Characterization of food spoilage fungi by FTIR spectroscopy



Volha Shapaval, Nofima AS, Centre for Biospectroscopy and Data modeling, Osloveien 1, N-1430 Ås, Norway. E-mail:



The objective of the study was to evaluate a high-throughput liquid microcultivation protocol and FTIR spectroscopy for the differentiation of food spoilage filamentous fungi.

Methods and Results

For this study, fifty-nine food-related fungal strains were analysed. The cultivation of fungi was performed in liquid medium in the Bioscreen C microtitre plate system with a throughput of 200 samples per cultivation run. Mycelium was prepared for FTIR analysis by a simple procedure, including a washing and a homogenization step. Hierarchical cluster analysis was used to study affinity among the different species. Based on the hierarchical cluster analysis, a classification and validation scheme was developed by artificial neural network analysis. The classification network was tested by an independent test set. The results show that 93·9 and 94·0% of the spectra were correctly identified at the species and genus level, respectively.


The use of high-throughput liquid microcultivation protocol combined with FTIR spectroscopy and artificial neural network analysis allows differentiation of food spoilage fungi on the phylum, genus and species level.

Significance and Impact of the Study

The high-throughput liquid microcultivation protocol combined with FTIR spectroscopy can be used for the detection, classification and even identification of food-related filamentous fungi. Advantages of the method are high-throughput characteristics, high sensitivity, low costs and relatively short time of analysis.


Moulds (filamentous fungi) – a group of fungi that grow in the form of multicellular filaments called hyphae – may be a problem in food industry. Several fungal species are air-borne or present in the environment or in raw materials. However, there are only a limited number of fungal species that are responsible for the mould contamination of a certain food product (Filtenborg et al. 1996). The associated mycobiota of a food product is normally constituted by less than ten species. Hence, the handling of undesirable moulds in the food industry should be focused on the associated mycobiota of the relevant product. However, differentiation and identification of moulds in the food industry are difficult (Pitt and Hocking 2009).

State-of-the-art identification schemes in mycology are based on phenotypic and genotypic methods. Phenotypic methods for fungal identification generally comprise plating on agar media followed by macroscopic and microscopic morphological examinations. In addition, a restricted assortment of physiological and biochemical reaction patterns are used as phenotypic methods. The disadvantages of microscopic examination are that it is time-consuming, it needs highly skilled personnel and that discrimination beyond the genus level is difficult. Moreover, fungal taxonomy is complicated due to the existence of anamorphs (asexual states) and teleomorphs (sexual states) leading to different genus and species names for the same fungus. Commercial identification systems, like the Biolog Filamentous Fungi system (Jordan et al. 2010), are mainly designed for clinical mould isolates and do not cover food spoilage moulds.

Genotypic methods, such as amplified fragment length polymorphism (AFLP) and polymerase chain reaction (PCR) of amplified DNA, using conserved sequences of mitochondrial and ribosomal genes and internal transcribed spacer regions (ITS), are used in the fungal taxonomy. While DNA-based methods are very reproducible and fast in the identification of different species and strains of bacteria, they still have some limitations for the identification of filamentous fungi. For example, < 50% of the Zygomycota isolates that were used in the study by Hall L. et al. could be correctly identified by use of the large-subunit ribosomal DNA sequencing kit (Hall et al. 2004). In general, genetic rapid techniques require highly skilled personnel and have high costs of consumables (Jordan et al. 2010).

