SEARCH

SEARCH BY CITATION

Keywords:

  • Epichloë;
  • fungal endophytes;
  • multivariate analysis;
  • visible-near infrared (Vis-NIR) reflectance spectroscopy

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results and discussion
  6. Conclusions
  7. Acknowledgements
  8. References

The aim of this work was to investigate the potential of visible and near-infrared (Vis-NIR) reflectance spectroscopy for the classification of three morphologically similar species of fungal endophytes of grasses. Vis-NIR spectra (400–2498 nm) from 34 isolates of Epichloë sylvatica, 32 of Epichloë typhina and 38 of Epichloë festucae were recorded directly from fresh mycelium growing in potato dextrose agar plates. Multivariate procedures applied to the spectral data were discriminant modified partial least squares regression, soft independent modelling of class analogy and discriminant partial least squares regressions (PLS1, PLS2). Several types of data pretreatments were tested to develop the classification models. The best predictive models were achieved with PLS2 analysis; with this method, 90% of E. typhina and 100% of E. festucae and E. sylvatica external validation samples were successfully classified. These results show the potential of Vis-NIR spectroscopy combined with multivariate analysis as a rapid method for classifying morphologically similar species of filamentous fungi.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results and discussion
  6. Conclusions
  7. Acknowledgements
  8. References

The earliest systems of classification of fungal species relied on the morphological characters, mainly those of reproductive structures. This method of classification has limitations: for example, when fungal cultures are sterile and do not develop reproductive structures, or in situations where members of different species are very similar morphologically. The incorporation of biochemical and molecular characters (e.g. isozymes, nucleotide sequences) to fungal taxonomy helped, at least in some cases, to overcome such problems (Alexopoulos et al., 1996; Guarro et al., 1999).

The purpose of the present study was to test if visible and near-infrared (Vis-NIR, 400–2498 nm) reflectance spectroscopy could be used to distinguish among fungal species whose cultures are morphologically identical. Other spectroscopic techniques such as Fourier transform infrared spectroscopy (FT-IR, 500–4000 cm−1) have been used to discriminate and classify filamentous fungi (Fischer et al., 2006) and yeasts (Kümmerle et al., 1998; Mariey et al., 2001).

NIR reflectance spectroscopy is a procedure that can detect differences in the chemical composition of biological materials based on the absorption of radiation by chemical bonds, chiefly C–H, O–H and N–H. These bonds generally have fundamental absorptions in the mid-infrared region (500–4000 cm−1), which are strong enough to produce overtones and combination bands that are detectable in the NIR region (750–2500 nm) (Osborne et al., 1993). Sample preparation is simple for this technique, no chemicals are used and analytical costs are lower than those of some methods used for fungal identification such as DNA sequencing.

For this test we have used three species belonging to the genus Epichloë. Epichloë sylvatica, Epichloë typhina and Epichloë festucae infect grasses and only cause symptoms when the plants develop inflorescences. These three species are adequate for this study because their cultures cannot be distinguished by their macroscopic (colour, growth rate) or microscopic (conidia or conidiophore size or shape) morphological characteristics. However, Epichloë species have been distinguished on the basis of molecular characters such as the nucleotide sequences of internal transcribed spacers, or β-tubulin gene introns, and the fertility of crossings made with members of the same species (Craven et al., 2001; Schardl et al., 2004).

To our knowledge this is the first time that the ability of Vis-NIR spectroscopy to identify different species of filamentous fungi has been evaluated.

Materials and methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results and discussion
  6. Conclusions
  7. Acknowledgements
  8. References

Plant sampling and endophyte isolation

Plants of Brachypodium sylvaticum, Dactylis glomerata and Festuca rubra were collected at 10 locations in the provinces of Salamanca and La Coruña in western and northern Spain, respectively. At each location, a distance of at least 10 m was left between the sampled plants. Plants of D. glomerata infected by E. typhina were recognized in the field by the presence of fungal stromata in their reproductive stems (Sampson, 1933). As the infection of E. festucae and E. sylvatica is often asymptomatic in their host plants, isolates of these endophytes were obtained from apparently healthy plants of F. rubra and B. sylvaticum (Leuchtmann et al., 1994; Leuchtmann & Schardl, 1998). In these plants the presence of endophytes were determined by microscopic examination of stem pith samples stained with 1% aniline blue (Bacon & White, 1994).

