived from IgA tomatoes. First, a protein profile analysis was performed to detect possible changes associated with the production of human IgA in the fruit. Next, the metabolomic profile of IgA tomatoes was compared to that of nontransgenic tomatoes to detect possible unintended effects in the semipolar secondary metabolite pool.

■

MATERIALS AND METHODS

Plant Material. Seven transgenic lines (A, B, D, F, G, H, and I) of tomato (Solanum lycopersicum cv. Moneymaker) expressing human IgA were used in this study along with the parental line, K, used for genetic transformation. Lines A, B, and D were obtained by Agrobacterium-mediated transformation of a plasmid carrying the heavy (HC) and light chains (LC) of human IgA under the cauliflower mosaic virus 35s constitutive promoter and the tomato NH fruit promoter,26 respectively. Lines F, G, H, and I were obtained by cotransformation with two Agrobacterium cultures, one carrying a plasmid with the 35s:IgA_HC and the second carrying a plasmid with the 35s:IgA_LC construct. A detailed description of the generation of these transgenic lines can be found in Juarez et al.11 Additionally, two nullizygote segregant lines, C and E, from the 35S:IgA_HC/ NH:IgA_LC transformation, and one wild type moneymaker line, J, obtained from a different laboratory were also included as additional controls. Finally, fruits from three widely consumed tomato varieties, plum (L), round (M), and cherry (N), purchased from a local grocery completed the experimental data set. Tomato seeds (excluding commercial lines L, M, and N) were grown in pots in a greenhouse with a light/dark cycle of 16/8 h. Fruits were harvested 15 days postbreaker. Five independent samples from each plant line were collected. After removal of the gel and seeds, tomato pericarp (with the peel) was ground to a fine powder in liquid nitrogen and stored at −80 °C for further analysis. Protein Analysis. For antibody analysis, frozen tomato powder samples from each group were pooled together. Total soluble protein (TSP) extraction, Western blots, and ELISA tests for IgA 8159

dx.doi.org/10.1021/jf502292g | J. Agric. Food Chem. 2014, 62, 8158−8168

Journal of Agricultural and Food Chemistry

Article

were defined by the symmetric maximum boundaries of the nonparametric 90% confidence intervals for the ratio to WT of the commercial variety farthest away from the WT. Safety was concluded when the confidence intervals for the ratio to control of the transgenic lines were within the safety limits. For the calculations of the confidence intervals the R package pairwise CI was used.

homozygous 35S_IgA line (line I) showed the highest IgA levels, with an average content of 37.5 μg/g FW. The IgA content of the remaining transgenic lines ranged from 25.5 to 1.9 μg IgA/g FW (see Table 1). To check the integrity of the IgA in the powder samples, 20 transgenic fruits (4 pools of 5 fruits) were analyzed by Western blot. As expected, specific bands corresponding to the HC (55 kDa) and the LC (28 kDa) were detected in all of the transgenic samples analyzed. A major degradation HC fragment band of 28 kDa HC was also observed in blots developed with anti-HC antibody. This degradation corresponds to the HC domains of reduced fragment antigen-binding (Fab) fragments and was earlier characterized by Juarez et al.11 In all cases the relative expression levels observed in Western blots were consistent with the ELISA quantification. Neither HC nor LC bands were detected in nontransgenic control samples (Figure 1).

■

RESULTS Evaluation of IgA Content in the Fruit. To detect unintended changes in fruit composition, two types of IgAproducing plants were selected for analysis. The first group was generated in cotransformation experiments (referred to as 35S_IgA) in which tomato explants were simultaneously cotransformed with two Agrobacterium cultures carrying HC and LC antibody chains, respectively, both of them expressed under the control of constitutive 35S promoter. In the second group (referred to as NH_IgA), both HC and LC were linked in the same T-DNA, with LC placed under the control of the tomato NH promoter.26 In the second generation (T2), different segregating lines were obtained. Some NH_IgA segregant lines lacking the transgene (nullizygotes) were selected to serve as appropriate negative controls in subsequent assays. All cotransformed 35s-IgA segregants showed at least one of the two antibody chains. As a consequence, no 35S_IgA nullizygote could be included in the analysis. In total, seven independent transgenic lines from both groups (NH_IgA and 35s_IgA) showing a broad range of IgA levels were used in the different comparative analyses (see Table 1). This includes an Table 1. IgA Levels Referred to Fresh Weight (FW) and to Total Soluble Protein (TSP) in Tomato Fruit in All Lines ID