In recent years, Fourier transform infrared spectroscopy (FTIR) has attracted attention as a method for characterization and identification of filamentous fungi. FTIR spectroscopy has been successfully applied for the differentiation of Aspergillus and Penicillium on genus, species and strain levels (Fischer et al. 2006), for the differentiation of wood-degrading basidiomycetes (Naumann 2009), various Fusarium species and phytopathogens (Nie et al. 2007; Linker and Tsror (Lahkim) 2008), for intraspecies characterization of airborne filamentous fungi (Santos et al. 2010) and for the identification of dermatophytes (Bastert et al. 1999). A few studies report results on the identification of spoilage fungi by FTIR spectroscopy in agricultural commodities and food, for example, the detection and differentiation of Fusarium oxysporum and Rhizopus stolonifer in tomatoes (Hahn 2002) and the detection of toxigenic fungi in corn by transient infrared spectroscopy (Gordon et al. 1999). These studies have shown that there is a clear potential for FTIR spectroscopy to be used as a routine method with high specificity for the identification of fungi. However, the protocols presented in the above-mentioned studies are not yet optimized for the use in an industrial setting: (i) all these protocols need at least 14 days for cultivation of fungi on agar media, which is time-consuming. In addition, the cultivation on agar media has limited throughput and is difficult to automate, (ii) all these protocols require the separation of the mycelium from spores by filtrating or other techniques because spores and mycelium should be analysed separately to obtain the highest possible discrimination, and (iii) they apply manual re-suspension of the mycelium and/or washing of the mycelium and/or spores before FTIR analysis.

We have recently established a high-throughput protocol for fungal characterization by FTIR spectroscopy (Shapaval et al. 2010), which is tailored for industrial use with the following characteristics: (i) It is high-throughput, as 200 samples can be measured simultaneously, (ii) the time for cultivation and analysis is only 5 days, (iii) handling is simple and can be automated, and (iv) it has high specificity. The protocol combines high-throughput FTIR spectroscopy and high-throughput liquid microcultivation in a microtiter plate system (Bioscreen C system; Oy Growth Curves AB, Helsinki, Finland), allowing cultivation in liquid media of several hundred fungi simultaneously. The mycelium obtained after cultivation in liquid media for five days has no spores or very little quantities of them. The protocol involves a simple sample preparation step, including washing of fungal mycelium with distilled water and short-time ultrasound sonication. The sample of the mycelium prepared for FTIR analysis does not contain any kind of externally added chemicals, which may affect FTIR spectra. Our previous study was focused only on protocol development and optimization and showed that 11 strains could be clustered according to species and genus level. In the present study, our objective was to use this protocol to develop an identification scheme for a broad range of fungal species associated with food. In particular, we wanted to evaluate the identification power of the developed protocol for the characterization of a large set of food spoilage fungal strains: fifty-nine strains belonging to nineteen species and ten genera were included in the study.

Materials and methods

Mould strains

Fifty-nine well-characterized fungal strains representing ten different fungal genera (Alternaria, Aspergillus, Eurotium, Geotrichum, Fusarium, Mucor, Paecilomyces, Phoma, Penicillium and Rhizopus), obtained from the mycological strain collection of the Norwegian Veterinary Institute (Oslo, Norway), were used in the study. All fungal strains have food relevance. Detailed information about the fungal strains is shown in Table 1.

Table 1. Fungal strains included in the study
  1. a

    All numbers refers to the VI strain number in the mycological strain collection of the Norwegian Veterinary Institute

Alternaria A. alternata 02814a02815 03120 04067  
Aspergillus A. fumigatus 02980 03680 03042   
A. niger 01880 04499 01350   
A. flavus 01656 01652 03121 04759  
Eurotium E. amstelodami 02040 02088 0216603344 04566
E. herbariorum 02039     
E. chevalieri 02038 02041   
Fusarium F. graminearum 01022 01028 01189 02875 02902
F. langsethiae 03266 01269 03401 03405 03276
Geotrichum G. candidum 03033 03865    
Mucor M. plumbeus 02019 02022 03754   
M. circinelloides 01914 04473    
M. hiemalis 01993 02201 02703  
Paecilomyces P. varioti 04645     
Penicillium P. roqueforti 04290 01614    
P. brevicompactum 03739 04660 04276   
P. verrucosum 03065 03384 04085 02109 03343
Phoma P. glomerata 03583     
Rhizopus R. oryza 02045 02047 02049 02087 02086

Sample preparation and FTIR measurements

In trouble shooting and microbial routine analysis of fungi, the initial step is an isolation procedure usually carried out on agar. In our protocol, this corresponded to a cultivation step on malt extract agar (MEA) (Oxoid, Basingstoke, UK) at 25°C for 5–6 days. Spore suspensions were prepared by collecting spores from the MEA plates with cotton tips followed by re-suspension in 1 ml malt extract broth (MEB, Oxoid). The second cultivation step – the microcultivation – was carried out in the automated Bioscreen C system (Oy Growth Curves AB), using liquid MEB as medium. The working volume of the spore suspension in the wells of the Bioscreen microplate (Oy Growth Curves AB) was 350 μl. Control wells needed for sterility test were filled with medium without spore suspension. Microcultivation was carried out at 25°C, with automatic continuous shaking, and the optical density of the fungal cultures was measured automatically at 540 nm every second hour, for five days.

