"Land and Food Systems, Faculty of"@en . "DSpace"@en . "UBCV"@en . "Vodovotz, Yael"@en . "2008-09-10T21:46:24Z"@en . "1992"@en . "Master of Science - MSc"@en . "University of British Columbia"@en . "The classification of the characteristics of two Mexican mango varieties (Tommy Atkins and Haden) ripened under two conditions (in cold storage and on the tree) was attempted using several multivariate analysis techniques. The aroma of the different mango purees was analyzed using both capillary gas chromatography and sensory analysis. A new portable gas chromatograph (SRI model 8610, Torrance, CA) equipped with a purge and trap was used and the area count of relevant peaks calculated. These results and those obtained from a sensory panel comprised the classification factors. In addition, parameters such as the Magness Taylor pressure values, pH and acid/sugar ratio were determined. Two familiar classification techniques namely Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) were used as well as a new program developed in our laboratory: PCA- Similarity (PCA-SIM). Results showed that differences exist between varieties as well as ripening conditions when analyzed both by GC and sensory techniques. Distinct groups were formed when the results of these techniques were independently subjected to LDA and PCA-SIM. PCA was more successful as a variable reduction technique than a classification method. Although both GC and sensory methods were successful in characterizing the differences between the groups, one could not replace the other since no correlation was found between the two methods. The other physico/chemical parameters were useful, but only to a limited extent, and none could account for both differences between varieties and ripening conditions."@en . "https://circle.library.ubc.ca/rest/handle/2429/1831?expand=metadata"@en . "5581754 bytes"@en . "application/pdf"@en . "CLASSIFICATION OF THE CHARACTERISTICS OF TWO MANGO CULTIVARSHARVESTED AT DIFFERENT STAGES OF MATURITY USING GASCHROMATOGRAPHY AND SENSORY DATAByYael VodovotzB. Sc. (Agr.) University of Illinois, 1989A THESIS SUBMITTED IN PARTIAL FULFILMENT OFTHE REQUIREMENTS FOR THE DEGREE OFMASTER OF SCIENCEinTHE FACULTY OF GRADUATE STUDIESDepartment of Food ScienceWe accept this thesis as conformingto the required standardTHE UNIVERSITY OF BRITISH COLUMBIAApril 1992\u00C2\u00A9 Yael Vodovotz, 1992In presenting this thesis in partial fulfilment of the requirements for an advanceddegree at the University of British Columbia, I agree that the Library shall make itfreely available for reference and study. I further agree that permission for extensivecopying of this thesis for scholarly purposes may be granted by the head of mydepartment or by his or her representatives. It is understood that copying orpublication of this thesis for financial gain shall not be allowed without my writtenpermission.(Signature)Department of t' ^Ley) eThe University of British ColumbiaVancouver, CanadaDate NjA^lqqz DE-6 (2/88)ABSTRACTThe classification of the characteristics of twoMexican mango varieties (Tommy Atkins and Haden) ripened undertwo conditions (in cold storage and on the tree) was attemptedusing several multivariate analysis techniques. The aroma ofthe different mango purees was analyzed using both capillarygas chromatography and sensory analysis. A new portable gaschromatograph (SRI model 8610, Torrance, CA) equipped with apurge and trap was used and the area count of relevant peakscalculated. These results and those obtained from a sensorypanel comprised the classification factors.^In addition,parameters such as the Magness Taylor pressure values, pH andacid/sugar ratio were determined. Two familiar classificationtechniques namely Principal Component Analysis (PCA) andLinear Discriminant Analysis (LDA) were used as well as a newprogram developed in our laboratory: PCA- Similarity (PCA-SIM).^Results showed that differences exist betweenvarieties as well as ripening conditions when analyzed both byGC and sensory techniques. Distinct groups were formed whenthe results of these techniques were independently subjectedto LDA and PCA-SIM. PCA was more successful as a variablereduction technique than a classification method. Althoughboth GC and sensory methods were successful in characterizingthe differences between the groups, one could not replace theother since no correlation was found between the two methods.The other physico/chemical parameters were useful, but only toiia limited extent, and none could account for both differencesbetween varieties and ripening conditions.iiiTABLE OF CONTENTSABSTRACT^TABLE OF CONTENTS^ ivLIST OF TABLES viiLIST OF FIGURES^ viiiLIST OF APPENDICES xiiACKNOWLEDGEMENT^ xiiiI.II.INTRODUCTIONLITERATURE REVIEW^14A. Maturity of Mangos 4B. Pretreatment of Mangos^ 6C. Storage of Mangos 8D. Flavour of Mangos^ 9E. Gas Chromatography 121. The Column^ 122. Traps 143.^Detectors^ 154.^Sampling Procedures^ 165.^Temperature Programming 18F. Optimization^ 191. Random Centroid Optimization^ 20G. Sensory Evaluation^ 21III. MATERIALS AND METHODS 22A. Samples^ 221.^For Optimization of GC Parameters^ 232.^For Actual GC Experimentation^ 23iv3. For Sensory Analysis^ 244. For GC/MS^ 24B. Sample Preparation for Analysis^ 251. Samples Ripened in Storage 252. Gas Chromatography^ 273. Samples Ripened on the Tree and ThoseUsed for Optimization of the GCParameters^ 304. GC/MS 305. Titratable Acidity and pH^ 316. Sugar Analysis^ 317. Sensory Analysis 328. Magness Taylor Pressure Determination ^ 35IV. RESULTS AND DISCUSSION^ 37A. Gas Chromatography 371. Optimization: Preliminary^ 372. Optimization: Random CentroidOptimization^ 523. Internal Standard 604. Repeatability of Chromatograms^ 615. Comparison of Peaks and their Selectionfor Multivariate Analysis 616. Effect of Fruit Size^ 64B. GC/MS^ 67C. Multivariate Analysis Techniques^ 731. Principal Component Analysis 732. Linear Discriminant Analysis^ 743. Principal Component Analysis-Similarity^ 75D. Multivariate Analysis of GC Results^ 791. Principal Component Analysis 792. Linear Discriminant Analysis^ 823. Principal Component Analysis-Similarity^ 84E. Sensory Analysis Results^ 891. Taste^ 892. Aroma 89F. Multivariate Analysis of SensoryAnalysis Results^ 1031. Principal Component Analysis^ 1032. Linear Discriminant Analysis 1053.^Principal Component Analysis-Similarity^ 108G. Comparison of Chemical and Sensory Datafor Taste^ 111H. Comparison of Sensory and GC Data^ 113I. PH and Magness Taylor Results 113J. Weight Loss in Mangos^ 119V. CONCLUSION^ 121VI. REFERENCES 126VII. APENDIX^ 134viLIST OF TABLESTable^ Page1. Experimental plan generated by Random CentroidOptimization (RCO) ^ 552. The mean, standard deviation and coefficient ofvariation of 5 peaks obtained from each of 3Tommy Atkins chromatograms run under optimizedconditions to test the repeatability of theGC^ 633. Results of PCA of the seven normalized peak areasfor the 112 gas chromatograms^ 804. Actual group membership (rows) vs. predicted(columns) for LDA of GC results using bothindividual variables and the first 3 principalcomponents from PCA^ 835. Averaged and normalized sensory results for 16mango samples^ 906. Averages and standard deviation of the normalizedresults for the nine sensory attributes^ 1047. Results of PCA of the sensory attributes 1068. Actual group membership (rows) vs. predicted(columns) for LDA of the individual descriptiveterms as well as principal components from PCA^ 1099. Averages for each lot of pH and Magness Taylorvalues^ 11810. Average weights and standard deviations of mangosbefore and after storage (for those ripened instorage) and after picking (for those ripened onthe tree). The % weight loss was calculated formangos ripened in storage^ 120viiLIST OF FIGURESFigure^ Page1. The SRI model 8610 gas chromatograph^ 282. Questionnaire for aroma determination of mangosamples using QDA^ 343. Questionnaire for sweet/sour determination ofmango using QDA 364. Blank run (10 mL distilled water) of a Tenaxtrap after 15 min of baking at 230 C^ 395. Blank run (10 mL distilled water) of a Tenaxtrap after 30 min of baking at 230 C^ 416. Five gram sample of Tommy Atkins heated to 45 Cfor 15 min. (dry purge 19 min.), concentratedon a Tenax trap and eluted onto a capillary DB-624 column^ 427. Tommy Atkins sample (2.5 grams) with the sameconditions as in Figure 6^ 438. Tommy Atkins sample (0.50 grams) with the sameconditions as in Figure 6 449. Tommy Atkins sample (0.10 grams) with the sameconditions as in Figure 6^ 4510. Tommy Atkins sample (0.25 grams) heated to 40 Cfor 15 min., concentrated on a Tenax trap andeluted onto a capillary DB-WAX column^ 4611. Tommy Atkins sample (0.10 grams) purged at 28 Cfor 15 min., concentrated on a Tenax trap anddesorbed onto a capillary DB-WAX column with aHelium flow rate of 1.96 mL/min^ 4812. Tommy Atkins sample with the same conditions asin Figure 11 except volatiles were concentratedon a charcoal trap^ 4913. Tommy Atkins sample (0.50 grams) purged at 39 Cfor 12 min., concentrated on a charcoal trap anddesorbed onto a Megabore DB-624 column with aHelium flow rate of 4.62 mL/min. The temperatureprogram was modified to ramp at 2 deg/min insteadof 4 deg/min^ 50viiiFigure^ Page14. Tommy Atkins sample with the same conditionsas in Figure 13 except sample was desorbed ontoa Tenax trap^ 5115. The resolution between two peaks was calculatedby dividing the value of f by g. The resultingP i value contributed to the ChromatographicResponse Function^ 5416. Mapping results of the first cycle of experimentsgenerated by RCO 5617. Mapping results of all 18 experiments generatedby RCO^ 5818. Tommy Atkins mango chromatograms obtained usingthe RCO program^ 5919. Reproducibility of three chromatograms of TommyAtkins using the optimized conditions^ 6220. Sample chromatogram of Tommy Atkins ripened onthe tree^ 6521. Sample chromatogram of Haden ripened in storage....6622. Simulated chromatograms of lot 1 (size 12)= Aand lot 3 (size 16)= B, of Haden mangos utilizingadjusted areas of the seven peaks used formultivariate analysis^ 6823. Simulated chromatograms of lot 2 (size 14)= Aand lot 4 (mixed sizes)= B, of Tommy Atkinsmangos utilizing adjusted areas of the sevenpeaks used for multivariate analysis^ 6924. Mass spectrometry identification of major peaksof Tommy Atkins sample^ 7025. Mass spectrometry identification of major peaksof Haden sample^ 7126. Mass spectrometer spectra and chemical structureof a-Pinene, Car-3-ene and 0-Phellandrene^ 7227. Flow diagram of Principal Component Analysis-Similarity program^ 77ixFigure^ Page28^Plot of PCA factor scores (principal components),derived from GC results for four representativelots^ 8129. Plot of factor scores (canonical variables)resulting from LDA for the individual peak areasin the GC analysis^ 8530. Principal Component Analysis- Similarity resultsof the seven GC peaks utilizing Tommy Atkins, lot1, individual 4, ripened in storage as a standard..8731^Corrected principal component scores of thestandard (Tommy Atkins , lot 1, individual 4,ripened in storage) compared