J. Chil. Chem. Soc, 55, N° 1 (2010), pág: 11-18

 

CLASSIFICATION OF GENOTYPES OF EUCALYPTUS GLOBULUS UNDER COLD CONDITIONS USING THEIR FREE AMINO ACIDS CONTENT ON LEAVES AND REGULARIZED DISCRIMINANT ANALYSIS (RDA)

 

ROSARIO DEL P. CASTILLO^1,2*, DAVID CONTRERAS ², JAME BAEZA ^1,2, MATTHIAS OTTO ³, CRISTIAN AGURTO¹, JUANITA FREER^1,2*

¹ Biotechnology Center, University of Concepción, Concepción — Chile E-mail: - Chile
² Faculty of Chemical Sciences, University of Concepción, Concepción — Chile.
³ Institute of Analytical Chemistry, TU Bergakademie Freiberg, Freiberg — Germany.

---

ABSTRACT

Eucalyptus globulus is an important forestry species for wood and pulp industry; however, its low cold resistance is a serious disadvantage. In this work, chemometric techniques were applied onthe leaves'free amino acid content of different E. globulus genotypes in orderto select cold tolerant genotypes. Pattern recognition methods based on regularized discriminant analysis (RDA) were used to classify the genotypes by cold resistance using samples' foliar damages as reference of cold resistance. Correlation between the amino acids content and cold resistance of genotypes was determined from a partial least squares (PLS) model. RDA models showed an excellent classification of genotypes (100% of correctly assigned genotypes in external validation), especially when RDA was reduced to quadratic discriminant analysis (QDA). Arg, Tyr and Ala, showed the highest correlations with the foliar damage in the PLS model and they could be responsible to improve cold resistance to the studied genotypes.

Keywords: regularized discriminant analysis (RDA), amino acids, cold resistance, Eucalyptus globulus.

---

INTRODUCTION

Chemometric tools, like pattern recognition and multivariate calibration methods, have applications in innumerable áreas including food industry, environmental analysis, pharmaceutical industry and others. Applications inthe forest industry are presently increasing, although most of the reported studies in this área are developed with the objective to quantify and predict the wood's chemical or physical properties ^1-6. Fewer studies using pattern recognition methods have been reported in this área^7-11 where unsupervised methods, such as principal component analysis (PCA), and some supervised methods like son modeling class analogy (SIMCA) and k-nearest neighbor (KNN) have been used for the classification and categorical prediction of new samples according to their wood physical properties, wood chemical composition or species classification ^7-11,18,12.

In this work, a statistical and a chemometric analysis using pattern recognition methods of the free amino acids of Eucalyptus globulus leaves is used to select cold tolerant genotypes of this species.

Eucalyptus globulus Labill is one of the most important forest species in Chile and the Southern Hemisphere, especially due to its extensive use in the pulp mili industry. This species is coveted for its rapid growth, high pulp yield and easy adaptation to new environments; however E. globulus has an important drawback: it is susceptible to cold conditions ¹³. As a result, this species is often replaced by other species in low temperature regions, often reducing wood quality and pulp yield. Propagation programs are presently being developed using clonal selection of genotypes in order to increase tolerant populations. Thus, the identification of these tolerant genotypes using physiological, chemical or genetic information is needed.

Cold tolerance in plants is still a little known process; however principal theories suggest that under cold conditions the tolerant plants support ice formation in their tissues possibly accompanied by changes in the chemical composition inthe cell, such as variation inthe content of soluble carbohydrates, free amino acids and abscisic acid¹⁴. Application of acclimatization processes can increase cold resistance in some plants ¹⁵. Cold acclimatization can be performed by expositions to low temperatures prior to colder or frost conditions.

The relationship between the cold tolerance at frost temperatures (freezing tolerance) and some chemical constituents, principally carbohydrates, has been studied in many plant species. Dionne et al. (2001)¹⁶ demonstrated an accumulation of sucrose and high concentrations of fructans in crowns of Poa annua L. ecotypes after freezing temperatures exposures; however these compounds were not related with the freezing tolerance presented by the ecotypes. Some others works such as the studies over barley crops ¹⁷, roots of Lonicera caerulia¹⁸, and cabbage seedlings¹⁹ show increments of sucrose, fructose and raffinose; but in transgenic Arabidopsis thaliana, the accumulation of raffinose by itself was not related to its cold tolerance²⁰, while a positive correlation between this sugar and cold tolerance was demonstrated in Petunia hybrida²¹. These results indícate that the tolerance depends on the species, involves organs, and that there is no agreement with respect to the accumulation of specific sugars.