Samples of the fungal mycelium were taken after 5 days of cultivation and measured by FTIR spectroscopy. Before FTIR measurements, each sample was prepared in the following way: (i) mycelium was transferred from the Bioscreen microplate with a bacterial loop to Eppendorf tubes and washed three times with 400 μl of distilled H2O, and (ii) to make the suspension of the fungal mycelium homogenous, mycelium was dissolved in 30–50 μl water and then sonicated in a tip-sonicator (Qsonica LLC, Newtown, CT, USA) for 20 s. Of each suspension, 8 μl was transferred to an IR-light-transparent Silicon 384-well microtitre plate (Bruker Optik GmbH, Germany). The samples were dried at room temperature for 45 min to form films suitable for FTIR analysis. FTIR measurements were performed using a High Throughput Screening eXTension (HTS-XT) unit coupled to a Tensor 27 spectrometer (both Bruker Optik GmbH). The spectra were recorded in the region between 4000 and 500 cm−1 with a spectral resolution of 6 cm−1 and an aperture of 5·0 mm. For each spectrum, 64 scans were averaged.

Experimental set-up

Two independent experiments were performed by two different operators (A and B). The first experiment was performed in 2009 and the second experiment in 2010.

The first experiment was performed in six independent cultivations, which were cultivated in different days. For each cultivation, two microcultivations were prepared in the Bioscreen C system for each of the 59 strains. From each microcultivation, two technical replicates were obtained for FTIR spectroscopy. The second experiment was performed in two independent cultivations, which were cultivated in different days as in the first experiment. All spectra were subjected to Opus spectral quality control test (Bruker® Analytik GmbH), and few technical replicates that did not pass the quality test were removed from the data set, resulting in a total of 1399 and 432 spectra for the first and the second experiment, respectively.

Data analysis

Cluster analysis

For the cluster analysis, the infrared spectra were preprocessed by taking first derivative spectra using a nine-point Savitsky–Golay algorithm and subsequently applying extended multiplicative signal correction (EMSC) (Kohler et al. 2009). For the EMSC correction, a replicate model was constructed as described by Kohler et al. 2009 (Kohler et al. 2009). To estimate the replicate variation, only technical replicates were used, that is, variation between spectra obtained from the same cultivation was estimated and build into the replicate correction model. In the EMSC replicate model, 5 principal components were used. The number of components was defined with respect to the explained variance; explained variance was 78% using 5 components.

To evaluate the natural clusters in the spectral data, hierarchical cluster analysis (HCA) and principal component analysis (PCA) were used. HCA was performed on the following spectra regions: 3000–2800 and 1800–900 cm−1. The construction of dendrogram was performed using Ward's linkage algorithm (Ward 1963). For calculating the spectral distance, the Euclidian distance was used.

Cluster analysis was performed by using in-house-developed algorithms in Matlab (The MathWorks Inc., Natick, MA, USA).

Artificial neural network-based FTIR analysis

Grouping information as obtained by the cluster analysis was used for building a hierarchy in the modelling scheme. A possible hierarchical structure was proposed and developed with artificial neural networks (ANNs). Prior to the ANN analysis, the spectral windows between 900 and 1800 cm−1 and 2800 and 3000 cm−1 were predefined in a data preprocessing step. The infrared spectra were preprocessed by taking first derivative spectra using a nine-point Savitsky–Golay algorithm. For the ANN analysis, 1399 spectra of the obtained data set were distributed into a training and test set. The training set contained data from first five independent cultivations and the test set contained data from the sixth cultivation. The training set of the first experiment contained 1168 spectra and the test set contained 231 spectra. The training set was used to build the model, and the test set was used for the validation. NeuroDeveloper software (Synthon, Heidelberg, Germany) was used to perform supervised feature selection and to establish a two-layer neural network with 61 input neurons, one hidden unit and two output units. The hierarchical decision tree as used in the algorithm is shown in Fig. 1. For each classification level, a fully connected feed-forward neural network was trained with the rprop algorithm (Rebuffo et al. 2006). The spectra form the second experiment were analysed all together by using the model, which was based on the training set from the first experiment.