Pure cultures of Epichloë endophytes were obtained from surface-disinfected pieces of stems and leaf sheaths plated on Petri dishes containing potato dextrose agar (PDA) (Bacon & White, 1994). All PDA plates used in this study were prepared at the same time and from the same batch of media. Therefore, they may be expected to be similar in composition. The morphological characters such as conidiophores and conidial size and shape were used for the taxonomic identification of the isolated endophytes in the genus Epichloë (White, 1993; Leuchtmann et al., 1994; Leuchtmann & Schardl, 1998).

Epichloë typhina is the only Epichloë species known to infect D. glomerata, E. festucae is the only species reported in F. rubra and B. sylvaticum is the only known host of E. sylvatica (Leuchtmann et al., 1994; Leuchtmann & Schardl, 1998; Clay & Schardl, 2002). Therefore, after morphological characters of cultures were used to identify the isolates to genus rank, host plant confirmed the species of each isolate.

Spectral collection

Five cultures of each isolate were prepared on Petri dishes containing PDA and maintained in the dark at 22–25 °C. Four weeks later, spectral recordings were made in random order. The reflectance spectra were collected between 400 and 2498 nm on an NIRSystem 6500 scanning monochromator (Foss NIRSystems, Silver Spring, MD) fitted with a sample transport module. Conventionally, the spectral data are recorded as log(1/R) where R is the intensity of reflected light at each wavelength – this convention derives from Beer's law, which states that the concentration of an absorbing species is related to the logarithm of the relative intensity of incident, monochromatic radiation. A 35-mm-diameter disc of mycelium and agar was cut from each culture plate and placed on a standard circular quartz reflectance cell. The temperature was kept at 24±3 °C in the laboratory during NIR measurements. Two spectra were recorded for each culture and the mean spectrum obtained from the five cultures of each isolate was used for subsequent data analysis. A total of 104 mean spectra (34 isolates of E. sylvatica, 32 of E. typhina and 38 of E. festucae), were stored in log(1/R) format (R=reflectance). Spectral collection took place over a period of 5 days. Instrument control and spectral file manipulation was performed using winisi III software (v. 1.50e, Infrasoft International, State College PA).

Data pretreatment

Because NIR spectra are affected by sample particle size, light scatter and consequent pathlength variation, pretreatments of the spectral data are often explored in an effort to improve calibration accuracy. In order to obtain the best discrimination model, nine different types of spectral pretreatments were tested. These included no treatment (raw spectral data), multiplicative scatter correction (MSC), standard normal variate (SNV) or standard normal variate and detrending (SNVD) and a combination of the mentioned pretreatments with first (1D) and second (2D) derivative transformations. These methods have found common application in NIR spectroscopy because they may remove the main extraneous complicating factors in reflectance spectra, i.e. light scatter and a sloping baseline. These phenomena arise from differences in sample particle size and variable spectral pathlengths deriving from variations in sample packing density. With regard to MSC, the spectrum of each sample is regressed against the mean spectrum of the collection; coefficients thus derived are used to correct for differences in scatter among the samples by application to each sample spectrum (Geladi et al., 1985). The equation used for transformation of individual spectra is as follows:

  • image

where X1 is the corrected spectrum and â, ĉ and Ê are regression coefficient estimates and X2 is the mean spectrum. SNV performs a similar function but in this case each individual spectrum is centred and scaled (Barnes et al., 1989). In the case of a spectrum comprising measurements at 700 wavelengths, the individual SNV-treated spectrum is calculated as follows:

