Research Article

Differential Detection of Potentially Hazardous Fusarium Species in Wheat Grains by an Electronic Nose

  • Jakob Eifler,

    Affiliations: Department of Crop Sciences, Georg-August-Universität Göttingen, Göttingen, Germany, Department of Neurophysiology and Cellular Biophysics, Georg-August-Universität Göttingen, Göttingen, Germany

  • Eugenio Martinelli,

    Affiliation: Department of Electronic Engineering, University of Rome, Rome, Italy

  • Marco Santonico,

    Affiliation: Department of Electronic Engineering, University of Rome, Rome, Italy

  • Rosamaria Capuano,

    Affiliation: Department of Electronic Engineering, University of Rome, Rome, Italy

  • Detlev Schild mail,

    Affiliations: Department of Neurophysiology and Cellular Biophysics, Georg-August-Universität Göttingen, Göttingen, Germany, Bernstein Focus of Neurotechnology, University of Göttingen, Göttingen, Germany

  • Corrado Di Natale

    Affiliation: Department of Electronic Engineering, University of Rome, Rome, Italy

  • Published: June 09, 2011
  • DOI: 10.1371/journal.pone.0021026


Fungal infestation on wheat is an increasingly grave nutritional problem in many countries worldwide. Fusarium species are especially harmful pathogens due to their toxic metabolites. In this work we studied volatile compounds released by F. cerealis, F. graminearum, F. culmorum and F. redolens using SPME-GC/MS. By using an electronic nose we were able to differentiate between infected and non-infected wheat grains in the post-harvest chain. Our electronic nose was capable of distinguishing between four wheat Fusaria species with an accuracy higher than 80%.


Fusarium species, F. graminearum and F. culmorum in particluar, are widespread pathogens in all cereal growing areas worldwide and one of the most important genera of plant pathogenic fungi on earth [1], [2]. Toxic metabolites, especially the mycotoxin deoxynivalenol (DON, Vomitoxin) and zearalenone (ZEA) are produced from these genera [3].

In 2001 the WHO/FAO carried out a world-wide study comprising Argentina, Brazil, Canada, China, Finland, Germany, Italy, the Netherlands, Norway, Sweden, the United Kingdom, Uruguay and the USA. It was estimated therein that, on the average, 57% of the wheat samples analyzed (11.444) contained deoxynivalenol (DON) [4]. Results from random analysis in Germany indicate that only 29% of all cereal-based products are devoid of Fusarium-toxins [5]. The mycotoxin thus appears to spread to food products becoming a potential and presumably serious health risk to humans [6]. Thus, the FAO has issued guidelines and regulatory limits for Fusarium mycotoxins [7].

As a consequence, there is an absolute need for early and readily applicable methods to detect Fusarium-infected grain and to distinguish between relevant and harmless species. So far Fusarium metabolites have been detected and analyzed using GC-MS [8], LC-UV or LC-MS [9], TLC [10], fluorescence immunoassays [11], NIRS [12], HPLC-MS [13], ELISA [14], whereas whole fungi were detected using ELISA or PCR techniques [15][17]. While some of these approaches are able to detect specific species, they commonly lack quantitative analysis [18]. Others do not allow differentiating between different Fusarium species (ELISA) [17]. Quantitative PCR turned out to be the most precise method, albeit expensive and time-consuming [17]. However, all of the methods used so far are laboratory–based, and none of them allows the on-line detection and quantification, possibly in the field.

Fusaria have been shown to emanate a number of volatile compounds, specifically carbonyls, hydrocarbons, ketones, terpenes and complex mixtures of alcohols [19][24]. While these studies suggested a number of relevant compounds, none of them turned out to be a specific marker for any of the Fusaria. It thus seems to be the pattern of chemicals that is characteristic for any of the species. In the laboratory, such patterns would commonly be analyzed using standard analytical equipment, e.g., gas chromatography–mass spectrometry (GC–MS) [25]. A similar objective can be obtained by using an electronic nose [26], i.e., an array of solid-state sensors that are non-selectively sensitive to the relevant chemicals and the responses of which reflect the chemical information contained in the sample. Note that this detection scheme is in many respects similar to natural olfaction where hundreds of different receptors allow to distinguish among tens of thousands of different odors [27]. Electronic noses have been applied in different fields providing useful identification and classification of samples [28]. The identification of contaminated grains was attempted by several groups. These studies were based on various sensor technologies such as chemosensitive field effect transistors [29], conducting polymers [30], metal oxide semiconductors [31] and quartz microbalances [32].