to the correctedprincipal component scores of Haden, lot 2,individual 3, ripened in storage^ 8832 Averaged and normalized panel ratings forsweetness and sourness using QDA for TommyAtkins and Haden^ 9133. Box plots of sensory data before and afternormalization procedure for the parameterfruity^ 9334. Box plots of sensory data before and afternormalization procedure for the parameterweed^ 9435. Box plots of sensory data before and afternormalization procedure for the parametersweet aroma^ 9536. Box plots of sensory data before and afternormalization procedure for the parameterpineapple/banana^ 9637. Box plots of sensory data before and afternormalization procedure for the parameterhoneydew melon^ 9738. Box plots of sensory data before and afternormalization procedure for the parameteroverall intensity^ 9839. Box plots of sensory data before and afternormalization procedure for the parametersour taste^ 99xFigure^ Page40. Box plots of sensory data before and afternormalization procedure for the parametertea^ 10041. Box plots of sensory data before and afternormalization procedure for the parametersweet taste^ 10142 Explanation of the Box plot structure andsymbols 10243. Sensory PCA results of principal component 1^ 10744. Plot of factor scores (canonical variables)resulting from LDA for the individual sensorydescriptive terms^ 11045. Scattergrams of individual peak areas (PA1-PA7) from GC analysis with individual sensorydescriptive terms: an attempt to find acorrelation between sensory and GC^ 11446. Scattergrams of factor scores from PCA of GCresults with individual sensory descriptiveterms: an attempt to find a correlation betweensensory and GC^ 11547. Magness Taylor and pH results of the samplesused for sensory analysis^ 116xiVII. LIST OF APPENDICESAppendix^ Page1. Principal Component Analysis- Similaritycomputer program^ 1342. Explanation of symbols for each mango sample^ 1413. Sample calculation for the box plot^ 142xiiACKNOWLEDGEMENTI would like to thank Dr. S. Nakai for sponsoring thisproject: his vision and steadfast support were vital in itscompletion. I am greatful to Belinda Cordoba and GuillermoArteaga for everything from scientific to moral support;without them this project would have never got off the ground.In Mexico, I would like to thank Dr. H. Weinstein and his wifefor their gracious hospitality and Ing. Balleza and his familyfor all their kindness and unquestionable trust. To theFernandez brothers for the cooperation of their employees, theuse of their processing plant and continual help, thank you.Last, but far from least, my gratitude to my parents, familyand many friends for being there when I needed them most.I. INTRODUCTIONMangos, although not grown in Canada, are becoming widelyaccepted by the public in the \"exotic\" fruit category. Asidefrom vitamins A and C, mangos provide minor amounts ofnutrients and are, therefore, considered a secondary foodsource providing variety and flavour to the diet (Caygill etal., 1976). This fruit can, in fact, be found in most foodmarkets in the Vancouver area. Although mangos are grown inmany tropical and subtropical regions of the world (i.e.India, Australia, Africa, South and Central America as well asparts of North America), the main suppliers to the Vancouvermarket are located in Mexico and the Philippines. Aside fromdifferent growing regions, there are hundreds of varietiesavailable and certainly more than one grown in each region.The two most popular cultivars supplied from Mexico areHaden (available from May to mid July) and Tommy Atkins(available in July and August) which are grown in the PacificCoast region. The Tommy Atkins is a medium to large sizefruit weighing 450-700 grams. This cultivar is oval in shapewith a broadly rounded tip. The skin is smooth and thickwhich protects it from disease. The external color is orange-yellow with a bright red blush. The yellow flesh issubstantially fibrous which may be objectionable to some(Campell, 1973). The Haden variety is similar in size, but ismuch less fibrous and lacks the attractive colouring of theTommy Atkins.1Because of the long transport time required for fruitfrom Mexican fields to reach Canadian markets, mangos need tobe harvested before reaching the final ripening stages. Thefruit is then ripened during shipment and, ideally, thesemangos that are ripened in storage should be similar incomposition to those ripened on the tree.Medlicott and co-workers in 1988, discussed thedifficulties in obtaining harvested mangos that will fullyripen simultaneously during transportation and the problems ofdifferent degrees of maturity in retail markets. Resultantdegree of maturity is an important parameter since consumersexpect the pleasing aroma and savory taste associated withproperly ripened fruit. There is, however, no accurate methodfor determining harvest maturity. Some physical parametersthat have been studied for this purpose are: size, shape, peeland flesh colour, lenticels, shoulder growth, pit around thepedicel, specific gravity and heat units. Chemicalcomposition parameters such as starch content, sugar/acidratio and phenolic content may be of value in assessing mangomaturity. However, none of the above factors have been foundto be reliable for this purpose (Salunkhe and Desai, 1983).Suryaprakasa Rao in 1972 also found no relationship betweendegree of ripeness and the size of the fruit and concludedthat the principal physical components of the fruit wereindependent of its maturity.Since, in the case of fruits, the aroma quality2constitutes the main portion of the overall sensory quality(Schamp et al., 1982), mango volatiles were analyzed. Gaschromatography (GC) is a common method used to detect volatilecompounds in food. As mentioned above, consumer acceptance isdue, partly, to the proper aroma development which may beindicative of maturity. Gas chromatography, however, has notbeen widely used for this purpose in mangos.The objectives of this research were the following:(1) To optimize the conditions of a new, portable, GC for theheadspace analysis of mango puree.(2) To group mangos according to cultivar and ripeningconditions using results obtained from gas chromatography andsensory analysis.(3) To compare a computer program developed in thislaboratory (PCA-Similarity) to other existing multivariateanalysis techniques in the classification of mango quality.(4) To relate physical and chemical properties of mangos tocultivar quality and ripening conditions.3II. LITERATURE REVIEWA. MATURITY OF MANGOS:The maturity stage of harvested mangos has a tremendousinfluence on the quality of the storage-ripened fruit. Maturebut unripe fruit has high total acidity and ascorbic acidvalues but low contents of sugars and total soluble solids(Askar et al., 1983). Mangos that were harvested at the semi-ripe stage had a longer storage life than mangos harvestedfully ripe but did not ripen evenly nor had the samecomposition than fruit that was harvested when fully mature(Medlicott et al., 1990). Also, fruit harvested too early wassubject to heavy spoilage during storage (Lakshminarayana etal., 1974). Therefore, mangos, regardless of cultivar ororigin, should not be harvested before they reach fullmaturity. Harvesting of semi-ripe mangos usually occurs whenthe skin colour just begins to change and one or two fruitsdrop from the tree (Jain,1961).A great deal of research has been carried out ondetermining and controlling the ripening process of mangos.Mangos are a climacteric fruit and Krishnamurth et al. in 1970found that colour break, odour development and softeningcoincided with the climacteric peak of advanced maturity inPairi mangos. Lakshiminarayana and co-workers in 1970 notedthat the growth climacteric may be due to high enzymaticactivity in the initial stages when the fruit is utilizing4enormous quantities of respiratory substrates.In Haden mangos, the rate at which flesh firmness waslost declined after the first three days as measured by aMagness Taylor Pressure tester with a plunger of 5/16 inchesdiameter. Likewise, a sensory panel found that the mangosbecame acceptable after the third day of storage (Soule etal., 1956). Density was found to be directly correlated withfirmness for various Indian mango cultivars (Verma andBajapai, 1971). Yet none of the aforementioned parametershave been successfully quantified so as to allow the shipperto use them as selection tools for the desired maturity.Bandyopadhyay and Gholap in 1973 examined therelationship of quantity and type of fatty acid and degree ofmaturity of mangos. Their experiments with Alphonso mangosrevealed that the degree of unsaturation of the fatty acidsbecame greater as fruits matured and that these UFA were moremetabolically active during ripening. Also, the ratio ofpalmitic to palmitoleic acid was found to be significant incontributing to the aroma and flavour of the mango. Strongaroma was noted when the ratio was smaller than one and a mildaroma when the ratio was greater than one.Other parameters which were investigated for theirrelationship to maturity include starch content and amylaseactivity. Pandey and co-workers in 1974, found that thegreatest change in carbohydrates during the development ofDashehary mangos was an increase of starch content from one to513%, a result also found by Pompanoe and co-workers (1957).They concluded that the analysis of the flesh for starchcontent would be useful to estimate the degree of maturity ofthe hard green fruit. Amylase activity was also found toincrease with the development of the fruit (Fuchs et.al .,1980) and this increase was parallel to the increase of fruitweight (Fuchs et al., 1980 and Tandon and Kalra, 1983).Little research has been carried out on comparing tree-ripe mangos to those harvested mature but green and thenripened in storage. Askar and co-workers (1983) comparedthese two conditions for the Egyptian mango variety, Zebda.They concluded that tree ripened fruits were better in tasteand colour than those ripened in storage at room temperature.Also, tree ripened fruits had more aroma compounds as comparedto those ripened in storage especially in the concentrationof cis-ocimene, a very important compound in Egyptian mangoaroma.B. PRETREATMENTS OF MANGOS:A common problem encountered when trying to extend thestorage life of tropical fruit is chilling injury. Thisphenomenon occurs when a fruit is stored at temperatures belowits tolerance level (Mukerjee and Srivastava, 1979). Chillinginjury, according to Thomas and Oke (1983) is manifested bypitting and browning of skin tissues, uneven ripening upon theremoval from low temperatures and failure to develop normal6aroma, flavour, skin and flesh colour on ripening anddecreased resistance to fungal diseases. Chilling injury maybe caused by a low production of certain essentials or an overproduction of some toxic products (Salunkhe et al., 1983).Farooqi and co-workers (1985), while investigating chillinginjury in Sensation and Samor Bahisht mango cultivars foundthat the skin showed the symptoms of chilling injury while