Authentication of dried distilled grain with solubles (DDGS) by fatty acid and volatile profiling.

Demand for ethanol substituted fuels from the utilisation of cereal based biofuel has resulted in an over production of dried distillers grains with solubles (DDGS) that are now readily available on the animal feed market. With this rapid emerging availability comes potential variability in the nutritional value of DDGS and possible risks of feed contaminants. Subsequently, the authentication and traceability of alternative animal feed sources is of high priority. In this study and as part of the EU research project "Quality and Safety of Feeds and Food for Europe (QSAFFE FP7-KBBE-2010-4) an attempt was made to classify the geographical origin of cereal grains used in the production of DDGS material. DDGS material of wheat and corn origin were obtained from Europe, China, and the USA. Fatty acid profiles and volatile fingerprints were assessed by gas chromatography flame ionisation (GC-FID) and rapid proton transfer reaction mass spectrometry (PTR-MS) respectively. Chemometric analysis of fatty acid profiles and volatile fingerprints allowed for promising classifications of cereals used in DDGS material by geographical and botanical origin and enabled visual representation of the data. This objective analytical approach could be adapted for routine verification of cereal grains used in the production of DDGS material.

Introduction

Emerging low cost animal feed stock, originating from bioprocessing and food production is increasing on a global scale. In particular, DDGS are obtained from ethanol distillation by drying solid residues of fermented cereal grains with the addition of pot ale syrup or evaporated spent wash. DDGS present a high concentration of energy, protein and ruminally undegradable protein, especially with respect to the original grains (Belyea et al., 2010; Rasco et al., 1987). This, together with low cost has promoted their use as a feed ingredient for several animal species (Aldai, Dugan, Rolland, & Kramer, 2009; Liu, 2011; Stein & Shurson, 2009; Swiatkiewicz & Koreleski, 2008). Despite the original source, from alcoholic distillation, DDGS are now increasingly available from the emerging biofuel industry (Holffman & Baker, 2011). With this new source of availability comes potential risks of feed contaminants (Liu, 2011). Subsequently, the global expansion of DDGS material has raised concerns for the animal feed industry, resulting in potential implications for compliance with regulatory agencies. DDGS are mainly obtained from corn (especially in the USA), followed by wheat. DDGS may also originate from other grains such as barley, sorghum and rice, as well as from grain blends (Aldai et al., 2009; Liu, 2011; Nuez-Ortin & Yu, 2009). One of the main problems hindering DDGS use as animal feed is their high variation in nutrient concentrations and nutritional quality among different sources. The inconsistency in nutrient value can contribute to imbalanced feed formulations leading to diminished animal productivity (Belyea et al., 2010; Belyea, Rausch, & Tumbleson, 2004). Apart from the effects on animal growth performance and other productivity parameters, it is well-known that feed FA content influences tissue FA composition, which highly determines the nutritional properties of animal products for human consumption (Bou, Codony, Tres, Decker, & Guardiola, 2009).Therefore, controlling all factors that contribute to DDGS composition (such as grain botanical source, geographical origin and production method) could help deliver a high quality standard for DDGS and ensure their market value. The authentication of botanical and geographical origin relies on the recognition of several markers typical of the variety of grains used in production and the country they were grown. However the natural variability of chemical composition prevents having one discriminative marker for any specific material. As an alternative strategy, fingerprinting techniques are nowadays used for authentication purposes (Bosque-Sendra, Cuadros-Rodriguez, Ruiz-Samblas, & de la Mata, 2012). Most studies on DDGS composition have evaluated DDGS nutrient composition (protein, oil, ash, carbohydrates, fibre contents) (Belyea et al., 2010; Liu, 2011; Nuez-Ortin & Yu, 2009; Rasco et al., 1987) but only a few have addressed its detailed chemical profiles (Aldai et al., 2010; Moreau, Liu, Winkler-Moser, & Singh, 2011; Winkler-Moser & Breyer, 2011; Winkler-Moser & Vaughn, 2009). FA fingerprinting has proven to be suitable for discriminate among wheat grain varieties, origin and year of harvest (Armanino, De Acutis, & Rosa Festa, 2002), but also to verify the authenticity of numerous food stuffs in relation to several identity aspects (Tres, Ruiz-Samblas, van der Veer, & van Ruth, 2013; Tres, van der Veer, Perez-Marin, van Ruth, & Garrido-Varo, 2012; Ulberth & Buchgraber, 2000). These previous findings, together with the fact that different DDGS sources led to variations in the FA profile of animal tissues and rumen fluid (Aldai et al., 2010, 2012) suggest FA composition as a promising strategy for the DDGS classification. In addition, the FA profile coupled with other important food constitutes such as volatile profiles is one promising approach capable of distinguishing high fat containing foods by geographical origin and/or botanical identity (Araghipour et al., 2008; Luykx & van Ruth, 2008; Tres et al., 2013; van Ruth, Rozijn, et al., 2010; van Ruth, Villegas, et al., 2010). Consequently, the aim of this study is to authenticate the botanical and geographical origin of DDGS by FA and volatile fingerprinting. This, together with the fact that the determination of FA composition is a quite common analysis in food laboratory, makes it a promising technique to help verify DDGS identity.