There are fewer reports in the literature about studies of amino acid content in plants under cold conditions. Most of those reports refer to the accumulation of free proline (Pro)^14,22. Meza-Basso et al. (1986)²³ presented evidence of a probable cryo-protector role of free Pro in Nothophagus dombeyi leaves due to its increase in fall and winter and its decrease in summer and spring. Several others works^24-26 found relationships between accumulation of Pro and freezing tolerance, although other results suggest that Pro increments are due to the exposure to cold temperatures rather than an enhanced freezing tolerance^22,27,28. Few studies have analyzed other amino acids ^29-31. Variations in the content of Glu, Gln, Ala, Asp and Asn after cold conditions have been reported in wheat ^30,31. Dionne et al. (2001)²⁹ found higher levels of Pro, Gln and Glu and greater differences of amino acid content for three ecotypes of Poa annua L. after an acclimatization at subzero temperatures (-2°C); however the amino acid levels were not related to the differential freezing tolerance of the ecotypes.

Additionally, the mentioned works used only classical statistical analysis to find the differences in the samples' amino acid content. We have previously reported a study classifying these genotypes under cold conditions in order to classify tolerant and sensible genotypes of E. globulus using regularized discriminant analysis (RDA) as a supervised pattern recognition method of carbohydrate content on leaves¹², where the classification did not show satisfactory results.

In this work, 28 genotypes of E. globulus were exposed to cold acclimatization treatments to improve their freezing tolerance, which was measured after frost conditions at the end of the acclimatization periods and expressed as foliar damage percentage. The leaves' amino acid content in the genotypes were quantified and used as data in a statistical and chemometric analysis to classify the genotypes according to cold resistance degree using RDA as supervised pattern recognition methods. Additionally, the differentiation of the plants' acclimated or non-acclimated stages and their provenance by forest companies was studied to determine whether the species are able to increase their tolerance during the applied acclimatization periods and whether each forest company's nursery practices are important factors in the cold tolerance of the genotypes, respectively

In our knowledge, actually there are no studies that use a plant's amino acid contení to select the genotypes according to cold resistance degree using pattern recognition methods.

Theory

Pattern recognition methods

Regularized Discriminant Analysis (RDA)

Developed by Friedman³², this technique has been used in different classifications problems^33-37. Some comparisons with linear discriminant analysis (LDA) and quadratic discriminant analysis (QDA), considering special cases oí RDA method, showed the advantages of its use, especially when ill-conditioned data like NIR spectra are used or when the ratio of objects to variables is not large ³⁷. In RDA, the classification problem can be described as follows:

The matrix X, containing n measurements x_{1, . . . . ,} x_n, each of p variables and the vector y = {1,2, ...K}, where K is the number of groups, is used to assign objects to one of several ATgroups. This assignment obeys a classification rule in such a way that the correct class of the object is predicted with high probability. The most often applied classification rules are based on the normal distribution of the different groups ƒ_k(x) for k {1,2, ...K}^32;

where ∑_k and m_k are the class k covariance matrix and mean vector, respectively, according to:

To minimize the risk of misclassification, x is allocated to the class of máximum probability score:

Where P(k) is the a-priori probability for class k. By assuming normal distribution in Equation (2), the expression in Equation (5) can be replaced by:

Equation (6) and equation (2) lead to the following expression:

where d_k(x) is the discriminant score for the kth class. The first term of equation (7) is the known squared Mahalanobis distance between X and m_k.

Equations (6) and (7) describe the quadratic boundaries between the classes in QDA. When the class covariance matrices are presumed to be identical, the quadratic terms of equation (7) are cancelled and linear discriminant analysis (LDA) is the resulting classification rule ³². In the case of LDA, the class covariance matrices are assumed to be equal, so that a pooled covariance matrix can be used in place of equation (3).

Bayes's theorem for calculation of the posterior probability P(k ‌ x) is used for the assignment of a new sample to class k³⁸:

where P(k ‌ x) is the a-posteriori probability for class k, ƒ_k(x) is the probability density function for class k p(x) is the probability density function over all classes and P(k) is the previously mentioned a-priori probability for class k.

A sample x is then assigned to the class k for which the largest posterior probability is found.