  • image

where SNV(1–700) are the individual standard normal variations for 700 wavelengths, y is the 700-wavelength log(1/R) values and ȳ is the mean of the 700-wavelength log(1/R) values. The detrending step involves the application of a second-degree polynomial to standardize variations in spectral curvilinearity. The application of derivative mathematical pretreatments to spectral data removes additive baseline effects (first derivative) or a linear baseline (second derivative) (Næs et al., 2002). Particular derivative parameters used with winisi III software were 1D (1,4,4,1) and 2D (2,12,2,2), in which the first number is the order of the derivative function, the second is the segment length in data points over which the derivative was taken, and the third (first smooth) and fourth (second smooth) are the segment lengths over which the function was smoothed (Shenk & Westerhaus, 1995). When the unscrambler software was used, data were transformed into 1D and 2D using the Savitzky–Golay derivative (5,5,1) and (6,6,2) (left and right number of side points and polynomial order, respectively, Savitzky & Golay, 1964). Different software packages were used because the data analysis options in each are different.

Multivariate data analysis

Discriminant modified partial least squares regression (MPLS) was carried out in winisi III software. Raw spectra were exported in JCAMP.DX format and imported into the unscrambler (v. 9.6; CAMO A/S, Oslo, Norway) to develop soft independent modelling of class analogy (SIMCA) and discriminant partial least squares regressions (PLS1 and PLS2). The accuracy of classification models using The unscrambler software was assessed on the basis of the percentages of correct classification (taking into account false negatives) and the percentages of false positives. A false-positive result occurs when a sample is wrongly identified as belonging to a specific class; conversely, a false-negative result occurs when a sample that does belong to a class is not classified as such (Downey & Daniel Kelly, 2004). False-positive results may be considered the more serious of the two because they result in failure to identify Epichloë species.

All classification models were developed using separate calibration and validation sample sets. Samples were assigned to each set based on their position in the spectral file; thus, every third sample was used for external validation (n=33; 11 samples of E. sylvatica, 10 samples of E. typhina and 12 of E. festucae) and the remainder for calibration (n=71; 23 of E. sylvatica, 22 samples of E. typhina and 26 of E. festucae).

Principal component analysis (PCA) was performed before classification for a preliminary data inspection to detect unusual spectra and to highlight sample clustering if any (Martens & Næs, 1996). Four multivariate techniques were used to differentiate samples on the basis of the corresponding Epichloë species at the 95% confidence level. In the case of techniques based on PLS regressions (MPLS, PLS1 and PLS2), dummy variables were used for statistical analysis. Thus, for MPLS analyses, samples of E. sylvatica were assigned a dummy value 1, E. typhina the value 2 and E. festucae the value 3. For PLS1 regression, a series of three binary classifications was applied with the aim of segregating each Epichloë species from the others (e.g. E. typhina vs. the others); there were only two classes to separate with one y variable for which 1 indicated an in-class member (the Epichloë species to discriminate) and 0 indicated a nonclass member (the other two species). In PLS2, the number of y variables equals the number of classes to be discriminated. Therefore, in this case, a new y variable was created for each Epichloë species studied; within each of these variables, one variety was assigned a dummy variable equal to 1 and all the others given the value 0. The optimal number of factors in the regression model was determined by full cross-validation; these models were then used to predict the dummy values of samples in the validation set from their spectral data. In SIMCA, each Epichloë species was modelled separately by PCA on the calibration sample sets. External validation samples were classified by predicting them with each of these three SIMCA models; the decision as to whether a sample belongs to a particular PCA model or not was based on the distance from the new sample to the centre of the model and the distance from the new sample to the model. In all of these techniques, it is necessary to develop models using class information about samples in the calibration sample set; the predictive ability of these models is determined on a separate sample set, which also includes this class information. Once efficient and effective models have been created, they may then be deployed to predict the identity of unknown samples. More detailed descriptions of these multivariate techniques can be found in the literature (Esbensen, 2002; Næs et al., 2002; Miller & Miller, 2005). Three wavelength ranges [400–2498 nm (Vis-NIR region), 1100–2498 nm (NIR region) and 750–1098 nm (short-wavelength NIR region)] were investigated by PLS2 and SIMCA methods in order to find optimal spectral segments.