Figure 1.

The hierarchical decision tree used in artificial neural network analysis.


Discrimination of food spoilage filamentous fungi by FTIR spectroscopy employing Hierarchical cluster analysis (HCA)

In this study, filamentous fungi of two of the most important phyla from the perspective of food spoilage, Zygomycota and Ascomycota, were analysed. The dendrogram from HCA of the average spectra for 1800–900 and 3000–2800 cm−1 regions is shown in Fig. 2, where the clustering is shown according to genus and species level. As seen from the dendrogram, all fungal strains clustered in two clusters, according to the phylum to which they belong: Cluster 1 contains species of genera Phoma, Fusarium, Penicillium, Paecilomyces, Aspergillus, Eurotium, Geotrichum and Alternaria; Cluster 2 contains species of genera Rhizopus and Mucor. The differentiation of fungi on the species level was studied for eight (of ten) genera (Fig. 2). The dendrogram showed frequently closer intraspecies clustering than interspecies clustering: (i) the dendrogram did not show clear clustering among Penicillium species, (ii) all strains of Eurotium amstelodami, Eurotium chevalieri and Eurotium herbariorum were clustering together with Aspergillus fumigatus strains: (i) all Eurotium amstelodami strains, except VI 04566, clustered between A. fumigatus VI 03680 and A. fumigatus VI 03042, (ii) E. herbariorum VI 02039, E. chevalieri VI 02041 and VI 02038 and Eurotium amstelodami VI 04566 clustered close to A. fumigatus VI 02980.

Figure 2.

Discrimination of filamentous fungi on genus, species and strain level, based on hierarchical cluster analysis (HCA) of FTIR spectra. For the HCA, average spectra were obtained by averaging all six independent cultivations and technical replicates.

Identification of food spoilage fungi by FTIR spectroscopy employing artificial neural network analysis (ANN)

Artificial neural network (ANN) analysis of 59 fungal strains, grown on MEB, was used to identify the strains on both species and genus level. Information about similarity of IR spectra was used to build the decision tree as shown in Fig. 1. Following this decision tree of different fungal species, we developed an ANN model with optimized classification through individual feature selection. The ANN model was used as a ‘supervised’ approach, which is performing classification of unknown samples into predefined groups based on a learning procedure. The basic architecture of the ANN model was built by breaking the whole training set into subunits, which represent main categories of spectral similarity. In Fig. 1 is shown which genera these subunits contain: ten major groups, which were used to establish the first level of the ANN model, determining the genus-specific networks (Fig. 1). When the first level was established, specialized networks were activated at the second level, determining species-specific subnetworks (Fig. 1).

Validation of ANN-based FTIR identification

The ANN-based FTIR identification model was validated based on two time-independent experiments, performed by two operators (operator A and operator B). The validation of the identification results obtained from operator 1 and operator 2 was performed by ANN on genus and species level for operator A and on genus level for operator B. The overall validation results obtained for operator A showed that the use of our protocol allowed the correct identification of 94·0% fungal strains on genus level (Table 2) and 93·9% of the strains on species level (Table 3). All strains of the species E. herbariorum and Mucor hiemalis and the genera Aspergillus and Penicillium were correctly indentified on species level. Less-satisfactory results were obtained for Fusarium graminearum, Mucor circinelloides and Eurotium amstelodami species with 95·0, 87·5 and 85·0% correct identification, respectively. A lower percentage of correct identification (80·0 and 75·0% respectively) was obtained for Fusarium langsethiae and E. chevalieri.

Table 2. Identification of food spoilage fungi on genus level by FTIR spectroscopy and artificial neural network analysis (ANN)
GenusIdentification Operator AaIdentification Operator Bb
  1. a

    Experiment performed in 2009.