To differentiate on-line between whole, dry wheat grains that were differentially contaminated by four Fusarium species, we here used an electronic nose based on an array of metalloporphyrin-coated quartz microbalances. The discrimination properties of this instrument were demonstrated in several applications to study food processes [33] and lung cancer diagnosis from breath analysis and medical diagnosis [34].

Materials and Methods


Grains from soft wheat (cv Isengrain, harvest season 2009, Germany) were used. The seeds were water-saturated for 24 hours to ensure rehydration and then autoclaved twice for 15 min at 121°C. For each sample 100 g sterilized kernels were inoculated with ten 0.5×0.5 cm2 slices of fungal mycelia derived from cultures grown on potato dextrose agar (PDA). Incubation was carried out for 5, 10 and 15 days at a relative humidity of 70% and at 27°C. Infected samples were dried to 13% moisture content and stored at 4°C to block further fungal growing. We used the fungus species Fusarium graminearum, Fusarium culmorum, Fusarium cerealis and Fusarium redolens. Sterilized, non-infected kernels incubated over 0, 5, 10 or 15 days as well as an untreated probe served as controls. The number of samples was chosen in respect of the sample variability artificially induced by inoculating microorganisms in homogeneous grain samples. Thus, the influence of the natural variability among grains, e.g., due to species variability and crop production is not taken into account here.

Headspace generation

For both GC/MS and Enose measurements, grain samples (3 g) were enclosed in Teflon-sealed vials. Prior to the measurements, the samples were kept in a thermal bath for 30 min at constant temperature. The Enose experiments were carried out at a sample temperature of 30°C. GC/MS analyses were done at 30°C and 70°C. In addition, an empty vial was added as a reference air source.

Electronic Nose analysis

The core of the electronic nose used in this paper consisted of an array of eight quartz microbalances (QMB), each being a quartz crystal resonator with mass-dependent eigenfrequency f. Slight mass changes (Δm) of the quartz surface result in frequency changes (Δf) of the electrical output signal of the oscillator circuit. The quantities Δm and Δf are linearly related to each other in the low-perturbation regime [35]. Our Enose consisted of QMB with a fundamental frequency of 20 MHz and a mass sensitivity in the order of a few nanograms. QMBs can be turned into chemical sensors by coating their surfaces with layers of chemically sensitive materials. In this work we coated the QMBs with layers of metalloporphyrins. Regarding their sensing properties metalloporphyrins host several interaction mechanisms from weak and non-selective Van der Waals forces to the more energetic and specific coordination of the central metal atom. The balance between these forces can be controlled by the nature of the porphyrins' peripheral group and the metal atom, so that metalloporphyrins with different sensitivities for volatile compounds can be obtained [36] and assembled to sensor arrays for electronic noses [37].

For the experiments of this paper, the grain samples were closed in a sealed vial with an inlet and an outlet. Vials were kept at a constant temperature in order to allow for a stable headspace composition. The headspace was extracted by a flow of ambient air, filtered through a CaCO3 bed. The flow was maintained constant at 7.5 ml/min by a peristaltic pump of the electronic nose. The filtered ambient air was also used to clean the sensors and to establish the reference signal. Sensor signals were calculated as the signal frequency shift, Δf = fs−fa, with fs and fa being obtained from the sample headspace and filtered ambient air. Sensors were exposed to a sample for 60 s, followed by a 5 min cleaning and regeneration phase with reference air. All measurements were repeated three times.

GC/MS analysis

GC/MS analysis were performed using a gas chromatograph (QP2010, Shimadzu, Japan) connected to a mass spectrometer. An EQUITY-5 capillary column (l, 30 m; i.d., 0.25 mm; 0.25 µm) was used in splitless mode with a programmed temperature time course (starting at 40°C and increasing, up to 250°C, at a rate of 10°C /min and beyond at a rate of 20°C /min, up to 300°C, followed by a hold time of 2 min). Helium (p = 14.5 kPa) was used as carrier gas with a total flow of 5.4 ml/min and a column flow of 0.59 ml/min. The mass spectrometer worked at an ionization energy of 70 eV and a mass range from 40 to 300 m/z.