Regularization parameters Lambda (l) and Gamma (g), Risk of Misclassification

There are two parameters that have influence on the covariance matrix³². The net covariance matrix is calculated by:

Where S_k the unweighted estímate of the covariance matrix, w_i is the weight assigned to each observation i, and W_k is the sum of w_i.

The RDA method seeks biased estimates of the covariance matrices, S_k, to reduce their variances by introducing two parameters, λ and γ, according to:

The first regularization parameter λ controls the degree of shrinkage of the individual covariance matrix estimates towards the pooled estimates; the second regularization parameter γ is responsible for weighting the Eigen values contrariwise³². Both parameters adopt values between 0 and 1. Net covariance matrix by introduction of l is calculated by:

And both parameters lead to the weighted regularized covariance matrix :

 

where tr characterizes the trace of a matrix and the values of l and g can be optimized in order to obtain the least risk of misclassification. If λ = 1 and γ = 0, RDA is equal to LDA; and if λ = 0 and y = 0, RDA corresponds to QDA³².

Misclassification risk of assigned sample to class k within matrix X can be calculated by:

Where: is the loss matrix for the assigned sample to class compared to correct classification, ƒ_k(X) is the probability density function for class k and P(k) the unconditional prior probability of observing a class k member

Materials and Methods

Samples and cold acclimatization treatments

Samples and cold acclimatization treatments were described in Castillo et. al., (2008)¹². Briefly, 28 genotypes of E. globulus, 6-months oíd, (60 ramets of each genotype) with standard nutritional states from two Chilean Forest Companies (13 from 'FC1' company and 15 from 'FC2' company) were used as samples. Twenty ramets of each genotype were maintained simultaneously in three cold chamber systems with different temperature profiles¹²: a) cold chamber A (control chamber at 20/12°C), b) cold chamber B, with weak acclimatization cycle (8/4°C), c) cold chamber C with strong acclimatization cycle (8/4/-2°C), during 31 days, with a total number of ramets of 1680 (28 genotypes x 20 ramets x 3 cold chambers). After that, the fourth superior pair of leaves was sampled and one pool of these for each genotype was used as sample for the chemical analysis (84 samples = 28 genotypes x 3 cold chambers). After this sampling, the genotypes in the three cold chambers were maintained 12 h at -6 °C for the posterior physiological assessment of the foliar damage percentage through a backlit visual examination to determine the necrotic área on leaves. This foliar damage percentage was used as reference indicator of the plants' cold resistance degree. The classification of the genotypes according to cold resistance degree was based on two classes: class 1 (Cl) consisted of tolerant genotypes (foliar damage < 50%) and class 2 (C2) consisted of sensible genotypes (percentage of damage ≥ 50%). Two genotypes from the cold chamber B were removed form this classification due to experimental errors, reducing the total number of samples to 82.

Extraction and Analysis ofAminoAcids

The pooled samples of each genotype were dried in an oven at 45°C for 24 h. Then, the leaves were pulverized using a grinder (Moulinex A505, France) for the extraction procedure. Free amino acids were extracted using 0.25 g of the dried and pulverized material with 20 mL of distilled water at 60°C by shaking at 150 rpm for 20 min. The extracts were filtered using 0.45 mm hydrophilic membranes, centrifuged at 12000g for 10 min, and the aqueous phase was collected, evaporated to dryness on a rotary evaporator at 40 °C and resolubilized in 2 mL of nanopure water.

Asp, Glu, Asn, Ser, Thr, Arg, Ala, Tyr, Met, Trp, Val, Phe, Leu e He, were determined by high performance liquid chromatography (HPLC), after a pre-column derivatization procedure with o-phtalaldehyde (OPA), based in the method described by Roshius et al (2006)³⁹ with some modifications. Briefly, 200 mL of the aqueous extract were mixed during 1 min in an amber vial with 1 mL of 0.5M sodium borate buffer pH 9,5 and 400 mL of OPA reagent (0.100 mg OPA +1.2 mL of methanol + 50 mL of 2-mercaptoethanol). Derivatized samples were filtered and injected into a Merck Hitachi chromatographic system equipped with an autosampler AS-4000 and a L-6200 Pump, using a LiChrospher RP-18(5 mm) column, maintained at 40°C in an oven (Eppendorf CH-430). Detection was performed using a fluorescence LaChrom F-1080 (Merck- Hitachi) detector with excitation (λ_exc) and emission (γ_cm) wavelengths of 334 nm and 425 nm, respectively.