  2. b

    Experiment performed in 2010.

  3. c

    Correctly identified/total number of samples.

Alternaria 15/15c32/32
Aspergillus 40/4079/86
Eurotium 27/3242/64
Fusarium 35/4062/70
Geotrichum 8/88/12
Mucor 31/3243/48
Phoma 3/46/6
Paecilomyces 3/46/6
Penicillium 36/3678/80
Rhizopus 20/2024/28
Correct identification (%)94·088·0
Table 3. Identification of food spoilage fungi on species level by FTIR spectroscopy and artificial neural network analyses (ANN) (operator A/2009)
SpeciesIdentificationCorrect identification (%)
  1. a

    Number of samples correctly identified at the species level/total number of samples analysed within that species.

Eurotium amstelodami 17/20a85·0
Eurotium chevalieri 6/875·0
Eurotium herbariorum 4/4100·0
Fusarium langsethiae 16/2080·0
Fusarium graminearum 19/2095·0
Aspergillus fumigatus 12/12100·0
Aspergillus niger 12/12100·0
Aspergillus favus 16/16100·0
Mucor plumbeus 12/12100·0
Mucor circinelloides 7/887·5
Mucor hiemalis 12/12100·0
Penicillium roqueforti 8/8100·0
Penicillium brevicompactum 10/10100·0
Penicillium verrucosum 18/18100·0

The validation results obtained for the operator B showed lower percentage correct identification on genus level (88%) (Table 2).


In the routine analysis in a food production process, it is often not sufficient to know whether fungi are present or not. The identification of the mycobiota is desired because this provides information on whether the detected fungi belong to the associated mycobiota of the product, the toxic properties of the fungi and may also be necessary to get an indication of the contamination route and measures to be taken. In the present investigation, the differentiation of various species of food spoilage fungi by FTIR spectroscopy has been demonstrated for the first time. The applied high-throughput liquid microcultivation and FTIR spectroscopy protocol for the identification of 59 strains of filamentous fungi showed successful differentiation of fungi on genus level (94·0% correct identification), species level (93·8% correct identification) and to some degree even on strain level. But the degree of correct identification for fungal species showed a strong variation from species to species (75·0–100%).

The set of fungi used in the study included two phyla Ascomycota and Zygomycota. A dendrogram of a HCA of the spectra of the 59 fungal strains showed a clear separation at genus level, while separation on species level in some cases was more pronounced within species of different genera then within species of the same genus (Fig. 2). The achieved differentiation of the fungi on the genus and species level reflected fungal taxonomy as discussed in the following.

Differentiation of food spoilage fungi by FTIR spectroscopy

The clustering of IR spectra, obtained by HCA analysis, has a strong connection to the fungal taxonomy. In particularly, clustering of the two genera Phoma and Fusarium, four genera Penicillium, Eurotium, Aspergillus and Paecilomyces and two genera Mucor and Rhizopus in three separate clusters (Fig. 2) can be explained by their belonging to the subphylum Pezizomycotina, family Trichocomaceae and family Mucoraceae, respectively.

The FTIR spectroscopy differentiation of Penicillium and Aspergillus species on the strain level showed low interstrain and high intrastrain variability, respectively (Fig. 2); similar results were obtained by Fischer et al. during their study of the application of FTIR spectroscopy for the identification and intraspecies characterization of airborne filamentous fungi (Fischer et al. 2006). High intrastrain variability of Aspergillus strains shown by FTIR spectroscopy might be due to the fact that the species A. fumigatus, A. niger and A. flavus belong to three different subgenera: Fumigati, Nigri and Flavi, respectively.