Material and methods

Samples

A total of 82 DDGS samples were collected in the framework of the QSAFFE EU funded project (QSAFFE FP7-KBBE-2010-4). DDGS samples were collected from six different countries (USA, Canada, France, Spain, Poland, and the Chinese provinces Jilin and Heilongjiang) between 2011 and 2012, they originated from two different botanical species (corn and wheat) and from the production of biofuel and alcoholic beverages (Supplementary Table 1). In addition a set of samples of unknown geographical origin were also obtained. All samples were pre-ground with a centrifugal mill (ZM 200, Retsch, Germany) using a mesh size of 0.5 mm. Prior to analysis, samples were stored in capped plastic vials, which were kept in the dark at −20 °C.

Reagents and standards

Sodium methoxide (0.5 N) was purchased from Sigma–Aldrich (St Louis, MO). Boron trifluoride methanol complex (35%) was obtained from Merck (Darmstadt, Germany). The FA methyl ester mixture of standards was supplied by Supelco (Supelco 37 Component FAME mix, Supelco, St. Louis, MO). All the other reagents were of ACS quality grade.

Fatty acid composition

Fat was extracted from DDGS with chloroform:methanol (2:1, v/v) as described in Tres and van Ruth et al., (2011), using half gram of sample. The FA methyl esters were obtained as described in Guardiola, Codony, Rafecas, and Boatella (1994), and were determined by gas chromatography using a Thermo Focus GC (Thermo Fischer Scientific Inc., Italy), fitted with a flame-ionisation detector and split–splitless injector port, set at 270 °C and 250 °C, respectively. The split ratio was 1:30. Chromatographic separation of FA methyl esters was performed on CP-Select CB for FAME (3-cyanopropyl polysiloxane) capillary column (50 m × 0.25 mm i.d.; film thickness 0.25 μm, Varian, Palo Alto, CA). Helium (18 psi) was used as carrier gas, and the oven was programmed as follows: initial temperature 100 °C, increased at 5 °C/min to 230 °C and held for 9 min. The sample volume injected was 1 μL. Fatty acids were identified by their retention times according to those found in the FAME standard mixture. All DDGS samples were analysed in triplicate and results were expressed as normalised peak areas (%). Data used was the average value of the three replicates for each sample.

Proton transfer reaction mass spectrometry

The volatile fingerprint of DDGS samples were measured in triplicate using a high sensitivity PTR-MS instrument (Ionicon Analytik, Innsbruck, Austria). All measurements were carried out under drift tube conditions of 120–140 Td (Td = Townsend; 1 Td = 10−17 V cm2 mol−1) over a mass range of m/z = 20 to m/z = 160 and a dwell time of 0.2 s mass−1, giving a cycle time of 32 s. Each ground DDGS sample was separately weighed (25 g) into 250 mL glass bottles (Schott Duran bottles, Germany) and allowed to equilibrate at room temperature (∼25 °C) for 30 min. Bottles were connected to the PTR-MS inlet flow that was heated to 60 °C via Teflon (0.25 mm) tubing and headspace air was sampled at a flow rate of 50 ml/min. Masses were analysed in a quadrupole mass spectrometer and detected as ion counts per second (cps) by a secondary electron multiplier (SEM). Mass ion intensities were converted to concentration (ppbv) according to (Lindinger, Hansel, & Jordan, 1998). Sample measurements were performed in 5 cycles resulting in an analysis time of 3.0 min. The mean of cycles 2–4 were represented in further analysis. Background air scans of five cycles were conducted from an empty bottle before each sample measurement and the mean signal was subtracted from the sample spectra (Aprea et al., 2007). Then, the three averaged mass spectrums of the three replicates for each sample were averaged to obtain a mean mass spectrum per sample. In this manner, a dataset containing mean mass spectra per sample analysed could be compiled. M/z 32 (O2+) and m/z 37 (water cluster ion) that are associated with the PTR-MS ion source were removed from the data set. The order of sample and triplicate measurements were randomised to account for possible memory affects.