Results and discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results and discussion
  6. Conclusions
  7. Acknowledgements
  8. References

Sample spectra

Figure 1 shows the visible and NIR mean spectra (raw and first derivative transformed) of the three fungal species; the break at 1100 nm in each spectrum is an artefact caused by a detector change at this wavelength. The main features of the raw mean spectra arise from water absorption. Water is the main component of fresh samples and poses a limitation in the use of Vis-NIR spectroscopy because it absorbs strongly and contributes to a significant amount of light scattering. The presence of water in a sample gives rise to characteristic absorption bands but also affects the overall spectrum as the scattering depends on the ratio of the refractive index of the particles to that of the surrounding medium (Osborne & Fearn, 1986). At 20 °C, pure water has maxima at 970, 1190, 1450 and 1940 nm (Segtnan et al., 2001).

image

Figure 1.  Vis-NIR mean spectra of the three Epichloë species studied: (a) raw spectra [log(1/R)], (b) first derivative transformation of the raw spectral data.

Download figure to PowerPoint

Along the whole spectrum, E. sylvatica showed absorbance values higher than those of E. festucae and E. typhina. These mean differences are removed by the first derivative transformation, which reveals greater spectral detail and differences in absorbance by the three species at particular wavelengths (Fig. 1).

Classification of Epichloë species

All the fungal isolates used for the study were identified as Epichloë species on the basis of morphological characters such culture colour and appearance, and the shape and size of conidia and conidiophores. The only Epichloë species known to infect D. glomerata, F. rubra and B. sylvaticum are E. typhina, E. festucae and E. sylvaticum, respectively (Clay & Schardl, 2002). Therefore, isolate identification to species rank was based in the grass host species.

All Vis-NIR spectra of the three Epichloë species were used for PCA, a technique commonly used visually to detect the presence of outliers and clusters in a sample set. Figure 2 shows a bidimensional representation of PC3 and PC6 scores for all samples labelled according to their Epichloë species; we chose these PCs for the graphical representation because they showed the best separation of the species studied. In this figure, some clustering of the samples according to fungal species is apparent although the separation of clusters is incomplete. This degree of clustering suggests, however, that it may be possible to discriminate between the three species using more powerful and specific discriminant mathematical techniques. No obvious outlying samples were detected and therefore all the samples (n=104) were included in calibration modelling. Given that PCA calculates sample scores on the basis of spectral information only, incomplete segregation by this technique is not uncommon with complex biological materials.

image

Figure 2.  Scores plot of the PC3 vs. PC6 from all 104 Epichloë samples using the Vis-NIR spectra (red, Ef, Epichloë festucae; green, Et, Epichloë thypina; blue, Es, Epichloë sylvatica).

Download figure to PowerPoint

A summary of the results obtained with the MPLS, SIMCA, PLS1 and PLS2 models is given in Tables 1–4. These results were obtained using 33 samples, which had not been used to develop the calibrations. Three different wavelength ranges (400–2498, 750–1098 and 1100–2498 nm) were evaluated for the SIMCA and PLS2 methods. The Vis-NIR region (400–2498 nm) generally produced best accuracy of the models; consequently, only the results for this wavelength region are shown (Tables 2 and 4). Several authors have investigated discrete regions in FT-IR spectra that contain information from different cell components (such as lipids, proteins, polysaccharides and nucleic acids) of microorganisms (Naumann et al., 1988, 1991; Kümmerle et al., 1998; Lin et al., 2005). However, the present study aimed to explore the possibility of using Vis-NIR spectra to assign unknown samples to the correct Epichloë species. To our knowledge there is no evidence about which spectral ranges are optimal for species discrimination in this 400–2498 nm wavelength region.

Table 1.   MPLS classification (% correct) for Epichloë typhina, Epichloë festucae and Epichloë sylvatica external validation samples
Data pretreatmentE. typhina (n=10)E. festucae (n=12)E. sylvatica (n=11)
  • Bold type indicates the best classifications obtained for each species.