Pro was quantified using the method of Bates⁴⁰ described by Dionne et al (2001)²⁹ using UV-VIS spectrophotometry at 521 nm by means of a colorimetric reaction with ninhydrin, using 300 mL of the aqueous extract, 300 mL of glacial acetic acid and 300 mL of acid ninhydrine reagent (1.25 g ninhydrine + 30mL of glacial acetic acid and 24 mL of 18 M phosphoric acid). The reaction was carried out for lh at 100°C with shaking at 150 rpm, and 2  mL of toluene were used to separate the organic phase for its colorimetric determination. An UV-VIS spectrophotometer 1650PC (Shimadzu, Kyoto, Japan) was used for this analysis.

Standard reagents of all the used amino acids (purity>99%) were supplied by Sigma (Darmstadt, Germany). Standard solutions with concentrations of 5, 10, 30, 50, 60, 80 and 100 mM were used to build calibration curves. Concentrations obtained in the extracts were reported in mg of each amino acid per g of dried weight of samples (mg g^-1 of dried weight). All samples were quantified in triplicate and the total number of measurements for each amino acid was 252 (3 extractions x 28 genotypes x 3 chambers). Met — Trp, and Leu — He were not completely resolved and they were used and reported as mixture of those amino acids.

Multivariate data analysis

Differences in the mean of amino acid contents in the genotypes were previously tested by two-way analysis of variance (ANOVA) in order to evaluate the possible influences of the forest company, the cold acclimatization treatments, and the possible effect of the interaction forest company - cold acclimatization treatment on the amino acid content. The data were log-transformed in order to produce normality and reduce heteroscedasticity Variance homogeneity and data normal distribution were tested with Cochrane 's test and normal probability plots, respectively. Post-hoc comparisons were applied using Tukey's honestly significant difference (HSD) test or the Spjotvoll-Stoline test (a generalization of Tukey's test applicable at unequal samples sizes).

In order to classify and to predict the classes of the genotypes by cold resistance degree and additionally by forest company and cold acclimatization treatments, chemometrics analysis using pattern recognition methods (PCA as exploratory analysis and RDA as predictive technique) were carried out. All the replicates were included in the classification methods used. The values of foliar damage of two genotypes were removed from the data due to experimental errors and the classification according to cold resistance degree was carried out using 246 samples.

The search for a good pair of regularization parameters l and g in the RDA classification involved an iterative process altering both parameters with constant step-widths (0.1) between 0 and 1. A surface plot with l, g and % of risk calculated by Equation (14) was used for fast determination of the parameters corresponding to the least risk.

Three techniques were used to valídate the predictive RDA models: resubstitution using all the training set as prediction set, cross validation with the 'leave-one-out' criterion and external validation using three sets in the next (training : prediction) rates: Set 1 (75 : 25 %), Set 2 (60 : 40 %) and Set 3 (50 : 50 %). The results obtained in the modeling sets were evaluated using their risk of misclassification and the results obtained on the prediction set in the external validation were evaluated using the percentage of correctly assigned samples in each class.

The correlation vector from a partial least squares (PLS) model was used to evaluate the correlation of each amino acid to the samples' cold resistance. PLS analysis was performed using the amino acid content as X independent matrix and the foliar damage percentage as dependent variable (Y). X matrix was previously pre-processed by austoscaling and orthogonal signal correction (OSC) in order to exelude the variation of data not correlated with the foliar damage.

Software

ANOVA and HSD test were performed using the General Linear Model (GLM) of Statgraphics Plus (Statistical Graphics Corp. Inc) program. PCA and PLS were performed using Pirouette 4.0 (Infometrix Inc) and RDA was carried out using MATLAB version 7.4.0.287 R2007a (The Math Works, Inc.) software with the routine developed by J. Friedman (1988) and reprogrammed in Matlab.

RESULTS AND DISCUSSION

Statistical analysis (ANOVA)

Table 1 summarizes the results obtained in the ANOVA applied over the amino acid content data, where most of them, excepting Arg and Ala, show differences due to the effect of the interaction forest company - cold acclimatization treatment. In the case of Ala, the genotypes show significant variation only for the cold acclimatization treatment and the effect of the forest company provenance of the samples on the amino acid content could be discarded. Arg shows significant variation in the genotypes by both, forest company and cold acclimatization treatment independently Those results show that a variation of samples by other factor besides cold resistance degree of the genotypes exists.