Statistics

The data matrix consisted of as many rows as samples (n = 82) and as many columns as variables (19 for FA composition matrix and 158 for VOC fingerprint matrix). For multivariate modelling and classification we used Pirouette 4.0 (Infometrix, Seattle, USA). With the 82 samples PCA was performed for both the FA and VOC data, to screen the multivariate data for outliers and to explore the presence of any natural clustering in the data. Several data pre-processing techniques including: none, auto-scaling (scaling to unit variance), mean centring, logarithmic transformation and a combination of these techniques for FA and VOC data sets were applied to evaluate the contribution of all variables (fatty acids and mass ions). The purpose of pre-processing in this case was to allow for all variables to give equal influence to the PCA model, regardless of their original variance. PLS-DA was conducted to develop a classification model to help verify the identity of DDGS samples. Several PLS-DA models were developed according to the feature to be classified: botanical origin (corn vs wheat), geographical origin of corn DDGS samples (USA vs Jilin province in China vs Heilongjiang province in China) and production method for corn DDGS samples (biofuel vs alcoholic beverages). To develop and validate the PLS-DA classification models, each sample set was divided into a training set (consisting of a random selection of 70% of the samples from each category) and a validation set (the remaining 30% of samples from each category). The training sets of samples were used to develop the models, and to internally validate them by leave-10% out cross-validation. Once the models were developed, they were externally validated by predicting the identity of the DDGS samples in the validation set. The performance of the models was assessed by their percentage accuracy (i.e. correctly predicted samples divided by the number of samples in the class) in predicting each class correctly.

Results and discussion

DDGS botanical identity

PCA was conducted as an exploratory technique on the DDGS FA and VOC data (n = 82) separately. Both the FA and VOC PCA scores plots revealed some natural clustering of the DDGS samples which was in agreement with their botanical origin (Fig. 1). No outlier samples were observed for the FA composition or the VOC data. In both approaches, corn DDGS clustered separately from wheat DDGS samples regardless of production method and or geographical origin. Therefore, a classification model was developed using PLS-DA on each data set (FA and VOC) to help verify differences among botanical origin of the grain from which DDGS were produced.

Fig. 1

First two factors of the PCA scores plot based on the fatty acid composition (A) and the volatile profile (B) of corn (◊) and wheat (♦) DDGS samples (auto-scaled data).

DDGS botanical identity by FA profile: classification model

The PLS-DA model on the FA profile (after data auto-scaling) correctly classified (both the internal and the external validation data sets) all wheat and almost all corn samples (Table 1). The unclassified corn samples were not assigned to wheat or corn as the models were not capable of classifying these samples. In this case an unclassified sample could be defined as somewhat different to other samples in terms of the presence or absence of fatty acids in relation to the class group of corn. Subsequently, these samples full outside the limits of the model and are difficult to classify. Despite the limited unassigned corn samples, these results showed that it can be possible to distinguish the botanical identity of the grains used to produce DDGS by the FA content. Comparisons of the PLS-DA scores and loadings plots revealed the FA responsible for differences among the corn and wheat samples (Fig. 2). A high contribution of C18:1 n-9 (oleic), C18:0 (stearic) and C20:0 was observed for the separation of the corn from the wheat samples. Separation of wheat from corn samples was driven by the monounsaturated FA n-6 and n-3 families, such as linoleic (C18:2 n-6) and linolenic acid (C18:3 n-3). This is in agreement with the differences encountered for individual FA between botanical sources of DDGS by univariate analysis (Supplementary Table 2), which is supported by Armanino et al. (2002), Winkler-Moser and Breyer (2011), and Moreau et al. (2011) findings, where FA composition depended on botanical variety of the grains.

Table 1

Identification of DDGS botanical identity and geographical origin: validation of PLS-DA models built on the fatty acid and VOC profiles.

	Botanical origin	Correct classifications (%)		Geographical origin	Correct classifications (%)
		Internal validationa	External validation		Internal validationa	External validation
Fatty acid composition	Corn	97.9%	95.2%	USA	87.5%	100%
	Wheat	100%	100%	Heilongjiang	100%	100%
				Jilin	100%	100%
Volatile organic compound fingerprint	Corn	97.9%	97.9%	USA	100%	100%
	Wheat	97.9%	100%	Heilongjiang	100%	100%
				Jilin	85.7%	100%

Internal validation by leave 10%-out cross-validation.