The obtained identification results for food spoilage fungi by the given high-throughput liquid microcultivation and FTIR spectroscopy protocol support the findings from phylogenetic studies of Ascomycota and Zygomycota (Schwarz et al. 2006). For example, eight strains of the three Eurotium species E. amstelodami, E. herbariorum and E. chevalieri are clustering close to Aspergillus flavus and A. fumigatus, but they are well separated from other strains (Fig. 2), supporting the findings from RAPD-PCR analysis of different Eurotium species (Gherbawy 2001), which showed distinct clustering of E. amstelodami, E. herbariorum and E. chevalieri. The identification of Aspergillus strains based on using 18S rDNA, 28S rDNA and ITS regions have been examined in recent years. FTIR spectroscopy, compared to genetic methods, allows identifying Aspergillus strains, which was proved by Fischer et al. In our study, we achieved 100% correct identification of Aspergillus on the species level. The identification of Zygomycota by our protocol also reflects the identification results obtained by different phylogenetic methods. For example, the identification of Mucor species reflects the finding of the phylogenetic analyses performed by Voigt and Wöstemeyer 2001.

An alternative method for characterization of spoilage fungi in food industry

Today, the identification of fungi in the food industry is carried out mostly by traditional morphological methods and rarely by genetic methods. Prior to the analysis by any microbiological method, sampling is performed by the industry according to Hazard Analysis of Critical Control Points (HACCP) routines. The starting point for our protocol is also after the sampling according to HACCP routines, when pure cultures are readily obtained. As FTIR spectroscopy combined with microcultivation has a high throughput and is less expensive than traditional morphological and phylogenetic methods, it provides a good alternative for routine analysis. The main problem concerning traditional morphological methods for the identification of fungi is they are time-consuming. The time needed until identification in our new protocol for FTIR analysis is 10 days, faster than traditional morphological methods, but longer than genetic methods.

The fact that identification results for the operator B were worse than for operator A reveals that the analysis of filamentous fungi by FTIR spectroscopy is dependent on standardization of the analysis. Because fungi are not unicellular as bacteria, they need longer cultivation time and multistage sample preparation procedure that make the FTIR identification protocol for fungi much more complex and sensitive than for bacteria and yeasts. In our study, the comparison and validation of the FTIR identification results for fungi obtained from two operators were performed for the first time and showed that the identification of filamentous fungi needs not only strictly controlled cultivation and sample preparation procedure, but such variability factor as ‘operator’ should be taken in account and minimized by automating different steps of the developed identification protocol.

Our protocol opens a wide range of possibilities for further necessary improvements: (i) the possibility to use different media in one cultivation run may increase the identification power, (ii) the possibility to make high-throughput liquid cultivation shorter decreases the identification time, and (iii) the possibility of automating by developing robotics.

In the food industry most often, only a selected range of fungi is of interest. The developed protocol gives an opportunity to establish tailor-made databases for the relevant associated mycobiotas, which might also contain all information about sampled fungi, microscopic images, FTIR spectra, that might be especially useful for source tracking of the fungal contamination along the food production chain.

The study presented in this article has demonstrated that FTIR spectroscopy can be used for the identification of certain filamentous food spoilage fungi. The method has a potential in food industries that have moulds as a dominant contaminant agent, for example, juice and soft drink industry, cheese industry and bakery. In addition, it might be applicable in pharmacy. In hospitals, where a high variation of micro-organisms is present and where genetically very close samples are present, FTIR spectroscopy may have a somewhat lower potential. In addition, our new protocol depends on the fact that media are used that enable growth of the targeted mould strains. In an environment like juice industry, the media can be chosen to be close to the juice matrix. While in more heterogeneous environments like hospitals, is it more difficult to know which media to use. On the other hand, different media could be used that support growth of the different moulds.

For survey purpose in the food industry, the presented protocol may have a similar identification power as molecular methods and a higher identification power than traditional morphological methods. In hospitals, the discrimination power might not be good enough and molecular methods may be favoured. Our results were validated by two independent test sets, which were acquired with one-year time difference to the training set. One of the test sets was acquired by a new operator. The identification level obtained by our protocol was high both on genus and on species level. In addition, the obtained clustering and separation patterns could be explained by other morphological and phylogenetic studies. That indicates a high potential for the application of the high-throughput liquid microcultivation and FTIR spectroscopy protocol in source tracking of fungal contamination and for routine survey of selected fungi in the food industry.


This work was supported by the grant 186909/I10 financed by the Research Council of Norway. Financial support from the Agricultural Food Research Foundation of Norway is also greatly acknowledged.