Figure 1 shows the variation of some amino acids in each genotype under the control and acclimatization conditions. Increases in levels of Pro, Asp, Asn, Gln, Thr, Arg, Ala, Tyr and Phe were observed. In most of the cases, amino acid content increases as the cold acclimatization conditions are stronger, except in the case of Glu and Met-Trp (and in some genotypes inthe case of Asp), where levels of these amino acid in cold chamber C genotypes are decreased.

These results are in agreement with some previously reported works, where Asp, Glu, Gln, Arg and Ala levels incremented under low non-freezing temperatures^30,31, and a diminution of some of them such as Glu, when freezing temperatures were applied over the plants was observed ²³; however, not all these amino acids could be responsible for the increased cold resistance of the genotypes.

PLS

PLS parameters are summarized in the Table 2. The calibration and validation coefficients (rCal and rVal, respectively) show lineal correlation between the amino acids content of the samples and their foliar damage percentage by use of four principal components (PC 's). A total number of forty five samples, majority from cold chamber B, were used in the PLS model with the objective to minimize the clustering samples by the effect of the acclimatization treatments. Model was improved using OSC transformation due to the main factors even are related to the cold acclimatization treatment or forest company of the samples. Results obtained in the PLS analysis to evaluate the correlation degree of each amino acid respect to the cold resistance of the samples is shown in the Figure 2, where the correlation vector of the PLS model show that Arg, followed by Tyr and Ala, are the amino acids most correlated with the foliar damage percentage in the studied genotypes (increases of these amino acids improved greater cold resistance in the samples). Inverse correlation between cold resistance and amino acid content was observed in Glu and Gln.

As we have previously mentioned, in most of studies about amino acid accumulation in plants under cold conditions, principally an accumulation of free proline (Pro) is correlated with cold acclimatization or cold freezing tolerance^24-26; however, in this work, free Pro levels do not show major correlation with the foliar damage of the samples. This result is in agreement with the previously mentioned works suggesting that an increase in free Pro could be merely a result of stress exposure than a cause of the enhanced freezing tolerance^22,27,28.

Pattern recognition methods: RDA

The determination of the regularization parameters in the resubstitution technique shows for all the studied classification objectives the lowest values of risk of misclassification in the range between 1=0.0 to 1.0 and g = 0.0. corresponding to the intermediates between LDA and QDA methods (Figure 3a-c), with a slight superiority of QDA over LDA in the case of the classification by cold acclimatization treatments (Figure 3b). These regularization parameters were also tested in the validation step accompanied by the evaluation of the models' predictive ability. The results obtained in the validation of these models are summarized in the Table 3. In this table, the classification of genotypes by cold resistance degree shows that QDA is the most adequate method to correctly classify most of the sensible and tolerant genotypes in both cross validation and external validation with a high percentage of correctly assigned samples in the prediction sets, especially in the Set 1 (100% of sensible and tolerant classes). As is observed in the Table 3, an increment of samples in the modeling set produces an increment in the percentage of correctly assigned samples; however an increment of the ratio of modeling : prediction samples could create a singularity problem in the apphcation of QDA (singularity refers to the impossibility of inversion of the covariance matrix in equation 7 and is related principally to the number of samples of each class, the total number of samples and the number of variables involved in the analysis ³²). Thus, the rate 75:25 established in set 1 could be considered the best model for future predictions.

Unfortunately, RDA classification cannot provide information about the importance of each variable for the model, but the high percentages of correctly assigned samples for the classification of genotypes by cold resistance degree in the models' external validation (100%) indícate that it outperforms the results previously reported with the use of RDA and carbohydrate content for the same data set¹². This result is very important since carbohydrates are the variables usually mentioned to be correlated to the cold tolerance process, and only a few studies about the variation of amino acids have been reported in a limited number of species under cold conditions.

The results obtained in the additional classification of genotypes by cold acclimatization treatments described in the Table 4 demonstrate the superiority of QDA in the correct classification of the genotypes as well as in the cross validation and external validation. The predictive ability of these models shows a high percentage of correctly assigned samples in all the prediction sets, and especially in the sets 1 and 2, where these percentages correspond to 100%. These results suggest that the studied amino acid content can be useful for classifying genotypes not only between the genotypes acclimated in the conditions of chamber C with respect the others, as is observed in PCA, where genotypes of cold chamber A and B belong to the same cluster (plot is not shown), but also between the three cold acclimatization states in the genotypes.