Fig. 2

Botanical identity of DDGS by fatty acid profile (auto-scaled data): PLS-DA scores plot of corn (◊) and wheat (♦) DDGS (A) and loadings plot of FA (B).

DDGS botanical identity by VOC fingerprinting: classification model

Good classifications were obtained by the PLS-DA model using the VOC data, with almost all samples being correctly classified in the internal and in the external validation sets. Observations along the positive factor of the PLS-DA plots (Fig. 3) indicated that corn were separated from wheat DDGS by mass ions [m/z] 34, 33, 35, 51, and 73, while along the negative axis wheat was separated from corn DDGS by m/z 68, 75, 76, 81, 85, 86, 113, 127, 128. Several VOC detected in DDGS have been associated with the fermentation process, the initiation of Maillard reaction (Seabolt, van Heugten, Kim, Ange-van Heugten, & Roura, 2010), or the ratio of amylopectin to amylose (Belyea et al., 2010) all of which might vary depending on the botanical origin of grains leading to detectable differences in volatile profiles.

Fig. 3

Botanical identity of DDGS by volatile finger-print (auto-scaled data): PLS-DA scores plot (A) of corn (◊), wheat (♦) and loadings plot (B).

DDGS geographical origin

A clear separation of corn samples in two different groups corresponding to DDGS from different geographical origins was observed, especially on the FA data PCA (Fig. 1). To further explore this clustering PCA was conducted on both FA and VOC profiles using only corn samples from known geographical origins (n = 54) (corn samples from unknown geographical origin were not included). For FA data the PCA scores plot showed three different clusters corresponding to USA, and the two Chinese provinces Heilongjiang and Jilin, while observations from the VOC PCA scores plot showed less evidence of cluster by country of origin (data not shown).

DDGS geographical origin by FA profile: classification model

Internal validation of the PLS-DA model based on the FA profile (after auto-scaling) revealed that almost all samples were correctly assigned to their geographical origin (Table 1). In fact, only two USA samples were not assigned to any origin. By external validation, all samples were correctly assigned to their geographical origin, which suggests that FA profiling could be used to help verify the geographical origin of corn DDGS. PLS-DA models discriminating among more than 2 categories, develop one sub-model for each category. In these sub-models, samples belonging to each category are discriminated from all the other samples (i.e. USA vs non-USA; Heilongjiang vs non-Heilongjiang; Jilin vs non-Jilin). For example, Fig. 4 shows that monounsaturated FA of the n-9 series such as C18:1 n-9 and C20:1 n-9 are important for the discrimination of the USA corn from all the other corn DDGS. Alternatively, samples from Jilin are discriminated from samples of Heilongjiang and USA by a high contribution of linoleic acid, while samples from Heilongjiang are discriminated by a high contribution of several n-3 polyunsaturated FA such as linolenic acid, some monounsaturated FA (except oleic acid and C20:1 n-9) and some saturated FA such as stearic palmitic acid (data not shown). These findings are in agreement with differences in the content of the individual FA between the three geographical origins (Supplementary Table 3). To the best of our knowledge, there are no studies in literature addressing the differences in corn DDGS FA composition between these geographical origins, making it difficult to explain the factors behind these results. However, differences in the FA composition according to the geographical location in other commodities such as palm or olive oil (Tres et al., 2013; Garcia-Gonzalez, Luna, Morales, & Aparicio, 2009) have been related to differences in the botanical identity grown in each origin, to pedoclimatic conditions as well as to variations in the production process. Therefore, it is likely that these factors are also responsible for the differences observed in corn DDGS. In fact, it has been reported that besides the botanical source, the production method (i.e. type of process, drying step, amount of solubles added back to distillers wet grains) influence the final nutrient composition of DDGS (Belyea et al., 2010; Stein, Connot, & Pedersen, 2009).Fig. 5

Fig. 4

Verification of the geographical origin of corn DDGS: PLS-DA scores plot (A) of USA (◊), Jilin (▵), Heilongjiang (□), and loadings plots (B) of the fatty acid based models for USA compared to Jilin and Heilongjiang provinces of China.

Fig. 5

Verification of the geographical origin of corn DDGS: PLS-DA scores plot of USA (◊), Jilin (▵), Heilongjiang (□) (A) and factor 1 loadings plots (B) of the volatile profile based model for USA compared to provinces Jilin and Heilongjiang from China.