therest of the fruit remained unaffected. Mature or partiallyripe Haden mangos developed chilling injury whereas ripemangos did not at the same temperature (Mukerjee andSrivastava, 1979).Several pretreatments had been suggested to preventspoilage, promote uniform ripening and avoid chilling injuryin mangos. Precooling, both by dipping the mature greenmangos in cold water for one minute (Mann and Singh, 1975) andcold adaptation by slow cooling from 10.5 C to 5.5 C (Mukerjeeand Srivastava, 1979) resulted in less spoilage of the fruit.However, the most successful treatment has been a hot waterdip (50 C for 7 min) prior to storage to decrease fungal decay(Thomas, 1975), especially in Haden mangos which are verysensitive to anthracnose spoilage. The hot water treatmentremoved part of the natural waxes on the fruit surface andfacilitated the exchange of respiratory gases by the increasedcell wall permeability. However, this treatment resulted inreduced storage life due to earlier ripening and slightlygreater weight loss of the fruit (Salunkhe and Desai, 1983).7C. STORAGE OF MANGOS:The optimum conditions of storage for mangos are highlydependant on the particular cultivar as well as the degree ofmaturity of the fruit. Storage temperature was consideredthe most influential part on the ripening time (Fornaris-Rollan et al., 1990), and various authors have looked at theinfluence of storage temperature on mangos (Tripathi, 1988;Krishnamurthy and Joshi, 1989; Medlicott et al., 1990; Thomas,1975; Kapse et al., 1985; Abou Aziz et al., 1975 and SaucedoVeloz et al., 1977). Lowering storage temperature slowed downthe physiological and biochemical activities of the fruitleading to senescence (Mukerjee and Srivastava, 1979). Higheracidity and a slower increase in sugar content resulted whenPairi mangos were held at 0 C for various number of days.Campbell in 1959 summarized the findings by recommendingstorage temperatures of 18-24 C at 85-90% relative humidity.Below 10 C chilling injury was common and above 27 C qualitywas seriously impaired. Similarly, Salunkhe and co-author in1983, recommended thee storege of mangos above 19 C sincelower temperatures caused a greyish scald-like discolorationof the skin and uneven ripening, a symptom of chilling injury.Holding the mangos at temperatures above 36 C resulted inover-ripe fruit which rotted quickly (Patwardhan, 1927).Bandyopadhyay et al., in 1973 found that ripening of Alphonsomangos at 25-30 C was accompanied by a change in fatty acidcomposition as well as an increase in triglyceride content in8the pulp, while ripening at low temperatures did not result insuch changes. Saha et al., in 1976 demonstrated the effect ofripeness level on both physical appearance and chemicalcomposition of the mangos.D. FLAVOUR OF MANGOS:Despite its large production volume and increasingeconomic importance, mango and its flavour characteristicshave not been widely studied (Shibamoto and Tang, 1990).Researchers who had analyzed flavour from the perspectiveof taste had utilized the sugar/acid ratio in their work. Thepredominant acid found in mangos was citric with lesseramounts of succinic and malic acids (Medlicott and Thompson,1985). The titratable acidity was shown to decrease asripening progressed (Saha et al., 1976; Selvaraj and Pal,1985; Sharaf et al., 1989; Askar et al., 1983; Pandey et al.,1974; Suryaprakasa Rao et al., 1972; and Mukerjee, 1959.)Tripathi in 1988 found that for mangos stored at roomtemperature, the acidity decreased for the first six days butgradually increased thereafter.The three predominant sugars in mango are sucrose,fructose and glucose. Their abundance was again highlydependant on the maturity of the fruit. Furan derivativesproduced by the Maillard reaction were found in abundance inmangos and probably contributed to its sweetness (Sharaf etal., 1989.) Lakshiminarayana and co-workers in 1970 observed9that total sugars increased gradually up to harvest maturity,but a slight decline was observed near ripening. A similartrend was observed by Patwardhal in 1927 and by Mukerjee in1959. Medlicott and Thompson in 1985 agreed with Fuchs andco-workers who in 1980 explained this phenomenon was caused bythe depletion of starch which initially replaces the sugarbeing metabolised.Upon maturation, the decrease in acidity and increase insugars resulted in a higher sugar/acid ratio and this balancedetermined, to a great extent, the taste of the mango(Selvaraj and Pal, 1989; Sharaf et al., 1989) and cultivarswith higher ratios were preferred by most consumers (Lodh etal., 1974). Suryaprakasa Rao and co-workers in 1972, however,could find no relationship between the maturity of mango andthe sugar/acid ratio.Some research has been conducted on identifying thecompounds which contribute to mango aroma, mostly utilizingGC/MS techniques. Unlike most other fruits, each cultivar ofmango differs greatly in its chemical composition as well asphysical appearance. Each cultivar has its own unique flavourand aroma, and even though, monotropene hydrocarbons are themajor group of volatiles in mango aroma, there is no singlechemical known to have a characteristic mango odour (Shibamotoand Tang, 1990). Engel and Tressl in 1983 found that limonenewas the main component in Baladi mango but was present in veryminute quantities in the Alphonso variety. Also, MacLeod and10Pieris in 1984 found that Venezuelan mangos contained mainlycar-3-ene. Abd El-Baki and co-workers in 1981, found thatcis-ocimene was a character impact compound in mango aroma.Difference in the aroma profile are dependent on thematurity of the mango. Ackerman and Traline in 1984 foundthat 2-butanoic acid esters were characteristic of green mangoaroma while 3-butanoic acid esters comprised ripe mango aroma.These facts contribute to the difficulty of identifying andcharacterizing typical mango flavour. Additionally, mostmethods of isolation of the volatiles consisted ofdistillation and organic solvent extraction (Engel and Tressl,1983; Yamaguchi et al., 1983; MacLeod and Pieris, 1984; Hunteret al, 1974; Gholap and Bandyopadhyay, 1977; Idstein andSchreier, 1985; Gholap et al., 1986; Kunishi and Seale, 1961;Abd El-Baki et al.,1981; Pino et al., 1989, Franco et al.,1991 and Sakho et al.,1985). These authors studied manycultivars (Alphonso, Baladi, Parrot, Willard, Jaffna, Keitt,Tommy Atkins, Zebda, Pairy, Corazon, Bizcochuelo, Super Haden,Haden Edward, Palmer and Zill) grown in several countries(Egypt, Brazil Philippines, Venezuela, Sri Lanka, India andMexico). The major components identified (and here againlargely dependent on the cultivar) belonged to the followinggroups: hydrocarbons, esters, alcohols, aldehydes, ketones,lactones and terpenes.Some criticism has surfaced as to the use of thesemethods for volatile compound isolation. Bartley and Schwede11in 1987, warned that both qualitative and quantitative changesmay occur to the volatile components when solvent extractionand distillation were used. In addition, a loss of volatilecomponents may take place due to the stripping away of lowboiling point compounds (Ackerman and Troline, 1984).Headspace concentration was recommended as more meaningfulmethod than distillation or extraction to ascertain volatilesreleased from fruits (Schamp and Dirinck, 1982).E. GAS CHROMATOGRAPHY:Before the analysis of the mango samples could be carriedout, there were various parameters of the GC which needed tobe considered in order to obtain the most information in areasonable time span.1. The Column The column is probably the most important part of a GCsince that is where separation of the volatiles occurs(Shibamoto, 1984). Many different columns are available soproper choice is based on several parameters. The first isthe type of the column, which is dictated by the type andthickness of the stationary phase, column diameter and columnlength. The type of stationary phase depends on its bondedphase which ranges from nonpolar to polar. Nonpolar columnsseparate on the basis of boiling point alone and, therefore,are rarely used for complex mixtures such as food. Polarcolumns separate not only on the basis of boiling point but12also on the basis of chemical functionality; however, they areextremely sensitive to oxidation and have a low tolerance forabuse. A medium polarity column resolves a large number ofcompounds with similar structure and combines the benefits ofboth polar and nonpolar phases (Rood, 1991).The diameter of a column is the main factor which deter-mines capacity: the maximum amount of sample that can beinjected onto a column before significant peak distortionoccurs (J&W, 1990). The greater the diameter the larger thecapacity. However, capacity is often sacrificed in order toobtain maximum resolution as is the case with capillarycolumns (0.35mm ID) as compared with packed columns (16mm ID).A good compromise is the megabore columns which have aslightly larger internal diameter than capillary columns(0.53mm) and have a substantially larger capacity. Thethickness of the stationary phase also affects the capacityand resolution (both these parameters would increase with athicker film). Greater film thickness increases the column'sinertness and retention and enables higher elutiontemperatures which may eliminate the need for expensive subam-bient operation. However, very thick films also result inlarge amounts of column bleed.The length of the column affects the resolution: thelonger the column the greater the resolution. However, todouble the resolution between two solutes, the column lengthmust increase by a factor of 4 which results in very long13analysis times (Schomburg et al, 1984; J & W Scientific, 1990and Pomeranz and Meloan, 1978).2. Traps For the analysis of trace compounds in a headspace, aconcentration step usually precedes the actual analysis.Several methods are used such as cryogenic concentration andtrapping on adsorbent materials. Using cryogenics results inexcellent volatile recovery but the cost of operation issometimes prohibitive especially if liquid nitrogen is used.However, cryo-trapping during purging may be a cheapalternative since liquid CO 2 can be used. Cryo-trappingdiffers from cryo-focusing in that only the trap is cooled andnot the GC oven. The adsorbent traps which are often used arecoated with porous polymers which have little affinity forwater but high affinity for organic compounds (Durr, 1983).Recovery of the volatiles on adsorbents may be achieved byelution with solvents or by thermal desorption. Activatedcharcoal, and Tenax (p-2,6-diphenyleneoxide) are commonly usedtraps. The activated charcoal trap has very strong adsorbingcapacity and can be heated to high temperatures but itsdesorption characteristics are very poor. Tenax is by far themost common adsorbent in use because of its excellent trappingof non-polar compounds, its inability to retain water as wellas efficient desorption with very low bleed levels. However,polar compounds are poorly retained and because of its low14desorption temperature limit, heavier compounds may remain inthe trap (Westendorf, 1981 and Leahy and Reineccius, 1984).3. Detectors There are several types of detectors that may be usedwith the GC. Sensitivity- the minimum amount of a compounddetected, selectivity- the type of compound detected andlinear range- the detector response proportional to compoundconcentration are important parameters that need to beconsidered when choosing an appropriate detector for aparticular application (Rood, 1991).The thermal conductivity detector (TCD) is sensitive toa change in resistance caused by cooling of a heated wire.The quantity of heat removed from the wire depends on thevelocity and mass of a molecule. This detector is sensitiveto all organic compounds and accommodates a variety of carriergasses including helium, hydrogen and nitrogen (Pomeranz andMeloan, 1978). However, because it responds to water andother non-hydrocarbons, it is usually used only when othermore sensitive detectors are unsuccessful (SRI, 1991). Also,its lack of sensitivity to certain compounds makes itinadequate for many applications (Durr, 1883).Another type of detector is the flame ionization detector(FID) which operates on the basis of a flame produced from acombination of hydrogen, air and the carrier gas. The flame(2100 C) burns the sample producing negative and positive ions15which create a current leading to a detectable signal. TheFID responds to any compound with a carbon-hydrogen bond andis 1000 times more sensitive than a TCD leading to its morefrequent use (Pomarenz and Meloan, 1978). This detector isalso very stable since its calibration stays constant as longas the gas flow rates don't change (SRI, 1991).An electron capture detector (ECD) may be used for addedselectivity. In this detector the electrons striking thesample molecules are captured decreasing the original signal.It can, therefore be used for halogenated compounds,conjugated carbonyls, nitriles, nitrates and someorganometallic compounds (Pomeranz and Meloan, 1978).However, the ECD is radioactive and is prone to contaminationby dirty samples and air leaks (SRI, 1991).4. Sampling Procedures There are two main types of sampling techniques availablefor GC analysis. The first is that of injection of the actualsample which is vaporized in the injector. Although thismethod reveals the content of the sample it is not always anaccurate representation of what the nose smells. Difficultiesalso arise when the sample is solid. There are four maininjection techniques: direct, split, splitless and cool oncolumn. In direct injection, the entire sample is completelytransferred onto the column while in split and splitlessinjection only a fraction of the sample reaches the column.16Split injection is used for very concentrated samples sinceonly a small fraction is allowed to enter the column while insplitless most of the sample reaches the column. Thesplitless mode is often prefered to the split injection due tolower discrimination against high boiling, less volatilecompounds. The cool on column technique involves thedeposition of the sample directly on the column withoutvaporization and is, therefore, used for trace level, highboiling or thermally unstable samples (Rood, 1991).If extraction with organic solvents is used, artifactsmay be formed as well as low boiling point compounds may belost (Dirinck et al., 1984). Quantitative and qualitativechanges in the volatile composition occur during distillationand extraction and these methods do not take into account therelative volatility of the flavour compounds (Schamp andDirinck, 1982). For example, a much larger number of peakswere eluted from a commercial room freshener extracted withpentane than from the head space sampling of the same sample(Shibamoto, 1982). Head space sampling, therefore, isconsidered a much better method since it measures thevolatiles released from the food matrix which can be bettercorrelated to sensory analysis. Headspace analysis does notdestroy the sample which may be of some advantage in certainapplications (Dirinck et al., 1984). It is considered anideal method for sample examination (Durr, 1983).The method of sample delivery and acquisition is also17important. If syringes fitted with steel needles are usedthey act as a miniature fractionation system and do not alwaysdeliver the full sample causing poor reproducibility andinaccurate results (Jennings and Takeota, 1984). Two alterna-tives are available: static (equilibrium) headspace analysisand dynamic headspace analysis. In the static method thevolatiles that are in equilibrium with a sample in a closedsystem are measured. This system has the benefit that it canbe easily automated, however, only the more volatile compoundsare detected and large sample volumes are necessary for properanalysis. In the dynamic headspace technique, a continuousflow of gas is passed through the sample to remove thevolatiles which are then concentrated on an adsorbent polymertrap (Dirinck et al, 1984). The purge and trap method is verysensitive and yields reproducible results and is used forsamples comprised of a large number of compounds with closelyrelated structures. A major problem with this technique isthe large quantity of water which is generated and introducedinto the column extinguishing the FID detector. Reducing thepurge time and desorption time may solve this problem.5. Temperature Programming Temperature programming (changing the oven temperaturewith time) is considered the most influential parameter on theselectivity and proper separation of an analysis (Schomburg etal, 1984). If isothermal temperature is used, low boiling18point compounds will elute with higher boiling point speciesresulting in broad bands with little to no separation. Theexact temperature program used would highly depend on thevolatiles present.F. OPTIMIZATIONAfter these parameters are established, an optimizationprocedure should follow to obtain the best results given thelimitation of the GC when using purge and trap for sampling.Optimizing the purge conditions such as purge volume, samplesize and sample temperature increases the sensitivity of thesystem and is, therefore, widely used. Although intuitivelyall three factors should be increased, doing so would lead topoor results. For example, sample size is often increased,but response is not linear with sample size unless the purgevolume is adjusted, especially for compounds with a low purgeefficiency. Also recovery of some volatiles may actuallydecrease due to breakthrough on the trap (caused bysupersaturation of the trap). Increasing the temperature ofa sample will increase recovery of the volatiles due to highervapour pressure and in solids greater mobility in the matrix.Higher temperature will, however, lead to a greater watervapour which may condense on the trap or the detector,therefore, temperatures greater than 60 C should not be used(Westendorf, 1981). Practicality should also be taken intoconsideration since a long purge time would lead to19excessively lengthy runs.1. Random Centroid Optimization In order to optimize the purge conditions (sample size,sample temperature and purge time), at least three levelsshould be used to obtain a proper trend. Changing one factorat a time while keeping the other two constant would result in27 experiments which may be costly and tedious. Several op-timization computer programs are available in which the numberof experiments are reduced and trends can be more easily seen(Nakai, 1990).Random Centroid Optimization, a modification of theMorgan-Deming simplex optimization, offered the followingadvantages: 1. Simple computer operation, 2. high searchefficiency, 3. no need for boundary constraints (limits as towhere the search for the optimum can take place) whichfrequently caused search stalling, 4. less chance on arrivingat a local optimum than a global optimum and finally, 5. thereis a chance of finding new combinations of the factors (Nakai,1990).The Random Centroid Program explores for the optimumconditions in a given search space. This space is defined bythe constraints placed on the factors being optimized. Theprogram calculates a random set of experiments which includethe centroid points. With the results of these experimentsand the mapping process, the search space is narrowed. This20procedure is repeated until the optimum is reached.G. SENSORY EVALUATIONAs a more objective tool than gas chromatography fordepicting the aroma and taste of the mango samples, sensoryanalysis in the form of Quantitative Descriptive Analysis(QDA) was used. QDA utilizes trained individuals to identifyand quantify the sensory properties of a product in astatistically reliable manner (Rutledge and Hudson, 1990 andStone et al., 1974). The first step in this method is thedevelopment of the language or descriptive terms tocharacterize both the aroma and taste of the given product.These terms are derived by the entire panel in a groupdiscussion setting with the panel administrator providingleadership and guidance but not actively participating in theproduct evaluation (Stone et al., 1974 and Larmond, 1985).The product itself (in this case mangos) were used for boththe language development, selection and training of the judges(Stone et al., 1974; Powers, 1988). The samples used forlanguage development should encompass a wide range oftreatments including the extremes to help minimize end-ordereffects (avoidance of extremes) and encourage full use of thescale (Stone and Sidel, 1985). The terms must be well definedand understood by all the judges in the panel and whenpossible designated an appropriate reference material (Stoneand Sidel, 1985).21A questionnaire with specific instructions and a list ofthese terms is subsequently presented to the judgesindividually in isolated booths. A 15 cm line with anchorpoints 1.5 cm from each end is designated for each term- theanchor point describing the full range for that term (Stone etal., 1974). Unlike the Flavour Profile Method, a centreanchor is not used in QDA for two reasons: the first is toallow the judges more freedom in manipulating the scale,avoiding the need to fit their perceptions into designatedcategories. The second is the need for statistical analysisof the data in which avoidance of a middle anchor would meetthe ANOVA assumption of errors being normally distributed(Powers, 1988). In fact, removal of the centre anchor wasfound to reduce variability by 10-15% (Stone and Sidel, 1985).Replications of the samples were recommended to improvethe reliability of the data and provide information about theconsistency of individual judges as well as the entire panel(Stone and Sidel, 1985).Several researchers have used QDA in conjunction with gaschromatography to characterize the flavour of a product. Forexample, Aishima et al. in 1987 analyzed and comparedstrawberry fruit flavour by gas chromatography and verifiedthe results with QDA.III. MATERIALS AND METHODSA. SAMPLES221. For Optimization of GC Parameters Tommy Atkins mangos were purchased from a localwholesaler (Van Whole Produce, Vancouver, B.C.). These mangoswere ready for distribution