The classification of genotypes byforest company (Table 5) showed less risk of misclassification when the analysis was carried out separate ly for the genotypes from each cold acclimatization system. In that case, QDA was the best technique with a risk of misclassification of 0% in resubstitution and external validation, a low risk in cross validation, and a highly predictive ability of the models, expressed by the 100% of correctly assigned samples in the prediction sets of external validation. These results show the important influence of the samples' provenances overthe amino acid content in genotypes in both non-acclimated (cold chamber A) and acclimated states (cold chambers B and C). Even when the nursery practices applied by the forest companies prior to the experiments are not reported, these practices result in important differences with respect to the amino acid content between the genotypes, and thus can produce effects when a classification by cold resistance is desired. In genotypes coming from cold chamber C, LDA also gave high percentage of correctly assigned samples.

Best results obtained using RDA reduced to QDA shows that the majority of the studied classes have quadratic boundaries and that the covariance matrices of the classes are different, except in the case of the classification of the genotypes of cold chamber C according to Forest Company, where LDA is also adequate.

It is important to mention the advantages to use RDA as predictive pattern recognition method for the evaluation of cold resistance in the genotypes, where methods based in principal component, such as SIMCA or PLS-DA (discriminantPLS) had no success due to the separation of genotypes by others factors (cold acclimatization treatment or forestry company).

CONCLUSIONS

The amino acid content evaluated in the genotypes after cold treatments was successfully used to classify the genotypes by cold resistance degree, with 100% of the samples correctly assigned in the external validation of the RDA predictive model. These results show the importance of the evaluation of the amino acids levels in the response of E. globulus to cold conditions and outperform the results obtained when the carbohydrates of these same genotypes were used. The multivariate calibration of the foliar damage percentage show the highest correlation of this parameter with the amino acids Arg, Tyr and Ala and consequently these amino acids could be considered responsible for the classification of the genotypes by cold resistance degree. As we have previously mentioned, to our knowledge there are no other studies using classification methods over amino acids content to discriminate tolerant and sensible genotypes in any species, and the results of this study could be used with high accuracy in a selection program.

 

REFERENCES

1.     Freer, J.; Ruiz, J.; Peredo, M. A.; Rodríguez, J.; Baeza, J., J. Chil. Chem. Soc. 48, 19, (2003),        [ Links ]

2.     Hauksson, J. B.; Bergqvist, G.; Bergsten, U.; Sjostrom, M.; Edlund, U.. Wood Sci. Technol, 35, 475, (2001)        [ Links ]

3.     Kelley, S. S.; Rials, T. G.; Snell, R.; Groom, L. H.; Sluiter, A., Wood Sci. Technol, 38, 257, (2004)        [ Links ]

4.     Monrroy, M.; Mendonca, R.; Baeza, J.; Ruiz, J.; Ferraz, A.; Freer, J., J. Near Infrared Spectrosc. 16, 121, (2008)        [ Links ]

5.     Olsson, R. J. O.; Tomani, R; Karlsson, M.; Joseffson, T; Sjoberg, K.; Bjorklund, C, Tappi J. 78, 158, (1995).        [ Links ]

6.     Rials, T. G.; Kelley, S. S.; So, C. L., Wood Fiber Sci. 34, 398, (2002)        [ Links ]

7.     Alsberg, B. K.; Goodacre, R.; Rowland, J. J.; Kell, D. B., Anal. Chim. Acta, 348,389,(1997)        [ Links ]

8.     Castillo, R.; Contreras, D.; Freer, J.; Ruiz, J.; Valenzuela, S., J. Chil. Chem. Soc, 54, 1565, (2008)        [ Links ]

9.     Gierlinger, N.; Schwanninger, M.; Wimmer, R., J. Near Infrared Spectrosc. 12, 113,(2004)        [ Links ]

10.   Michell, A. J.; Schimleck, L. R.,Appita J. 51, 127, (1998)        [ Links ]

11.   Schimleck, L. R.; Michell, A. J.; Vinden, P., Appita J. 49, 319, (1996).        [ Links ]

12.   Castillo, R.; Otto, M.; Freer, J.; Valenzuela, S., J. Chemom, 22, 268, (2008)        [ Links ]

13.   Florence, R., Stan dynamic processes in: Ecology and silviculture of eucayptus forest. Australia, 1996.        [ Links ]