Headspace collection was performed by solid phase microextraction (SPME) with the carboxen/polydimethylsiloxane (CAR-PDMS, 75 µm) fiber (Supelco). After exposure of the fiber to the pre-heated headspace of the grains for 30 min, its contents were injected into the GC for 1 min at an injection temperature of 250°C.

Compound identification was done using the NIST library.

Data analysis

The QMB frequency differences, Δf, between the steady-state reference (air) and the recording phase was used as feature vector. The classification of the different samples was based on a discriminant analysis. We used the partial least squares discriminant analysis (PLS-DA), which is an algorithm originally developed for quantitative regression [38]. Before the application of PLS-DA, Enose data were properly auto-scaled (zero mean, unitary variance). Discrimination models have been cross-validated by the leave-one-out method in order to estimate the classification performance. Data evaluation was done using Matlab (Mathworks).


Species identification

As a first step we investigated the capability of the Enose to differentiate between Fusarium strains. The PLS-DA model used for this discrimination in the infected samples covered five latent variables, the first two of which are shown in Figure 1A. The smallest variance within a group occured in the samples of F. cerealis. Samples of this group were overlapping with the highly contaminated samples of F. redolens and F. culmorum. The widest distribution was found in the samples of F. graminearum showing a partial overlap with F. culmorum and F. cerealis. It can be noticed that the gas chromatographic profiles from F. redolens and F. culmorum are similar. However, there was a large interclass variance detectable. Thus, the model has a satisfactory capability of differentiating different classes, as confirmed by the confusion matrix (Table 1). In the cross-validation F. culmorum and F. cerealis were perfectly recognized (100% correct). The classification rates of F. graminearum and F. redolens were 83% and 89%, respectively. The correct classification rate across all fungi was 94%.


Figure 1. Enose results indicating different Fusarium species and infection levels.

A: Scores plot of the first two latent variables, showing groups of F. culmorum, F. graminearum, F. redolens and F. cerealis. B: Discrimination between high- (15 days incubation) and low -infected (5 to 10 days incubation) samples of Fusarium. The diagram shows the scores of the first two out of 3 latent variables.


Table 1. Confusion matrix of true vs. estimated values of species classification.


Classification of infection level

Correctly classifying the level of infection turned out to be slightly more difficult due to a high interspecies variability and the need for many data samples at the various concentrations. Thus, a binary classification with three latent variables in total was done (Figure 1B). For that purpose samples with an incubation time of 15 days were classified as highly infected, while samples having been incubated over 5 or 10 days were merged and classified as low-infected. This led to a higher variability in the low level probes. Nevertheless, there was a clear difference detected. Classification of the infection level (Table 2) showed that 91% of the samples were correctly classified. However, 18% of highly infected samples were classified as low-infected, all of them being F. graminearum samples.


Table 2. Confusion matrix of true vs. estimated values of binary classification of infection levels.


Fungi vs. Control

For a qualitative analysis all measurements were split into two groups (infected vs. control). Figure 2A shows average sensor responses of infected vs. control samples or ambient air. For the infected samples, the responses of all sensors were on the average 16 Hz higher than for the controls. Ambient air gave very low responses. The signals were lower than 50% as compared to grain measurements. Furthermore, a PLS-DA model was performed on these data with a total of four latent variables. The scores plot of the first two variables (Figure 2B) showed a segregation of infected from control samples, although with a slight overlap. Accordingly, the major portion (91%) of infected probes are recognized correctly by cross-validation (confusion matrix, Table 3). Eight out of 47 measurements led however to a misclassification due to large variances within the groups. All of the false negatives originated from F. redolens at a high infection level. The variance of the F. redolens group might thus be explained by the fact that the high concentrations of those samples were associated with changes in the relative components of the patterns. These probes also led to false classifications during species identification (Table 1). Further, the data of the infected group showed also a rather large variance, most presumably due to the heterogeneity of the dataset consisting of probes from different fungi and at different levels of infection. In fact, using GC/MS yielded a complex composition of the headspace (Table 4). While some substances such as ethanol or hexadecane appeared in all samples, several other compounds reported to be infection-specific (2-methyl-1-propanol, 3-methyl-butanol, 1-octen-3-ol and 3-octanone [8], [23], [30], [39]) were differentially detected in some but never in all samples.


Figure 2. Enose results indicating Fusarium infection.

A: Frequency shift of the quartz microbalance sensors during measurements of infected (n = 32) and control (n = 15) samples, and of ambient air (n = 5). Error bars indicate standard deviations. B: Scores plot of the first two latent variables of the PLS-DA model showing groups of infected and non-infected samples. The Model covers four latent variables in total.