to local outlets and, therefore,judged ready to eat. The wholesaler had received the mangosfrom Mexico and kept them in cold storage (15 C) until theirsale.2. For Actual GC Experimentation Sample collection of both Tommy Atkins and Haden tookplace in Tecoman, Colima which is a town located on thePacific coast of Mexico. Four lots were collected for eachcultivar: size 12, 14, 16 and mixed (the size reflected thenumber of mangos which were able to fit in a prefabricated 5Kg. box). The first 3 lots (sizes 12, 14 and 16) for bothcultivars and both ripening conditions were obtained fromDesarrollo Fruticola Del Valle S.P.R. de R.L. (an orchardowned by the Fernandez brothers). The miscellaneous lot (lot4) for each variety consisted of a mixture of sizes and wascomprised of the following: the mangos ripened on trees camefrom different orchards (Tommy Atkins from Empacada Colimanand Haden from Empacadora de Mango Chulavista both in Tecoman,Colima) and those ripened in storage were bought in Vancouverfrom Van Whole Produce Ltd. instead of processed in Mexico.Each lot was comprised of 14 samples: 7 ripened on a treeand 7 ripened in storage. Of each of these 7 samples, 5 were23individual mangos and the other 2 consisted of a blend of 10mangos each (total number of mangos used per lot= 50) Allmangos were randomly chosen within their size and method ofripening.Total number of samples:2 varieties x 4 lots x 14 samples/lot = 112 samples foranalysisMangos that were ripened on a tree were harvested at aready-to-eat stage and no further processing took place exceptfor the sample preparation described below.Mangos that were to ripened in storage were harvested atthe fully mature but green stage (determined by localpickers.)3. For Sensory Analysis The following samples from the ones used for G.C.analysis were also utilized for the sensory analysis: TommyAtkins ripened on tree, Tommy Atkins ripened in storage, Hadenripened on the tree and Haden ripened in storage. The secondmix of 10 mangos for all four lots (sizes 12,14,16 and mixed)was tested for a total of 16 different samples.In addition, to train the sensory panel two othercultivars of mango were obtained from Mexico: Irwin andManila. Both were ripened on the tree.4. For Gas Chromatography/Mass-Spectroscopy GC/MS was used to identify the compounds present in four24representative samples (Tommy Atkins ripened on the tree andripened in storage, and Haden ripened on the tree and ripenedin storage). The first mixed sample of lot 2 (size 14) foreach of the four conditions was used for the analysis.B. SAMPLE PREPARATION FOR ANALYSIS1. Samples ripened in storage These mangos underwent the following processing steps:Pre-storageStep 1: washingFruit was soaked in water at ambient temperature withdetergent added. They were then passed under brush rollers andsponges which scrubbed the fruit clean.Step 2: dryingMangos were dried in a current of hot air.Step 3: classification by sizeA specially designed machine separated the fruit according tosize.Step 4: hot water treatmentMangos were soaked in a bath at 47.5 C as required by the USDAto prevent anthracnose disease. The time of submersion of thefruit depended on its size (and therefore, its weight). Allfruits in this experiment were held in the hot water for 75min.Step 5: cold water treatment25Immediately after the mangos were removed from the hot waterbath, they were immersed in cold water (21-25 C) for 20 min tocool the fruit and,thereby, minimize heat damage.Step 6: storageMangos were then classified by hand according to size, weighedindividually on a table-top electronic balance, and placed inspecially designed boxes which allow air to circulate. Theboxes were then stacked on racks and stored at a temperatureof 15 C until fully ripe.Mangos were judged to be fully ripe by the local ownerafter 10 days of storage (Aina, 1990). Judgement was basedmainly on the softness of the fruit as it yielded to handpressure.Post-storage1) Mangos were peeled, then cut and the pit removed.2) A puree was made with a Waring blender and then passedthrough a mesh strainer (2.0 mm openings) to remove thefibrous material. For the mixed samples, the puree of 10mangos was blended thoroughly. The pulp was used since El-Samahy et al. in 1982 pointed out that the majority of theflavouring compounds were present in the pulp portion.3) The puree was then placed in 40 mL vials with anappropriate code and frozen at -20 C. The mango pulp may bestored from 3 to 6 months at -20 C without the need for heat26treatment (Pino et al., 1989 and Sakho et al., 1985).Samples were stored in dry ice and then transported fromMexico to Canada by plane and immediately placed in -20 C atarrival.2. Gas chromatography OptimizationThe Tommy Atkins samples purchased in Vancouver weredefrosted and the appropriate weight of the sample measuredinto a test tube. The sample was then diluted with distilled,deionized water to 10.0 g. The analysis of volatiles was thencarried out.The SRI Gas Chromatograph:Recently SRI in Torrance, California, developed acompact, low cost and portable gas chromatograph equipped witha purge and trap and an FID detector (8610 model) which wassuitable for routine analysis (Figure 1). The purge and trapunit was heated with a temperature controlled hot wand. Thetraps used were supplied by SRI (Torrance, CA) and when cryo-trapping was utilized, the liquid CO 2 was acquired fromMedigas Pacific Ltd., and connected directly to the GCThe hydrogen and helium gases used for the detector and thenitrogen purge gas were of ultra high purity and acquired fromMedigas Pacific Ltd., Vancouver.^Gas filters and purifiers(Supelco chromatography products) were installed in all gaslines between the gas cylinders and the chromatograph to2710.a,,13.1.17.15.14.16.12.t COLUMN IN OVENZ SCREW-IN OVEN HEATER3. TRAP4. SPARGING HEAD & TOGGLE VALVE5. DISPOSABLE TEST TUBE6. HOT WAND7. CARRIER GAS DIGITAL CONTROLLERB. COLUMN HEAD PRESSURE GUAGE5. DIGITAL TEMPERATURE DISPLAYB SELECTOR SWITCHIS INJECTION PORT11. FID DETECTOR\u00E2\u0080\u00A212 TEMPERATURE SETPOINT ADJUSTMENTS13. ZERO 8 ATTENUATOR CONTROLSFOR TWO DETECTORS14. PID DETECTOR & LAMPCURRENT ADJUSTMENT KNOB15. OVEN COOLING FANS16. DETECTOR OUTPUT SIGNAL CABLESREMOTE CONTROL CABLE17. TILT-UP COVER WITHINTERLOCK SWITCHFigure 1. The SRI (Model 8610) gas chromatograph28remove water, oxygen and oil from the gas. An air compressorand filter were built into the GC so no air tank was required.The nitrogen gas flow rate was 10 mL/sec and the helium(carrier gas) flow rate was 3.34 mL/min in the capillarycolumn and 4.62 mL/min for the megabore column.The flow rates were obtained from the retention time ofmethane after injection at 45 C. The formula used tocalculate the flow rate was:Flow rate(mL/min)= nr 2Lt,where r= column radius in cmL= column length in cmt,= retention time of methane in minTo operate the GC and collect and analyze the data, theG.C. was linked to an IBM compatible 386K personal computerloaded with Peaksimple II, a software package provided by SRI.This program contains many features including various optionsfor the integration of the resultant peaks.2) Routine analysisThe appropriate sample vial was defrosted under runningcold water. As the optimization results dictated, 0.11 gsample was weighed out into a test tube and diluted withdistilled, deionized water to 10.0 g. The internal standardwas added (20 microliters of diluted standard) and thecontents mixed thoroughly. The analysis was then carried out29using the optimized conditions and the GC parameters outlinedin the section on optimization.The needles used for the internal standard were cleanedwith purge and trap grade methanol and distilled water betweenapplications and then left to dry in a drying oven.3. Samples ripened on the tree and those used for the optimization of the GC parameters There was no pre-harvest control of these samples and thepost-harvest treatment was the same as for those mangosripened in storage.4. Gas chromatography/ mass spectroscopy The conditions for the GC/MS analysis were the following:A Hewlett Packard Model 5840 GC was equipped with a Tenax trapwithout cryotrapping. A 30 m X 0.53 mml.D. DB-624 megaborecolumn with 3.0 g film thickness was temperature programmed toincrease by 8 deg/min from 30 C to 225 C. The optimized purgeconditions were used with the mass spectrometer except thesample was not diluted since water accumulation became aproblem. Instead, 6 g of the sample were heated to 51 C in awater bath and purged for 8 min.The mass spectrometer was a Hewlett Packard Model 5985BQuadropole with a Teknibent data system operated with thefollowing parameters:ion source temperature: 200 C30ionization voltage: 70 eVinterface temperature: 250 Cscan rate: 40-350 AMU at 0.5 sec.5. Titratable acidity and pH The appropriate sample was defrosted in cold water and10.0 g weighed out. The sample was then diluted with theaddition of 25.0 mL of distilled, deionized water. The pH wasmeasured with a Fisher pH/ion meter (Model 420) as the samplewas being stirred. It was then titrated to a pH of 8.1 with0.10 N NaOH. Titratable acidity was calculated as % citricacid per gram sample.6. Sugar analysis The appropriate sample vial was defrosted with coldwater. A 1.0 g aliquot was weighed out and diluted withdistilled, deionized water in a 10.0 mL volumetric flask. Thesample was centrifuged for 15 min at 10,000 RPM in a Sorvallcentrifuge (Model RC2-B). For the analysis, 1.0 mL of thesupernatant was diluted again with distilled, deionized waterin a 10 mL volumetric flask.A Boehringer and Mannheim (Biochemica) sucrose/D-glucose/D-fructose enzyme kit was subsequently utilized toanalyze the sample. In this ultra-violet method the VarianCarry 210 spectrophotometer was set at a wavelength of 340nm.The analysis is based on determining the glucose content31before and after the hydrolysis of sucrose into glucose andfructose as well as the determination of fructose separately.The results of the separate sugars were then added to obtainthe total sugar content of the sample.7. Sensory analysisa) Odour analysisMango samples were defrosted and subsequently heated to51 C in a hot water bath to simulate the purging temperatureoptimized for the GC analysis.Training of panel members: Development of the DescriptiveTermsTen judges (4 males and 6 females) from the Food ScienceDepartment at the University of British Columbia gathered ina round-table discussion session to determine the appropriateterms that should be used to describe each sample. Each judgewas given the test samples and asked to chose as manydescriptive terms as he/she could to characterize the sample'saroma. The test samples included heated mango puree of Hadenand Tommy Atkins both ripened on the tree and in storage aswell as heated samples of Irwin and Manila mangos. Theresulting terms were then compiled by the panel leader anddiscussed with all the judges until a clear set of terms wasdeveloped and all the judges agreed on their meaning. To aidthe judges in their task, standards representing the32descriptive