  • *

    Multiple scatter correction.

  • Standard normal variate and detrending.

  • First derivative.

  • §

    § Second derivative.

Raw80.058.345.5
MSC*70.058.363.6
SNVD80.066.772.7
1D90.058.372.7
MSC*+1D10066.763.6
SNVD+1D90.058.363.9
2D§80.066.772.7
MSC*+2D§90.066.754.5
SNVD+2D§10066.754.5
Table 2.   SIMCA classification for Epichloë typhina, Epichloë festucae and Epichloë sylvatica external validation samples
Data pretreatmentE. typhina (n=10)E. festucae (n=12)E. sylvatica (n=11)
% Correct classification% False positives% Correct classification% False positives% Correct classification% False positives
  • Bold type indicates the best classifications obtained for each species.

  • *

    Multiplicative scatter correction.

  • Standard normal variate.

  • First derivative.

  • §

    § Second derivative.

Raw9052.27571.47350
MSC*10039.18380.98245.4
SNV9043.58380.97340.9
1D9065.28380.97363.6
MSC*+1D10021.783100739.1
SNV+1D10060.9831006459.1
2D§10082.68395.28272.7
MSC*+2D§10065.2831007359.1
SNV+2D§10078.3921008272.7
Table 3.   PLS1 binary classification to segregate each Epichloë species (Et, Epichloë typhina; Ef, Epichloë festucae; Es, Epichloë sylvatica) from the others
Data pretreatmentEt (n=10) vs. Ef−EsEf (n=12) vs. Et−EsEs (n=11) vs. Et−Ef
% Correct classification% False positives% Correct classification% False positives% Correct classification% False positives
  • Bold type indicates the best classifications obtained for each species.

  • *

    Multiplicative scatter correction.

  • Standard normal variate.

  • First derivative.

  • §

    § Second derivative.

Raw908.7929.5919.1
MSC*9013924.8829.1
SNV908.71004.87313.6
1D908.710008213.6
MSC*+1D9013924.89113.6
SNV+1D9013924.89113.6
2D§908.71004.87313.6
MSC*+2D§908.79208213.6
SNV+2D§908.7924.8829.1
Table 4.   PLS2 classification for Epichloë typhina, Epichloë festucae and Epichloë sylvatica external validation samples
Data pretreatmentE. typhina (n=10)E. festucae (n=12)E. sylvatica (n=11)
% Correct classification% False positives% Correct classification% False positives% Correct classification% False positives
  • Bold type indicates the best classifications obtained for each species.

  • *

    Multiple scatter correction.

  • Standard normal variate.

  • First derivative.

  • §

    § Second derivative.

Raw908.71004.8829.1
MSC*904.31004.8829.1
SNV904.31004.8829.1
1D908.710008213.6
MSC*+1D908.71000824.5
SNV+1D908.71004.8829.1
2D§908.710008213.6
MSC*+2D§904.39214.31004.5
SNV+2D§904.39214.3824.5

The results obtained by the MPLS method are shown in Table 1. In the case of models produced using MSC+1D and SNVD+2D, a 100% correct classification rate was achieved for E. typhina validation samples. However, the results obtained for E. festucae and E. sylvatica differentiation were not satisfactory as, in general, no more than 70% of the external validation samples were correctly classified.

In the case of the SIMCA, PLS1 and PLS2 methods (Tables 2–4), the best models for each Epichloë species were selected according to the highest correct classification rates and the lowest percentage of false positives (Woodcock et al., 2007).

The SIMCA approach was not sensitive enough to be of practical utility in this particular application because it produced a high rate of false positives (Table 2). This agrees with the results obtained by Oust et al. (2004) for the identification of closely related species of Lactobacillus using FT-IR; these authors found that SIMCA models were not as good for identification as those developed by PLS regression.