Table 3. Confusion matrix of true vs. estimated values, summarizing the classification results of infection.


Table 4. Volatile compounds released by the analyzed fungi and an untreated control (sampling temperature, 70°C).


The total amount of volatiles found in our GC/MS analysis clearly increased with increasing sampling temperature. Samples infected at 70°C emitted approximately 5-fold more volatiles than at 30°C, which expectedly led to a larger variance. In control samples the corresponding increase was just 60% (Figure 3).


Figure 3. Temperature-dependency of abundancies.

Comparison of total volatile abundances of four Fusaria species and controls as obtained from the GC/MS measurements taken at two different sampling temperatures.



F. graminearum and F. culmorum are known mycotoxin producers and thus potentially hazardous. It is therefore important to differentiate between Fusarium species. They are the most pathogenic and most frequently occurring Fusaria species [1]. In wheat and barley the pathogens lead to a destructive disease called Fusarium head blight (FHB). The fungi affect mainly living plants, especially during warm and wet weather conditions in the time of flowering [40]. The occurrence of FHB leads to a reduction in grain size, kernel weight, germination rate as well as a depression in quality parameters [1], [41]. Thus, high economic losses are going along with FHB infection. Furthermore there are hazardous effects of the toxic metabolites produced by Fusarium. Contaminated cereals, if used in livestock feed, cause a reduction in food uptake and eventually economic losses [42]. The relevance for humans should not be underestimated since mycotoxins, when entering food products, represent a potential toxicity [6].

We used a linear classifier for our data analysis. A more complex classifier such as a neural network would presumably improve the classification rate. However, there is a balance between the classifier performance and the generalisability of the results. If the complexity of the classifier is kept low, the results are usually more general, whereas results obtained with more complex classifiers tend to be less prone to generalization. For this study we therefore preferred a simple and robust classifier. The result that 18% of highly infected samples were classified as low infected ones appears to be the price for the simple classifier. As a consequence, the risk of samples classified as low infected must not be underestimated. A more complex analysis, the results of which can at the same time be generalised, is of course desirable and would certainly be preferred in many cases. We are currently developing a bionic algorithm of this kind hopefully achieving both requirements.

Our Enose analyses performed at 30°C turned out to be more reproducible than those performed at 70°C. This somewhat unexpected behavior may have the following explanations. First, while transferring the gas from the source to the sensor array, the temperature was not kept constant, so that a temperature drop might have induced condensation phenomena of some volatile compounds resulting in a change of the composition of the sample prior to entering the sensor cell. Second, the adsorption of volatiles onto a porphyrin layer is a temperature-dependent process, the efficiency of which decreases with increasing temperatures.

The false positive classifications observed are brought about by a high variance in our control groups. There are several possible reasons for this. As our model performed a binary classification, it was unable to differentiate between species or levels. Second, the control groups contained samples exposed to different incubation times. Third, incubation of non-inoculated grains may have changed the composition of odor compounds. In practical terms, false negative classifications would seriously compromise the Enose approach, while false positive classifications were considered acceptable, though unsatisfactory.

As we aimed at a qualitative recognition of infected samples, we took the classification rates as efficiency measure. The fact that all classification rates were higher than 83%, most of them being much higher, clearly indicated that chemometrical fingerprints allow the detection of fungal infection as well as the discrimination between specific Fusaria species.

In this paper we tried to bridge the obvious gap between the relevance of Fusarium infestation and its simple, fast, on-line, portable and quantitative measurement. We show that the metalloporphyrin-based Enose can be used to qualitatively detect and correctly classify dry, whole, Fusarium-infected wheat grains. Even low-contaminated grains were accurately detected, allowing them to be excluded from the food or feed chain. Second, in some practical cases a Fusarium infestation–specific signal must be recognised in the presence of a background of interfering molecular species. This would commonly be achieved–if necessary–by increasing the number of sensors. Third, the Enose meets most analytical requirements needed. It is a mobile, inexpensive and relatively fast electronic device, capable of differentiating hazardous grain from innocuous grain and thereby guaranteeing the compliance with existing health standards.

Author Contributions

Conceived and designed the experiments: JE EM MS RC DS CDN. Performed the experiments: JE EM MS RC. Analyzed the data: JE EM MS RC DS CDN. Contributed reagents/materials/analysis tools: JE EM DS CDN. Wrote the paper: JE EM DS CDN.