terms were provided.Training of panel members: Practice SessionsPanelists were introduced to the questionnaire format andasked to judge several samples individually. The results ofthe tests were discussed amongst all the panelists and fine-tuning of the terms took place.Questionnaire Format:Each judge was given a set of specific instructionsdetailing how he/she should place a vertical line across thehorizontal lines at the point which best reflected themagnitude of the perceived intensity of that particularattribute. The horizontal line was 15cm long with anchorpoints 1.5cm from each end and a descriptive term at each ofthese anchors. The descriptive terms were those determinedduring the training sessions. An example of a questionnaireis shown in Figure 2. Results were transformed into usabledata using SigaScan (Jandel Scientific). This programtransforms a point into a numerical value after the endpointsof a line and its intervals are defined.b) Taste AnalysisIn order to verify the validity of the sugar/acid ratiodetermined by instrumental means, panelists were asked totaste the defrosted mango puree and determine its degree ofsweetness and sourness using the same questionnaire format as33Name:^ Date:Que:,tionaire for Descriptive Analysis of MangosPlease evaluate the odour of the coded samples by making avertical line on the horizontal line to indicate your rating ofthe particular attribute.sample code # ^overall intensityvery weak^ very strongtea-likevery weak^ very strongweed-likevery weak^ very strongfruityvery weak^ very strongpineapple/bananavery weak^ very stronghoneydew melonvery weak^very strongSweetvery weak^ very strongFigure 2. Questionnaire for aroma determination of mangosamples using QDA.34for the odour analysis (Figure 3).Prior to the evaluation of the mango samples, MagnitudeEstimation was used to assess judges' perception of sweetnessand sourness. Four concentrations of citric acid (0, 0.01,0.03 and 0.1 percent) and sucrose (0, 0.15, 0.55 and 1.0percent) were used independently. Testing was performed inindividual booths and samples were served at room temperature.Samples were encoded with a three digit code and tested inrandom order. The panellists were asked to rank the 4 samplesin descending order of intensity (ie most sour to least sourand most sweet to least sweet).Data Collection:Both taste and odour were analyzed in the same session.The 32 samples (16 in duplicates) were presented to each judgein a random order. Four samples were evaluated per session.Judges evaluated heated samples in individual booths with redlight to avoid errors due to colour differences among thesamples.8. Magness Taylor pressure determination A Magness Taylor penetrometer (Hoskin Scientific Ltd.,Vancouver, B.C.) fitted with a 5/16\" diameter plunger was usedto test the firmness of the fruit. The plunger was insertedwith even pressure into the tissue in the plump side of eachmango. The reading was made in kilograms.35Name:^ Date:Sweetness and SournessPlease taste these samples and evaluate them for their degree ofsweetness and sourness. Place a vertical line on the horizontalline to indicate the intensity you perceive.sample #not sweet^ very sweetnot sour^ very sourFigure 3. Questionnaire for sweet/sour taste determination ofmango using QDA36IV. RESULTS AND DISCUSSIONA. GAS CHROMATOGRAPHY1. Optimization: preliminary Before any computer optimization could be utilized,preliminary work had to be done to establish some generalparameters for the gas chromatograph. The sampling procedurewas chosen to be dynamic headspace analysis for the reasonsdiscussed earlier. The column, trap, and temperature programneeded to be decided upon for the mango sample using the SRIchromatograph.Temperature programA conservative temperature program was used at first toobtain a general perspective of the chromatograms. Thetemperature program began with holding the oven at 35 C for 10min, then the temperature was increased at 4 deg/min until210 C was reached and then holding the temperature at 210 Cfor 5 min. Modifications of this program were introducedafter other parameters were established.ColumnsThree columns were used (all acquired from J&WScientific, Rancho Cordova, CA). The first was a DB624capillary column which was 30 m long, had a 0.32 mm diameterand a film thickness of 1.8 p.. The stationary phase of thismedium polarity column was comprised of cyanopropyl,phenyl,dimethyl polysiloxane crosslinked and bonded to fused silica.The Megabore counterpart of the DB-624 (30 m long, diameter of370.53mm and film thickness of 511) was the second column used.Finally, a highly polar capillary column DB-WAX (30 m long,0.32 mm diameter and 0.5 g film thickness) with a stationaryphase of polyethylene glycol crosslinked and bonded to fusedsilica was tested. This column is highly desirable for purgeand trap when separating low boiling, polar molecules (Takeokaand Jennings, 1984). A non-polar column was not used sincemango aroma consists of complex mixture of polar and nonpolarcompounds.TrapsCharcoal and Tenax traps were compared without cryo-trapping at first. When liquid CO 2 is used to cool the trap,there is a greater chance for water accumulation when the trapis thermally desorbed.^This extra moisture can easilyextinguish the FID.^Later cryo-trapping proved to be agreater benefit than a problem.A dry purge (an empty test tube attached to the purgingapparatus) of 19 min followed immediately after the purging ofthe sample so that excess water accumulated during purging waseliminated (Westendorf, 1981). The traps were initially bakedfor 15 min at 230 C (Tenax) and 450 C (charcoal) to eliminateany retained volatiles, a standard procedure in most analyses(Westendorf, 1981).^However, this method did not provesufficient for the mango samples. Figure 4 shows ablank run (10.0 mL distilled water) after a 15 min bakingperiod at 230 C. The eluted peaks were probably due to the38_L/1111111-11111 1 111111111111 1111111Figure 4. GC chromatogram of a blank run (10 mL distilledwater) of a Tenax trap after 15 min of baking at230 C.39trap, necessitating longer baking. Figure 5 shows the sameblank run after 30 min of baking at 230 C. Therefore, thebaking time was increased to 30 min.Purge conditions were to be optimized by Random CentroidOptimization but the upper and lower limits of the factors hadto be established. Therefore, a sample size of 5.0 grams wasarbitrarily chosen as the starting point.Figure 6 shows a chromatogram of the 5.0 g of TommyAtkins sample (diluted to 10.0 mL with distilled water) andheated to 45 C for 15 min. The column used was the capillaryDB624 with a Tenax trap. Immediately it was noted that therewas a substantial overloading of the column (huge peak at theright hand side of the chromatogram). This overloading washighly undesirable since the one huge peak may be maskingother peaks. There were various ways to remedy this problemand the first was to reduce the sample size thereby reducingthe quantity of the volatiles. Figures 7-9 show chromatogramswith the same conditions as in Figure 6 but reduced samplesize (2.5g, 0.5g and 0.1 g respectively). Although theoverloading of the column decreased substantially, reductionand elimination of other peaks also occured. To test theeffect of the stationary phase on the peak separation, a DB-WAX column was installed. A 0.25g mango sample was heated to40 C for 15min. The resultant chromatogram shown in Figure 10indicated a decrease in the resolution of the peaks as well asincreasing the baseline. Both these factors can be40E00.\"\"14*JAILIJIIIILL.11111111 I I I I I I dij11111 I I I I II_LLEILLI (III LI LL111 I I I 1Figure 5. GC chromatogram of a blank run (10 mL ditilledwater) of a Tenax trap after 30 min of baking at230 C.411111111111111_111111111111111111111111111111111111111111111111111111111111111111111111 Figure 6. GC chromatogram of volatiles from five grams sampleof Tommy Atkins heated to 45 C for 15 min \u00E2\u0080\u00A2(drypurge 19 min), concentrated on a Tenax trap andeluted onto a capillary DB-624 column.42II1 \u00E2\u0096\u00BAU I'FtE001^1111 1 HI L1111111^H 111 1 111 1 1 1 I^1 1 I I I^1 1 1 1 1^1 1 1 I H11111111(1111^1111111)1 mil H11 Figure 7. GC chromatogram of volatiles froM a Tommy Atkinssample (2.5 g) with the same conditions as inFigure 6.431 1 1111111h 1 11111IIII1iI1IIHIH11111I1h1111111111HIHIHIIIIHIMLI111111111H1 EOE0tDFigure 8. GC chromatogram of volatiles from a Tommy Atkinssample (0.50 g) with the same conditions as inFigure 6.44EOlatiI H 11 WI !Milli! 11,1 1,111 1 1 h1111111111i1111111111111111111 1 111111 IIH I 11^(1 1 1111Figure 9. GC chromatogram of voltiles from a Tommy Atkinssample (0.10 g) with the same conditions as inFigure 6.45$it.41M0A1/444-A,14,\1111111(111 1 11(1111111111111111 1 11111111111 1 111 1 1i1 1.1411 1 1111111,Figure 10. GC chromatogram of volatiles from a Tommy Atkinssample (0.25 g) heated to 40 C for 15 min,concentrated on a Tenax trap and eluted onto acapillary DB-WAX column.46attributed to the instability of the stationary phase(Takeoka and Jennings, 1984). In an attempt to improve theresolution and separation of the peaks the following changeswere made and the results shown in Figure 11: lowering thecarrier gas flow rate (1.96 mL/min), decreasing the purgingtemperature (28 C) and using a smaller sample size (0.1 g).A slight improvement was noted but overloading (even at suchsmall sample size and no heating during purging) was apparent.Figure 12 shows a chromatogram obtained using the sameconditions as in Figure 11 except a charcoal instead of aTenax trap was used to concentrate the volatiles.^Nosubstantial improvement resulted from this change and,therefore, the capillary DB-624 was judged better than thecapillary DB-WAX column.To try and further improve the resolution, a Megaborecolumn (DB-624) was installed. The helium flow rate wasincreased to 4.62 mL/min since higher flow rates are requiredfor larger diameter columns (to keep the head-pressureconstant). Again a 0.5 g sample was used and it was purgedfor 12 min at 39 C. The temperature program was also modifiedslightly to ramp at 2 deg/min instead of 4 deg/min. Theresults are shown in Figure 13 (charcoal trap) and Figure 14(Tenax trap). The Tenax trap resulted in a better resolutionof the first clump of peaks and was regarded as superior tothe charcoal trap. Westendorf in 1981 pointed out that if acompound that is normally trapped on Tenax is carried into the47E00'.O .1114.04.4110-4.1..0.4,m4ve,711111111111HIIIIIIIIIIIII111111111111111111111111111111111 Figure 11. GC chromatogram of volatiles from a Tommy Atkinssample (0.10 g) purged at 28 C for 15 min,concentrated on a Tenax trap and desorbed onto acapillary DB-WAX column with a helium flow rateof 1.96 mL/min.48tqjEO0111111111111 111111 1111111j1111111111111U1 1_11111111111111111111111111 Figure 12. GC chromatogram of volatiles from a Tommy Atkinssample with the same conditions as in Figure 11except volatiles were concentrated on a charcoaltrap.49rI- ...iFigure 13. GC chromatogram of volatiles from a Tommy Atkinssample (0.50 g) was purged at 39 C for 12 min,concentrated on a charcoal trap and desorbed ontoa Megabore DB-624 column with a helium flow rateof 4.62 mL/min. The temperature program wasmodified to ramp at 2 C/min instead 4C/min.