Identification of DDGS production process

DDGS are a by-product of ethanol production through fermentation of grains to produce fuel or alcoholic beverages. As reported in literature, differences in production processes might lead to differences in DDGS composition. Therefore, we investigated whether FA profile or VOC fingerprinting could be a useful strategy to help verify the production process by which DDGS had been obtained (fuel vs beverage). However, our sample set only contained a few samples for which the full information on the production process was not available. Since we have observed that the botanical source of grains has a remarkable influence on DDGS composition, wheat DDGS were not included in the DDGS production model. Therefore, 31 corn samples were considered, of which 19 were from ethanol beverage production and 12 were from biofuel ethanol based production. Results from PCA on the VOC fingerprint showed some natural grouping in agreement with the DDGS production method (fuel vs beverage). However, there was some overlap between both clusters in the PCA on the FA composition data (data not shown). PLS-DA was attempted with all 31 samples. Due to the low number of samples that could be included in the production method, all samples were used as training set and the model was only internally validated by leave-10% out cross-validation.

Identification of DDGS production process by FA profile: classification model

FA fingerprinting did not seem a promising technique to discriminate DDGS production method because of several (35%) misclassifications occurring during model internal validation. However, considering the high influence of the geographical origin on DDGS FA composition in this study, it is likely that the geographical identity of DDGS masked any possible differences between samples derived from fuel or beverage industries. Unfortunately, there were not enough DDGS samples produced as co-products of the fuel and the beverage industries from the same country to perform a classification model (fuel vs beverage) for each country. It is therefore advisable to increase the DDGS sample set before discarding FA fingerprinting for this verification purpose.

Identification of DDGS production process by VOC fingerprint: classification model

VOC PLS-DA classification demonstrated that 100% of DDGS from beverage production and 91.6% of DDGS from biofuel production were correctly classified, leaving only one sample unassigned. Thus volatile profiling showed improved results when compared to FA profiling for distinguishing method of production. Beverage DDGS were separated from biofuel produced DDGS along the negative axis of factor 1 as a result of a high contribution of m/z 43, 44, 61, 62, 63, 64 and 110 (Fig. 6). In contrast a high contribution of m/z 42, 49, 87, 88, 89, 101, 102, 133 and 134 were responsible for separating biofuel from beverage produced samples along the positive axis of factor 1. In correspondence with literature and where compounds are present in DDGS, masses could be tentatively assigned to propylene glycol and/or 3-Methyl-1-butanol (m/z 43), acetic acid and/or a fragment of ethyl acetate (m/z 61), dimethyl sulphide (m/z 63), diacetyl (m/z 87), and ethyl acetate (m/z 88). In particular acetic acid and diacetyl are associated with by-products from fermentation and therefore dependent on the process (Seabolt et al., 2010). These findings are in line with previous reports that suggest that processing technique including fermentation and drying parameters yield high variability between DDGS composition (Belyea et al., 2010; Liu, 2011; Nuez-Ortin & Yu, 2009). For example Nuez-Ortin and Yu (2009) suggested that during fermentation some biofuel plants add significant amounts of sulphuric acid to adjust the pH, resulting in sulphur rich DDGS products. Despite the tentative assignment of some masses due to production process and as consequence further studies are required to help verify the identity of compounds derived from DDGS samples.

Fig. 6

Verification of the production method of corn DDGS: PLS-DA scores plot of beverage (◊), biofuel (□) (A) and factor 1 loadings plots (B) of the volatile profile based models for production method.

Conclusions

In summary, the classification of the botanical and geographical origin and production process of DDGS using both FA and VOC profiles showed promising results. While volatile profiling using PTR-MS offered comparable results to FA fingerprinting; the latter is an analytical determination quite common in most food analytical laboratories. Regarding the identification of the DDGS production process, VOC fingerprinting offered a more successful classification approach than FA profiles, which is likely to be related to specific volatiles generated during the different fermentation, drying and manufacturing processes occurring in beverage or biofuel production. The classification models developed in this study represent a promising proof of concept study that could be expanded in the future to other DDGS botanical sources and geographical origins (i.e. for wheat DDGS), provided that new authentic samples are included in the sample sets. Apart from authentication purposes, DDGS compositional data and classification of its botanical, geographical origin and production process might be important information for feed producers, because it is well-known that feed composition might determine feed digestibility, animal growth, health and composition of animal products such as meat or milk.