  1. 1. Parry DW, Jenkinson P, McLeod L (1995) Fusarium ear blight (scab) in small grain cereals; a review. Plant Pathol 44: 207–238.
  2. 2. Yazar S, Omurtag GZ (2008) Fumonisins, Trichothecenes and Zearalenone in Cereals. Int J Mol Sci 9: 2062–2090.
  3. 3. Bottalico A, Perrone G (2002) Toxigenic Fusarium species and Mycotoxins Associated with Head Blight in Small-Grain Cereals in Europe. Eur J Plant Pathol 108: 611–624.
  4. 4. (2001) Fifty-sixth report of the Joint FAO/WHO Expert Committee on Food Additives 906: World Health Organ Tech Rep Ser.
  5. 5. Schollenberger M, Suchy S, Jara HT, Drochner W, Müller H-M (1999) A survey of Fusarium toxins in cereal-based foods marketed in an area of southwest Germany. Mycopathologia 147: 49–57.
  6. 6. Peraica M, Radic B, Lucic A, Pavlovic M (1999) Toxic effects of mycotoxins in humans. Bull World Health Organ 77: 754–766.
  7. 7. Van Egmond HP, Jonker MA (2004) Worldwide Regulations for Mycotoxins in Food and Feed in 2003. FAO Food Nutr Pap 81:
  8. 8. Olsson J, Börjesson T, Lundstedt T, Schnürer J (2002) Detection and quantification of ochratoxin A and deoxynivalenol in barley grains by GC-MS and electronic nose. Int J Food Microbiol 72: 203–214.
  9. 9. Razzazi-Fazeli E, Böhm J, Luf W (1999) Determination of nivalenol and deoxynivalenol in wheat using liquid chromatography–mass spectrometry with negative ion atmospheric pressure chemical ionization. J Chromatogr A 854: 45–55.
  10. 10. Schaafsma AW, Nicol RW, Savard ME, Sinha RC, Reid LM, et al. (1998) Analysis of Fusarium toxins in maize and wheat using thin layer chromatography. Mycopathologia 142: 107–113.
  11. 11. Maragos CM, Plattner RD (2002) Rapid fluorescence polarization immunoassay for the mycotoxin deoxynivalenol in wheat. J Agric Food Chem 50: 1827–1832.
  12. 12. Dowell FE, Ram MS, Seitz LM (1999) Predicting Scab, Vomitoxin, and Ergosterol in Single Wheat Kernels Using Near-Infrared Spectroscopy. Cereal Chem 76: 573–576.
  13. 13. Plattner RD (1999) HPLC/MS analysis of Fusarium mycotoxins, fumonisins and deoxynivalenol. Nat Toxins 7: 365–370.
  14. 14. Abouzied MM, Azcona JI, Braselton WE, Pestka JJ (1991) Immunochemical assessment of mycotoxins in 1989 grain foods: evidence for deoxynivalenol (vomitoxin) contamination. Appl Environ Microbiol 57: 672–677.
  15. 15. Iyer MS, Cousin MA (2003) Immunological detection of Fusarium species in cornmeal. J Food Prot 66: 451–456.
  16. 16. Martin KJ, Rygiewicz PT (2005) Fungal-specific PCR primers developed for analysis of the ITS region of environmental DNA extracts. BMC Microbiol 5: 28–39.
  17. 17. Brandfass C (2006) Establishment and application of real-time PCR-based methods to study the epidemiology of Fusarium Head Blight. Dissertation. Universität Göttingen.
  18. 18. Atkins SD, Clark IM (2004) Fungal molecular diagnostics: a mini review. J Appl Genet 45: 3–15.
  19. 19. Pasanen AL, Lappalainen S, Pasanen P (1996) Volatile organic metabolites associated with some toxic fungi and their mycotoxins. Analyst 121: 1949–1953.
  20. 20. Schnürer J, Olsson J, Börjesson T (1999) Fungal Volatiles as Indicators of Food and Feeds Spoilage. Fungal Genet Biol 27: 209–217.
  21. 21. Korpi A, Järnberg J, Pasanen A-L (2006) The Nordic Expert Group for Criteria Documentation of Health Risks from Chemicals: 138. Microbial volatile organic compounds (MVOCs). Arb Hälsa 13:
  22. 22. Börjesson T, Stöllman U, Adamek P, Kaspersson A (1989) Analysis of volatile compounds for detection of molds in stored cereals. Cereal Chem 66: 300–304.