\ILE0 ^0ti)Figure 14. GC chromatogram of volatiles from a Tommy Atkinssample with the same conditions as in Figure 13except sample was desorbed onto a Tenax trap.51charcoal matrix, broader peaks and even some loss in recoverymay occur before the breakthrough volume is reached resultingin inferior chromatograms. Therefore, Tenax was prefered tocharcoal as the adsorbent for trapping the mango volatiles.Subsequently, the use of liquid CO 2 to lower thetemperature of the Tenax trap was tested. As expected, thelower boiling point volatiles were better retained thanwithout the cryo-trapping. Since the liquid CO 2 was used inbursts, not enough water accumulated to extinguish the FID.In addition, dry purging was no longer required saving 19 minon the run time. Therefore, for the Random CentroidOptimization, cryo-trapping was used.The capillary column, having a more linear baseline, wasprefered to the Megabore column. The larger capacity Megaborecolumn tended to exhibit a rising baseline due to its muchthicker stationary phase which leads to increased bleeding.In addition, the capillary column is more frequently used byresearchers enabling comparison with existing literature.2. Optimization: Random Centroid Optimization (RCO) With the basic parameters established, the purgeconditions were then optimized using RCO. The preliminaryoptimization narrowed down the limits for the sample size,purge temperature and purge time. The search space for theoptimum, therefore, was confined to sample size ranging from0.05 g to 0.25 g; purge temperature varying from 35 C to 55 C52and purge time ranging from 5 min to 15 min.In RCO, a response factor was used to numerically comparethe results of the prescribed experimental conditions. In gaschromatography the Chromatographic Response Function (CRF) hasbeen used to quantitatively represent a chromatogram (Aishimaet al., 1990). The CRF takes into account both the number ofpeaks and their resolution and was calculated with theequation below:nCRF= n + E Pii=1where n is equal to the number of peaks in the chromatogramand P i was equal to f divided by g as demonstrated in Figure15. The larger the CRF value the greater the number of peaksand the better the resolution between them.The first nine experiments outlined in Table 1 were theresult of the first cycle of the RCO program. The vertexnumbers were provided by the program, while the actual numbersrepresent the order of experimental execution that resultedfrom manual random selectivity. The outcome of theseexperiments as measured by the CRF were used to generate thecentroid experiments (Table 1). All this data was mapped toaid in narrowing the search space for the optimum (Figure 16).The mapping process aids in visualization of the experimentalresponse surface indicating the trend of the data (Nakai, etal., 1984). The mapping process, therefore, helped confinethe search space for the second cycle to a sample size of53Figure 15. The resolution between two peaks was calculated bydividing the value of f by g. The resulting P ivalue contributed to the Chromatographic ResponseFunction (CRF).54Vertex^Actual^Sample Sample Purgenumber number*^size^temp.^timeCRF**1 3 0_20 40 9.44 12.462 8 0.11 50 6.24 36.843 5 0.10 37 11.36 19.25 first4 4 0.22 40 8.59 19.03 cycle5 9 0.08 48 6.99 26.426 7 0.12 52 10.56 23.377 1 0.25 50 5.45 21.048 6 0.08 49 12.33 23.979 2 0.21 40 14.24 19.0010 10 0.09 49 8.52 25.48 centroid11 11 0.09 49 7.25 24.1812 13 0.12 51 5.30 29.1813 15 0.10 50 6.31 26.12 second14 16 0.18 48 8.94 22.43 cycle15 12 0.17 48 6.90 22.3316 14 0.09 51 9.55 26.2517 17 0.10 50 7.05 17.72 centroid18 18 0.10 51 7.30 32.25* random selection of vertex numbers** Chromatographic Response FunctionTable 1. Experimental plan generated by Random CentroidOptimization (RCO). The vertex number representsthe RCO,order while the actual number was obtainedby randomizing the vertex numbers. The latter wasused to carry out the experiments.5540363228242035 4339 55514740B36 -3228 -2420sample temperature (C)0.05^0.10 0.15^0.20sample size (gm)0.25^0.30C5^7^9^11^13^15purgetimetWOFigure 16. Mapping results of the first cycle of experimentsgenerated by RCO. Figure A, B and C outline thetrends for sample temperature, sample size andpurge time, respectively.LLQ403632282420560.08 g to 0.18 g, sample temperature ranging from 47 C to52 C and a purge time varying from 5 min to 10 min. Usingthese limits, the same procedure was followed as in cycle 1and the optimum was reached after 18 experiments. Results ofall the experiments are found in Table 1. Although Vertex 2has the largest CRF value, in actuality the chromatogram wasless acceptable than the one obtained in Vertex 18 due toinferior peak resolution. The larger number of peaks obtainedusing condition in Vertex 2 contributed heavily to theresultant CRF value. This example demonstrates the importanceof subjective evaluation to arrive at the best conditions.Subsequent mapping of all the results indicated the optimumconditions for the purge parameters (Figure 17). Therefore,using the RCO approach, best results were obtained when a 0.11g. mango sample was purged at 51 C for 8 min. Figure 18compares a chromatogram obtained using Vertex 3 conditions ofTable 1 and the optimum conditions. A substantial improvementwas apparent in the resolution of the peaks as well as thelowering of the baseline. The RCO program eliminated theproblem with overloading of the later eluting peaks whileretaining the minor ones. Although the resolution of thepeaks does not seem appropriate for a capillary column,limitations inherent to this GC model play an important role.The oven may not be properly insulated to obtain satisfactorytemperature control leading to trailing peaks and some bandbroadening. However, for the current application it was5716^0.17^0.18^0.190.11^Oat^0.1.1^0.14^0.15^0sample size (gm)4034LLCEAU4)^oo^so^ 32^53sample temperature (C)CFigure 17. Mapping results of all 18 experiments generatedby RCO. Figure A, B and C outline the trends forsample temperature sample size and purge time,respectively. The arrow indicates the predictedoptimum conditions for each factor.582 4 .619.614.6g .64.60Figure 18. Reproductions of Tommy Atkins mango chromatogramsobtained using the RCO program. The dotted linechromatogram resulted from the conditions outlinedin vertex 3, Table 1, while the solid linechromatogram was obtained after optimization ofthe GC parameters with RCO.59judged satisfactory.3. Internal Standard An internal standard was required to quantitate the gaschromatography peak areas. The compound used as the internalstandard should not be part of the mango effluent as well aselute in an area where no other volatiles are present.Previous mango researchers have used tridecane and 1-heptanolwith Alphonso and Baladi varieties (Engel and Tressl, 1983),ethyl acetate with Tommy Atkins and Kitts varieties (MacLeodand Snyder, 1985), alloocimene, methylphenylacetate andundecane-l-ol with the Alphonso variety (Idstein and Schreier,1985) and heptan-l-ol for Corazon, Bizcochuelo and Super Hadenvarieties (Pino et al., 1989). Following these authors'results, several alcohols and hydrocarbons were tried: 2-methy1-1-propanol, 1-pentanol, 3-methyl-1-butanol, n-octadecane and n-dodecane (all acquired from PolyScienceCorp., Niles, Illinois), however, none gave adequate results.The problem may have been due to the method of analysis(headspace vs. distillation and extraction) as well asvarietal differences. A ketone, 4-methyl-2-pentanone, wasfinally chosen since it showed good recovery with the purgeand trap system and it eluted in an area where no other peakswere found. To normalize the data, all the peak areas fromthe eluted volatiles were divided by the peak area of thestandard.60area of peakNormalized peak area=area of standardThis method allowed the comparison of the differentchromatograms for all the samples (Mohler Smith, 1990).4. Repeatability of Chromatograms Three chromatograms were run with the optimum conditionsobtained from the RCO to determine the repeatability of thegas chromatograph. The internal standard was added at aconcentration of 0.1 ppm after the sample was diluted andbefore beginning the purge sequence. The mango sample was thesame as used for the optimization procedure. Five major peakswere compared for their repeatability after they had beennormalized with the internal standard as discussed above.Figure 19 compares the three chromatograms and Table 2 showsthe mean, standard deviation and the coefficient of variationobtained by dividing the standard deviation by the mean.These results indicate that the G.C. showed good repeatabilityfor these samples.5. Comparison of peaks and their selection for multivariate analysis Peaks were compared by their retention times- that is apeak in one chromatogram that had the same retention time(given a narrow range) as a peak in another chromatogram weresaid to be the same compound. The range for the retention61210rep. 1 rep. 2 rep. 31^2^3^4^5peak numberFigure 19. Repeatability of three chromatograms of TommyAtkins using the optimized conditions.62peak 1 rep 1 rep2 rep 3 avg std cv1 1.59 1.31 1.48 1.46 0.14 0.102 0.84 0.67 0.51 0.67 0.17 0.253 0.25 0.24 0.31 0.27 0.04 0.144 0.23 0.17 0.16 0.18 0.04 0.205 0.86 0.59 0.78 0.74 0.14 0.17Table 2. The average, standard deviation and coefficient ofvariation of 5 peaks obtained from each of threeTommy Atkins chromatograms run under optimizedconditions to test the repeatability of the SRIGC.63time was chosen by site and careful comparison of all thechromatograms.A total of twenty eight different major and minorvolatiles were detected in both varieties in differingquantities and frequency.Of these twenty eight, seven were chosen for themultivariate analysis processing. The selection was based onthe degree of area fluctuation exhibited by each volatile-those volatiles whose area changed the most from sample tosample were chosen for the multivariate analysis. The averageareas for the seven peaks ranged from 0.15 to 0.87 withstandard deviations ranging from 0.06 to 0.76. The differencebetween the samples within the lot, especially for the onesripened on the tree, contributed to the large standarddeviation.Figures 20 and 21 show sample chromatograms of TommyAtkins and Haden ripened on the tree and ripened in storage,respectively. Few peaks were seen due to the samplingprocedure being purge and trap as noted by Bartley and Schwedein 1987.6. Effect of fruit size The lot numbers assigned to each sample designated theirsize (ie lot 1= size 12, lot 2= 14, lot 3= 16 and lot 4 was amix). The four lots showed similar chromatograms for eachvariety and ripening condition including lot 4 which came from64ItE00Figure 20. Sample chromatogram of Tommy Atkins mango ripenedon the tree.651141AA,'II.......................................................................Figure 21. Sample chromatogram of Haden mango ripened instorage.66different orchards. As an example, Figure 22 shows lots 1 and3 of the Haden variety ripened in storage. Peak 1 of lot 1volatiles has a slightly larger area than the same peak in lot3 but otherwise the two lots were very alike. Similarly, lots2 and 4 of Tommy Atkins ripened in storage show comparablepeak distribution (Figure 23). Individual number three in lot4 shows no peak 2, but it is an exception (probably anoutlier) and can't really be compared with lot 2 since itdiffered from other samples within its own lot. SuryaprakasaRao et al. in 1972 also noted that the size of Baneshan mangosdid not affect the quality of the ripe fruit nor the rate ofripening. In addition, Soule Jr. and Harding in 1956 foundlittle effect of size of the mango on the chemical parametersstudied. Therefore, the mango size was not considered as asuitable means of classification and the different lots wereused interchangeably in the computer analysis.B. GAS CHROMATOGRAPHY/ MASS SPECTROMETERYFigures 24 and 25 show the identification of the majorpeaks in Tommy Atkins and Haden mangos.