14.   Buchanan B; Gruissem W; R., J., Biochemistry and Molecular Biology of Plants. American Society of Plant Physiologist. John Wiley and Sons: Mariland, EE.UU, 2000.        [ Links ]

15.   Hale M., Temperature stress. In The Physiology of Plants under Stress, Orcutt, D.; Nilsen, E. John Wiley: USA, 1987.        [ Links ]

16.   Dionne, J.; Castonguay, Y.; Nadeau, R; Desjardins, Y. Crop Sci, 41, 443, (2001)        [ Links ]

17.   Tabaei-Aghdaei, S. R.; Pearce, R. S.; Harrison, P, J. Exp. Bot, 54, 1565, (2003)        [ Links ]

18.   Imanishi, H. T; Suzuki, T; Masuda, K.; Harada, T, Scien. Hort. 72, 255, (1998)        [ Links ]

19.   Sasaki, H.; Ichimura, K.; Okada, K.; Oda, M., Scien.a Hort. 76, 161, (1998)        [ Links ]

20.   Zuther, E.; Buchel, K.; Hundertmark, M.; Stitt, M.; Hincha, D. K.; Heyer, A. G.,FEBS Letters, 576, 169, (2004).        [ Links ]

21.   Pennycooke, J. C; Jones, M. L.; Stushnoff, C, Plant Physiol. 133, 901, (2003)        [ Links ]

22.   Haré, P. D.; Cress, W. A.; van Staden, J., J. Exp. Bot. 50, 413, (1999)        [ Links ]

23.   Meza -Basso L. Guarda P, R. D. A. M., Phytochem. 25, 1843, (1986)        [ Links ]

24.   Dörffling K., A. M., Bradersen U., Dörffling H, Melz G., Involvement of abscisic acid and proline in cold acclimatation of winter wheat. In Plant cold hardiness, Pergamon Press: New York, 1998.        [ Links ]

25.   Koster, K. L.; Lynch, D. V., Plant Physiol. 98, 108, (1992)        [ Links ]

26.   Paquin, R.; Pelletier, G., Physiol. Vég. 19, 103, (1981)        [ Links ]

27.   Delauney, A. J.; Verma, D. P. S., Plant J. 4, 215, (1993)        [ Links ]

28.   Wanner, L. A.; Junttila, O., Plant Physiol. 120, 391, (1999)        [ Links ]

29.   Dionne, J.; Castonguay, Y; Nadeau, P; Desjardins, Y, Crop Sci. 41, 1862, (2001)        [ Links ]

30.   Kaldy M., F. S., J. Plant Nut. 7, 1103, (1984)        [ Links ]

31.   Naidu, B. P; Paleg, L. G.; Aspinall, D.; Jennings, A. C; Jones, G. P, Phytochem. 30, 407, (1991)        [ Links ]

32.   Friedman, J. H, J. Amer. Stat.Assoc. 84, 165, (1989)        [ Links ]

33.   Baldovin, A.; Wen, W.; Massart, D. L.; Turello, A., Chem. Intel. Lab. Syst. 38, 25. (1997)        [ Links ]

34.   Guo, Y. Q.; Hastie, T; Tibshirani, R., Biostatistics, 8, 86, (2007).        [ Links ]

35.   Knauthe, B.; Steffen, S.; Barth, M.; Niewoehner, L.; Otto, M., J. Chemom. 22, 252, (2008)        [ Links ]

36.   Quinn G., K., Experimental Design and Data Analysis for Biologists. 1 ed.; Cambridge University Press: Cambridge, 2002.        [ Links ]

37.   Wu, W.; Mallet, Y; Walczak, B.; Penninckx, W.; Massart, D. L.; Heuerding, S.; Erni, F, Anal. Chim. Acta, 329, 257, (1996)        [ Links ]

38.   Otto, M., Chemometrics. Statistics and computer application in analytical chemistry. 2nd ed.; Wiley-VCH: Weinheim, 2007.        [ Links ]

39.   Rohsius, C; Matissek, R.; Lieberei, R., Eur. Food Res. Technol. 222, 432, (2006)        [ Links ]

40.   Bates, L., Plant Soil, 39,205,(1973).        [ Links ]

 

(Received: August 6, 2009 - Accepted: November 11, 2209)

* e-mail addresses: rosariocastillo@udec.cl