  23. 23. Jelen HH, Latus-Zitkiewicz D, Wsowicz E, Kamiski E (1997) Trichodiene as a volatile marker for trichothecenes biosynthesis. J Microbiol Methods 31: 45–49.
  24. 24. Keshri G, Magan N (2000) Detection and differentiation between mycotoxigenic and non-mycotoxigenic strains of two Fusarium spp. using volatile production profiles and hydrolytic enzymes. J Appl Microbiol 89: 1365–2672.
  25. 25. Perkowski J, Buśko M, Chmielewski J, Góral T, Tyrakowska B (2008) Content of trichodiene and analysis of fungal volatiles (electronic nose) in wheat and triticale grain naturally infected and inoculated with Fusarium culmorum. Int J Food Microbiol 126: 127–134.
  26. 26. Nagle T, Schiffmann S, Gardner J, Pearce T, editors. (2003) Handbook of machine olfaction. Weinheim, Germany: J. Wiley–VCH.
  27. 27. Korsching S (2002) Olfactory maps and odor images. Curr Opin Neurobiol 12: 387–392.
  28. 28. Rök F, Barsan N, Weimar U (2008) Electronic nose: current status and future trends. Chem Rev 108: 705–713.
  29. 29. Jonsson A, Winquist F, Schnürer J, Sundgren H, Lundström I (1997) Electronic nose for microbial quality classification of grains. Int J Food Microbiol 35: 187–193.
  30. 30. Magan N, Evans P (2000) Volatiles as an indicator of fungal activity and differentiation between species, and the potential use of electronic nose technology for early detection of grain spoilage. J Stored Prod Res 36: 319–340.
  31. 31. Falasconi M, Gobbi E, Pardo M, Della Torre M, Bresciani A, et al. (2005) Detection of toxigenic strains of Fusarium Verticilloides in corn by electronic olfactory system. Sens Actuators B 108: 250–257.
  32. 32. Paolesse R, Alimelli A, Martinelli E, Di Natale C, D'Amico A, et al. (2006) Detection of fungal contamination of cereal grain samples by an electronic nose. Sens Actuators B 119: 425–430.
  33. 33. Santonico M, Bellincontro A, De Santis D, Di Natale C, Mencarelli F (2010) Electronic nose to study postharvest dehydration of wine grapes. Food Chem 121: 789–796.
  34. 34. D'Amico A, Pennazza G, Santonico M, Martinelli E, Roscioni C, et al. (2010) An investigation on electronic nose diagnosis of lung cancer. Lung Cancer 68: 170–176.
  35. 35. Ballantine DS, White RM, Martin SJ, Ricco AJ, Zellers ET, et al. (1997) Acoustic wave sensors: theory, design and physico-chemical applications. Academic Press.
  36. 36. Paolesse R, Mandoj F, Marini A, Di Natale C (2004) Porphyrin based chemical sensors. In: Nalwa HS, editor. Encyclopedia of Nanoscience and Nanotechnology, Vol. 9. Valencia CA: American Science Publishers, 21:
  37. 37. Di Natale C, Paolesse R, D'Amico A (2007) Metalloporphyrins based artificial olfactory receptors. Sens Actuators B 121: 238–246.
  38. 38. Wold S, Sjöström M, Eriksson L (2001) PLS-regression: a basic tool of chemometrics. Chemom Intell Lab Syst 58: 109–130.
  39. 39. Miyazawa M, Kimura M, Yabe Y, Tsukamoto D, Sakamoto M, et al. (2008) Use of Solid Phase Microextraction (SPME) for Profiling the Volatile Metabolites Produced by Glomerella cingulata. J Oleo Sci 57: 585–590.
  40. 40. Racca P, Jörg E, Weinert J, Von Tiedemann A (2007) Die DON-Werte voraussagen, DLG Publikationen 4: 60–61.
  41. 41. Schaafsma AW, Tamburic-Ilinic L, Miller JD, Hooker DC (2001) Agronomic considerations for reducing deoxynivalenol in wheat grain. Can J Plant Pathol 23: 279–285.
  42. 42. Rotter BA, Prelusky DB, Pestka JJ (1996) Invited Review: Toxicology of Deoxynivalenol (vomitoxin). J Toxicol Environ Health 48: 1–34.