A third discriminant approach that produced better results than SIMCA was the PLS1 method. It can be seen in Table 3 that PLS1 models identified Epichloë species with a degree of accuracy ranging from 90% (E. typhina vs. E. festucae and E. sylvatica) to 100% (E. festucae vs. E. typhina and E. sylvatica). The best model for the binary classification of E. festucae from the others involved the use of 1D pretreatment and correctly classified 100% of samples with no false-positive classifications.

Of all the methods of classification considered, the most accurate models were obtained with the PLS2 method because it produced the best classifications with the lowest levels of false positives (Table 4). The correct classification of 100% of E. festucae samples with no false positives using three different data pretreatments (1D, MSC+1D and 2D) was a highly satisfactory result. The corresponding classification rate of E. sylvatica was also very high (100%; MSC+2D) and with a relatively low percentage of false positives (4.5%). The only species that did not show the 100% correct discrimination was E. typhina. In this case 90% of the external validation samples were correctly classified with several data pretreatments and low percentages of false positives (4.3%) (Table 4).

Conclusions

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results and discussion
  6. Conclusions
  7. Acknowledgements
  8. References

On the basis of all results obtained, discriminant PLS2 analysis was the best method for the classification of the three species of Epichloë considered.

This research shows that Vis-NIR spectroscopy can be used to identify morphologically similar species of a filamentous fungus. Although three Epichloë species were used as experimental material, this technique might be useful in similar situations with other fungal species.

Morphological similarity is quite common among the fungal species belonging to the same genus, and it is often necessary to use molecular techniques such as sequencing to identify to species level. DNA sequencing methods may be better than the spectroscopy method used here for species classification when a rigorous identification is required. However, in other types of work such as surveys of fungal incidence in hosts where more than one Epichloë species is known to occur (e.g. Lolium perenne, Clay & Schardl, 2002), or screenings of large numbers of isolates in situations where several morphologically similar species are studied, the Vis-NIR method could present substantial advantages.

To our knowledge, this is the first time that Vis-NIR spectroscopy has been used for the identification of different fungal species. The combination of Vis-NIR spectroscopy and multivariate techniques could be a complementary, rapid and reliable tool in microbiological studies for screening and discriminating morphologically similar fungal species.

Acknowledgements

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results and discussion
  6. Conclusions
  7. Acknowledgements
  8. References

This work was supported by the project AGL2005-02839/AGR of the MECD (Spain). C.P. acknowledges financial support from the FPU predoctoral grant awarded by the MECD (Spain). C.P. also wishes to thank G.D. and I.M. for welcome to Ashtown Food Research Centre (Ireland) and Scottish Agricultural College (UK).