line

A B C D E F G H I J K L M N

NH-IgA NH-IgA NH-IgA nullizygote NH-IgA NH-IgA nullizygote 35s_IgA 35s_IgA 35s_IgA 35s_IgA homozygous Moneymaker seed batch 1 Moneymaker seed batch 2 plum round cherry

total IgA (μg/gFW) 2.5 5.7 0 1.9 0 13.0 25.5 20.9 37.5 0 0 0 0 0

± 0.3 ± 1.6 ± 0.3 ± ± ± ±

2.2 6.5 5.2 5.2

total IgA (% TPS) 0.2 0.4 0 0.2 0 1.0 2.0 1.6 2.9 0 0 0 0 0

Figure 1. Western blot analysis of IgA expression in tomato fruit: (a) Western blot under reducing conditions, developed with anti-heavy chain antibody (HC); (b) Western blot under reducing conditions developed with anti-light chain antibody (LC). Lanes: A and D, NH_IgA lines; F and I, 35S_IgA lines; C and E, nullizygote; J and K, wild type Moneymaker samples from two different seed batches; HS, human serum sample.

± 0.1 ± 0.1 ± 0.1 ± ± ± ±

0.2 0.5 0.4 0.4

Identification of Differentially Expressed Proteins in IgA Tomatoes by 2D-DIGE and LC-MSMS. The first assessment of the fruit content was directed to the analysis of the protein composition by 2D-DIGE. A total of eight tomato lines were included in the analysis: four selected IgA lines covering the whole IgA expression range were paired with samples from four nontransgenic lines, which included two inhouse-grown Moneymaker plants as well as two NH_IgA nullizygote plants. The NH_IgA tomato samples (A and D from Table 1) were paired with their corresponding nullizygote samples (C and E), in two gels. The other two gels were loaded with 35s_IgA samples (F and I), each one in combination with a different wild type sample (J and K). A total of 1001 spots were located in the four gels. Differential proteins are shown as green spots (with higher levels in transgenic line) or red spots (with higher levels in the control line) (Figure 2). All spots were quantified, and the resulting data were used for the comparative analysis between the composition of the different samples. A principal component analysis (PCA) was performed with the quantitative data extracted from the quantification of protein spots in the gels (Figure 3). The first two principal components of this PCA model explain 28 and 23% of the total

“elite” T4 homozygous IgA line (line I) previously selected from the 35S_IgA group for its high antibody levels. A number of IgA-free control lines were also included, which comprised (i) the above-mentioned nullizygote plants from the segregating NH_IgA lines, (ii) two wild type Moneymaker lines from different seed batches, and (iii) three commercial tomato varieties purchased from a local supermarket (L, M, and N). Tomato plants of all noncommercial lines were grown in the same greenhouse conditions. Fruits were collected 15 days after breaker stage, and frozen tomato powder of each fruit was prepared. The first analytical step consisted on the determination of the antibody content of the different lines. For this, frozen tomato powder samples were extracted in PBS buffer, and the IgA content in the resulting crude extracts was quantified by means of ELISA. As expected, the “elite” 8160

dx.doi.org/10.1021/jf502292g | J. Agric. Food Chem. 2014, 62, 8158−8168

Journal of Agricultural and Food Chemistry

Article

Figure 2. Representative 2D DIGE gel of tomato fruit proteins: (a) fluorescence images of gel loaded with equal amounts (50 μg) of the reference sample (Cy2-labeled, blue), nontransgenic samples (Cy5-labeled, red), and transgenic samples (Cy3-labeled, green); (b) overlay of the three fluorescence images (differential proteins appear in green or red; unaffected proteins appear in white); (c) silver-stained 2D DIGE gel (spots showing significant differences between transgenic and nontransgenic samples are outlined in green).