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results and discussion
  6. Conclusions
  7. Acknowledgements
  8. References
  • Alexopoulos CJ, Mims CW & Blackwell M (1996) Introductory Mycology, 4th edn. John Wiley & Sons Inc, New York.
  • Bacon CW & White JF (1994) Strains, media and procedures for analysing endophytes. Biotechnology of Endophytic Fungi of Grasses (BaconCW & WhiteJF, eds), pp. 4756. CRC Press, Boca Raton, FL.
  • Barnes RJ, Dhanoa MS & Lister SJ (1989) Standard normal variate transformation and de-trending of near-infrared diffuse reflectance spectra. Appl Spectrosc 43: 772777.
  • Clay K & Schardl CL (2002) Evolutionary origins and ecological consequences of endophyte symbiosis with grasses. Am Nat 160S: 99127.
  • Craven KD, Hsiau PTW, Leuchtmann A, Hollin W & Schardl CL (2001) Multigene phylogeny of Epichloë species, fungal symbionts of grasses. Ann Missouri Bot Gard 88: 1434.
  • Downey G & Daniel Kelly J (2004) Detection and quantification of apple adulteration in diluted and sulfited strawberry and raspberry purées using visible and near-infrared spectroscopy. J Agric Food Chem 52: 204209.
  • Esbensen KH (2002) Multivariate Data Analysis in Practice. An Introduction to Multivariate Data Analysis and Experimental Design, 5th edn. CAMO Process AS, Norway.
  • Fischer G, Braun S, Thissen R & Dott W (2006) FT-IR spectroscopy as a tool for rapid identification and intra-species characterization of airbone filamentous fungi. J Microbiol Methods 64: 6377.
  • Geladi P, McDougall D & Martens H (1985) Linearisation and scatter correction for near infrared reflectance spectra of meat. Appl Spectrosc 39: 491500.
  • Guarro J, Gené J & Stchigel A (1999) Developments in fungal taxonomy. Clin Microbiol Rev 12: 454500.
  • Kümmerle M, Scherer S & Seiler H (1998) Rapid and reliable identification of food-borne yeast by Fourier-transform infrared spectroscopy. Appl Environ Microb 64: 22072214.
  • Leuchtmann A & Schardl CL (1998) Mating compatibility and phylogenetic relationships among two new species of Epichloë and other congeneric European species. Mycol Res 102: 11691182.
  • Leuchtmann A, Schardl CL & Siegel MR (1994) Sexual compatibility and taxonomy of a new species of Epichloë symbiotic with fine fescue grasses. Mycologia 86: 802812.
  • Lin M, Al-Holy M, Chang S, Huang Y, Cavinato AG, Kang D & Rasco BA (2005) Rapid discrimination of Alicyclobacillus strains in apple juice by Fourier transform infrared spectroscopy. Int J Food Microbiol 105: 369376.
  • Mariey L, Signolle JP, Amiel C & Travert J (2001) Review: discrimination, classification, identification of microorganisms using FTIR spectroscopy and chemometrics. Vib Spectrosc 26: 151159.
  • Martens H & Næs T (1996) Multivariate Calibration. John Wiley & Sons Inc, New York.
  • Miller JN & Miller JC (2005) Statistics and Chemometrics for Analytical Chemistry, 5th edn. Pearson Education Limited, London.
  • Næs T, Isaksson T, Fearn T & Davies T (2002) A User-Friendly Guide to Multivariate Calibration and Classification. NIR Publications, Chichester, UK.
  • Naumann D, Fijala V, Labischinski H & Giesbrecht P (1988) The rapid differentiation and identification of pathogenic bacteria using Fourier transform infrared spectroscopic and multivariate statistical analysis. J Mol Struct 174: 165170.
  • Naumann D, Helm D & Labischinski H (1991) Microbiological characterizations by FT-IR spectroscopy. Nature 351: 8182.
  • Osborne BG & Fearn T (1986) Near Infrared Spectroscopy in Food Analysis. Longman Group, London.
  • Osborne BG, Fearn T & Hindle PT (1993) Practical NIR Spectroscopy with Applications in Food and Beverage Analysis, 2nd edn. Longman Group, London.
  • Oust A, Moretro T, Kirschner C, Narvhus JA & Kohler A (2004) FT-IR spectroscopy for identification of closely related lactobacilli. J Microbiol Methods 59: 149162.
  • Sampson K (1933) The systemic infection of grasses by Epichloë typhina (Pers.) Tul. Trans Br Mycol Soc 18: 3047.
  • Savitzky A & Golay MJE (1964) Smoothing+differentiation of data by simplified least squares procedures. Anal Chem 36: 16271639.
  • Schardl CL, Leuchtmann A & Spiering MJ (2004) Symbioses of grasses with seedborne fungal endophytes. Annu Rev Plant Biol 55: 315340.
  • Segtnan BH, Sasic S, Isaksson T & Ozaki Y (2001) Studies on the structure of water using two-dimensional near-infrared correlation spectroscopy and principal component analysis. Anal Chem 73: 31533161.
  • Shenk JS & Westerhaus MO (1995) Routine Operation, Calibration, Development and Network System Management Manual. NIRSystems Inc, Silver Spring.
  • White JF Jr (1993) Endophyte–host associations in grasses. XIX. A systematic study of some sympatric species of Epichloë in England. Mycologia 85: 444455.
  • Woodcock T, Downey G, Daniel Kelly J & O'Donnell C (2007) Geographical classification of honey samples by near-infrared spectroscopy: a feasibility study. J Agric Food Chem 55: 91289134.