spots showing average ratio higher than 2 (|ratio| > 2), only 19 of them showed an increase in both 35s_IgA and NH_IgA lines when considered separately, and only 5 of these 19 were found statistically significant (P < 0.05) (Figure 2c, outlined in green). The identification of all 19 selected spots by LC-MS/MS is shown in Table 2. All spots were identified as HC or LC with the exception of spot 783, which corresponded to a mix of human HC and tomato ACC oxidase. Metabolic Profiling by UPLC-MS for the Evaluation of Unintended GM Effects. Unintended effects on the metabolome of transgenic tomato fruits generated by the expression of human IgA were also evaluated by nontargeted metabolomics. Tomato methanol extracts were used for the analysis, which was carried out by means of reverse-phase UPLC coupled to mass spectrometry (UPLC-QTOF). This technology is especially suited for the detection of semipolar secondary metabolites, primarily tomato polyphenols and alkaloids.33 Four or five ripe fruits from seven transgenic lines containing different levels of IgA, two nullizygote plants, two wild type Moneymaker lines, and three commercial varieties (L, M, and N) were analyzed. The mass profile obtained was first processed with XCMS, an integrated metabolomics analysis platform for the determination of metabolic profile differences and metabolite identification.30 Data processing and filtering generated a total of 356 reliable peaks, which were introduced in a single matrix for analysis. A PCA was performed as the first step in data exploration to examine the relationship between the different sample classes. The first two components of the PCA model explained 30 and 17% of the total variance, respectively. As can be seen in the scores scatter plot depicted in Figure 4a,

Figure 3. PCA of proteomic data of transgenic and nontransgenic tomato lines. Capital letters represent the different tomato lines as described in Table 1. Sample J, which behaves as an outlier, is not included in the plot to improve visualization.

variance, respectively. PC1 slightly separates 35s_IgA lines (F and I) from the rest of the group, which remain together in the scores space. PC2 further discriminates between both 35s_IgA lines (F and I). Interestingly, all samples derived from NH_IgA transformation events (both transgenic and nullizygote) remain closely grouped in the PCA model. To identify those spots that are the main contributors for class separation, a quantitative comparison between transgenic and nontransgenic lines was performed. From a total of 37 8161

dx.doi.org/10.1021/jf502292g | J. Agric. Food Chem. 2014, 62, 8158−8168

Journal of Agricultural and Food Chemistry

Article

Table 2. Identified Proteins with Ratio >2 spot P < 0.05

Paragon score

266 844 885 849 855 spot P > 0.05

17.6 22.5 6.1 12.6 14.7 Paragon score

% cov

843 896 783

6.3 6.2 2 1.7 9.0 2.4 8.1 9.3 3.6 2.0 10.1 2 2.37 2 1.82

34.3 22.4 52.3 10.5 17.9 10.5 45.3 45.3 23.6 17.0 38.0 10.5 10.5 15.5 23.6

792 797 852 868 869 875 886 899 956 1186 1169

% cov

accession

name

29.8 41.1 19.2 42 39.7

Sp|P01876|IGHA1_HUMAN Sp|B9A064|IGLL5_HUMAN Sp|B9A064|IGLL5_HUMAN LV301_HUMAN Sp|B9A064|IGLL5_HUMAN accession

Ig α-1 chain C region immunoglobulin λ-like polypeptide 5 immunoglobulin λ-like polypeptide 5 Ig λ chain V−III region SH immunoglobulin λ-like polypeptide 5 name

sp|P01714|LV301_HUMAN sp|P01876|IGHA1_HUMAN sp|P10967|ACCH3_SOLLC sp|P01876|IGHA1_HUMAN gi|46561796 sp|P01876|IGHA1_HUMAN sp|P0CG04|LAC1_HUMAN sp|P0CG04|LAC1_HUMAN sp|P0CG04|LAC1_HUMAN sp|P0CG04|LAC1_HUMAN gi|21669631 sp|P01876|IGHA1_HUMAN sp|P01876|IGHA1_HUMAN gi|7438718 sp|P0CG04|LAC1_HUMAN

Ig λ chain V−III region SH Ig α-1 chain C region OS=Homo sapiens 1-aminocyclopropane-1-carboxylate oxidase Ig α-1 chain C region rpL23−ScFv fusion protein Ig α-1 chain C region Ig λ-1 chain C regions Ig λ-1 chain C regions Ig λ-1 chain C regions Ig λ-1 chain C regions immunoglobulin λ light chain VLJ region Ig α-1 chain C region Ig α-1 chain C region Ig λa chain NIG250 precursor Ig λ-1 chain C regions

species

peptides (95%)

human human human human human species

9 30 3 4 18 peptides (95%)

human human Solanum lycopersicum human

4 6 1 1 4 1 4 4 2 1 5 1 1 1 1

human human human human human human human human human human

Figure 4. PCA scores plots of LC-MS spectra: (a) complete data set of tomato cultivars including transgenic lines expressing IgA, nullizygote lines, wild type lines, and commercial lines; (b) PCA including only Moneymaker lines grown in-house. Each point represents a biological replicate (pool of two to three different fruits). Capital letters represent the different tomato lines as described in Table 1.

PC1 clearly separated the commercial varieties from the remaining samples, whereas PC2 allowed further discrimination within the commercial group between N and L and M lines. The samples from IgA-producing lines and control nullizygote and Moneymaker wild type lines, however, grouped together in the scores space, not being separated by any of the components of the PCA model. As the largest variance in the data set was observed between commercial lines purchased in a supermarket and Moneymaker lines, a new PCA was performed in which commercial lines were excluded from the analysis to thus better explore the differences associated with IgA production. The obtained PCA model (Figure 4b) did not reveal any clear group structure in the data set; only 35S_IgA lines F, G, and H slightly separated from the remaining samples, whereas

significant confusion occurred between the homozygous 35S_IgA line I, NH_IgA lines, and the control lines. To directly search for specific metabolites that discriminate between IgA and control lines, an orthogonal partial leastsquares discriminate analysis (OPLS-DA) was set. OPLS-DA makes use of a priori information on class membership (e.g., IgA content) to extract class-related variance in a predictive component and class-uncorrelated variation in one or more orthogonal components. Thus, even in the presence of uncontrolled covariates, OPLS-DA allows a direct identification of class-discriminant variables. As 35s_IgA lines and NH_IgA lines contain very different IgA levels,11 they were treated separately so that each transgenic group could be compared with its closest control (non-IgA) group. Thus, three separate 8162

dx.doi.org/10.1021/jf502292g | J. Agric. Food Chem. 2014, 62, 8158−8168

Journal of Agricultural and Food Chemistry

Article

Figure 5. OPLS-DA score plots of LC-MS spectra: (a) 35S_IgA versus WT; (b) NH_IgA versus WT; (c) NH_IgA versus nullizygote. Each point represents a biological replicate (pool of two to three different fruits). Capital letters represent the different tomato lines as described in Table 1.

masses is provided in Table S1 as Supporting Information. The three models were then compared by means of shared and unique structures (SUS) plots, an analysis that combines the p(corr) profiles from two models, constructed with all of the selected variables (Figure 6). The first SUS plot showed the correlation between models 35s_IgA versus WT and NH_IgA versus WT (Figure 6a). The second SUS plot correlated models 35s_IgA versus WT and NH_IgA versus nullizygote (Figure 6b). The last SUS plot showed the correlation between models NH_IgA versus WT and NH_IgA versus nullizygote (Figure 6c). Only two discriminating ion masses were shared by the three models, namely, masses Mp232 and Mn44. In general, there was little overlap between the 35s_IgA versus WT and NH_IgA versus nullizygote models, except the two mentioned masses, Mp232 and Mn44, sharing high correlation loadings and covariance. Comparison of 35s_IgA versus WT/NH_IgA versus WT and NH_IgA versus WT/NH_IgA versus nullizygote demonstrated greater correspondence, with nine and six common discriminating features, respectively (Figure 6d). To determine the statistical significance of the differences in the metabolic profiles observed in the OPLS-DA, a univariate analysis was also carried out. A t test was performed for the comparison between NH_IgA versus WT, NH_IgA versus nullizygote, and 35s_IgA versus WT. Seventeen masses, 4.8%

comparisons were made involving 35s_IgA versus WT, NH_IgA versus WT, and NH_IgA versus nullizygote lines (Figure 5). The fitted models comprised one predictive (IgA content) plus one, two, and one orthogonal components, respectively. The variance related to class separation explained by the models was 82, 98, and 84% with a cross-validated predictive ability Q2(Y) = 0.31, 0.84, and 0.40, indicating a higher uncertainty in the 35s_IgA versus WT and NH_IgA versus nullizygote models regarding class separation. The homozygous line I formed a separate subgroup within the 35s lines as revealed by the first orthogonal component of the model. The most relevant metabolite ions for class discrimination were selected, according to the following rules: the selected ions should present a high variable influence on projection (VIP ≥ 1), a high correlation with IgA content (|p(corr)| ≥ 0.5), and narrow jack-knifed 95% confidence intervals of the variable coefficient (confidence interval not crossing zero). Totals of 41, 39, and 27 discriminating mass peaks (counting 91 different masses) were chosen according to these criteria from the 35s_IgA versus WT, NH_IgA versus WT, and NH_IgA versus nullizygote models, respectively. A large proportion of these peaks, however, showed a very modest variation between the transgenic lines and their counterpart controls with