^The three majorcompounds identified were a-Pinene, Car-3-ene and 0-Phellandrene (their structure and mass spectrum are shown inFigure 26). All three compounds have a molecular weight of136 and a general molecular formula of C10H16\u00E2\u0080\u00A2Car-3-ene was found to be the most abundant volatile inTommy Atkins from Florida (MacLeod and Snyder, 1985) and from67A6^7^8 11 12 13 14 15 16 17 18 19 20peak number11^12 _k_ 13 _a_ 14^O 15 _a_ M1B1.50.51^2^3^4^5^6^7^8 11 12 13 14 15 16 17 18 19 20peak number11^12^13 _El_ 14^15^MlFigure 22. Simulated chromatograms of lot 1 (size 12)= A andlot 3 (size , 16)= B, of Haden mangos utilizingadjusted areas of the seven peaks used formultivariate analysis.681.50.511 12 13 14 15 16 17 18 19 20peak number1^22.50.5A11^12^13^a_14 -_ 15 _A_ M11^5^6^7^8 11 12 13 14 15 16 17 18 19 20peak number11^12^13^14^15^.. ^M1Figure 23. Simulated chromatograms of lot 2 (size 14)= A, andlot 4 (mixed sizes)= B, of Tommy Atkins mangosutilizing adjusted areas of the seven peaks usedfor multivariate analysis.6925^7 S1^i\1 1 III11i11 ^ti %,\\u00E2\u0080\u00A2^4^6.--.,..-\.^- I---Peak 1=Peak 2=Peak 4=Peak S=Peak 5=Peak 6=Peak 7=unknownunknownunknownstandard (4-methyl-2-pentanone)a-PineneP-PhellandreneCar-3-eneFigure 24. Mass Spectroscopy identification of major peaks ofTommy Atkins sample.70S7E00OPeak 1= unknownPeak 5= a-PinenePeak 6= P-PhellandrenePeak 7= Car-3-eneFigure 25. Mass Spectroscopy identification of major peaks ofHaden sample.71^ ,,^I,O 10^26^30MW 13GIC , .1 1 1 1 1^I 0 50 60 70 00 90 100 110 120 130 140 150-Pine.le^C10141G3110 120 130..\u00E2\u0080\u009E,./..,,.O 10 20 30 40 50 60 70 80 90 100M14 13G^ 3-CareweII l l 11 t , .`,. 1 1 1V, . l I 1... -, r^I-, .1.i, , ^140 150C10H1G9 31^.... r ,,,.,.. 0- ,^il ... ill,...,1. 1N1..J1i..j .,\u00E2\u0080\u009E\u00E2\u0080\u009E M\u00E2\u0080\u009E..e, 1-,...,.. I, O 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150MN 136^ a-PhellandreHe^ C101-116Figure 26. Mass Spectrometer spectrums and chemicalstructures of a-Pinene, Car-3-ene and 0-Phellandrene.72Brazil (Franco et al., 1991), in Cuban mangos (Pino et al.,1989), in Venezuelan mangos (MacLeod and Pieris, 1984) as wellas in Haden from Brazil (Franco et al., 1991). This compoundwas also found in lesser quantities in African mangos (Sakhoet al., 1985) and in the Indian variety of Alphonso (Idsteinand Schreier, 1985). Car-3-ene was found to have a \"mangoleaves\" aroma and, therefore, considered highly desirable inmangos (MacLeod and Snyder, 1985). a-Pinene and 13-Phellandrene were also found in most of these varieties, butin varying quantities. The first couple of peaks could notbe identified probably due to the lack of cryo-trapping in theGC/MS system (i.e. the Tenax trap by itself failed to retainthese highly volatile compounds).C. MULTIVARIATE ANALYSIS TECHNIQUESIn an attempt to classify the mango samples into twocultivars and two ripening conditions, three multivariateanalysis techniques were used. These will be brieflydiscussed below. All statistics and graphing were performedby SYSTAT (Wilkinson, 1989) and Lotusl-2-3 version 3.1.1. Principal Component Analysis (PCA)PCA is one of the most commonly used multivariateanalysis techniques in food science. PCA aims at reducing thenumber of variables required for data manipulation, and insome cases it was also used for classification purposes. PCA73proceeds by searching for linear combinations of variableswhich account for the maximum possible proportion of variancein the original data (Piggott, 1986).The first step of the analysis is the orthogonal rotationof the N axes of the original pattern space. The N axes canthen be expressed in terms of the original axes. Each newaxis is expressed in the following linear function:PC' = auX, + ^ aniX,where:PC = principal component (N)a = constants of transformation (loadings)X = vectors defining the location of the variablepoints in object spacei = (1,N)n = number of variablesThe first principal component needs to explain the maximumvariance, the second principal component which is orthogonalto the first must explain the largest amount of the residualvariance etc. (Derde and Massart, 1985 and Powers and Ware,1986). The PCA loadings indicate the importance of theoriginal parameters in the direction of the principalcomponent (Derde and Massart, 1985).2. Linear Discriminant Analysis (LDA)LDA is a supervised technique in which mutually exclusivegroups are obtained from a predetermined classification. Asstated by Powers in 1986, the basic concept of LDA is to find,74through a transformation of the original data, a linearcombination of variables so that the mean distance betweenclasses can be maximized. In some cases, it is advantageousto use Principal Component Analysis first to reduce the numberof variables and subsequently apply LDA (Mohler Smith andNakai, 1990). The basic equation for the discriminantfunction was the following:L= blx, + b2x2 + . . . +bpxpwhere:L= linear composite scoreb= the weight of the variableThe goal, therefore, is to obtain the set of b's that willmaximize the between group variability while minimizing thewithin group differences. The b's are selected in such a waythat bl reflects the largest group difference, b2 accounts forthe largest group difference not accounted for in b l etc.Ideally, the first few functions are adequate to expressalmost all the vital group differences (Manly, 1986 and Powersand Ware, 1986).Once the discriminant function has been established, aneed arises to determine its success in classification of thegroups. The computer program utilized in this work, SYSTAT(Wilkinson, 1989), used the Mahalanobis distance from theobservation to the group centre in order to classify thesamples into their respective groups.3. Principal Component Analysis- Similarity 75(PCA-SIM)A new classification method was developed in ourlaboratory in which a sample is compared to a predefinedstandard. Principal Component Analysis-Similarity is anonsupervised pattern recognition technique that uses multi-criteria to discriminate between desired and undesiredcharacteristics.IBM BASIC was used to write a PCA-SIM (Apendix 1) whichconsisted of five parts (Figure 27). Principal component(factor) scores were obtained by any appropriate statisticalprogram (ie Systat, SAS) and used in the PCA-SIM program.This step was critical since PCA reduces the number ofvariables required to be analyzed. The number of factorscores used in the analysis may be determined in several ways.The first, and most common, is by choosing those factorswhose eigenvalues are equal to or greater than one. Thiselimination criteria is based on the idea that a componenthaving an eigenvalue of one is explaining the averageproportion in the data so a value greater than one explains adisproportionately large proportion of variance.A second method is that of cross-validation and itincludes performing the analysis several times with differentsubsets of objects omitted and estimating these omitted scoresfrom the model. Once the number of scores has been chosen, anew file is created which contains just the relevant factorsand this becomes the working file which can be76PCA FACTOR SCORES iSELECT REFERENCELINEAR REGRESSION ANALYSIS1SLOPE VS COEF. OF DETER. PLOTPLOT OF SAMPLE VS REFERENCEFigure 27. Flow diagram of Principal Component Analysis -Similarity (PCA- SIM) program.77used with the new program (Powers and Ware, 1986). For thepresent analysis, the eigenvalue method was chosen as inrelated literature (Cortoneo et al., 1990, Mohler Smith andNakai, 1990).A reference or standard needs to be selected.^Thisreference is one of the samples obtained or some ideal sampleto which all other data would be compared.^An arbitrarymagnification coefficient was subsequently chosen. Thiscoefficient (which is randomly selected) should usually bebetween the magnitudes of 10-30. It aids in magnifying thedifference between the regression lines, elucidating clearlythe differences/similarities between the samples or between asample and the reference.The corrected principal component scores (V i ' areobtained for each sample by using the following equation:V,' = (S i-R 1 ) M+Viwhere:Vi'= corrected PCSS i = principal component scores of the sampleRi - principal component scores of thereferenceM = an arbitrary magnification value to obtainvisually clear resultsVi = an adjusted variance obtained fromPCA for a particular principal component78Linear regression is performed on the corrected PCS ofthe reference against each of the samples from which a slope(S) and a coefficient of determination (R) are obtained.These values are then plotted (i.e. S vs. R) where therelative similarity of each of the samples to the standardcould be visualized. If a particular sample was of interest,its corrected PCS could then be plotted against those of thestandard. A perfect match would be obtained if the pointswere to lie on a 45 degree line, therefore, any variation wascompared to such a line.^The differences between thecorrected PCS of the reference and sample can then beaccounted for by referring to the factor loadings of PCA.D. MULTIVARIATE ANALYSIS OF GAS CHROMATOGRAPHY RESULTS1. Principal Component Analysis PCA was performed on the seven normalized areas of all112 mango samples. The first three principal components hadan eigenvalue greater than one and accounted for 78% of thetotal variance (Table 3a). The factor loading for eachprincipal component (factors) are shown in Table 3b- thehigher the absolute value of the loadings the more thatcomponent (peak area) contributed to the variance between thesamples. A 3-dimensional plot of a representative lot fromeach group (i.e. Tommy Atkins ripened on the tree,\"A\", andripened in storage, \"B\" and Haden ripened on the tree,\"C\", andripened in storage,\"D\") is shown in Figure 28. Seven points79Factor #^Eigenvalues % var. explained1 2.852 40.742 1.515 21.643 1.096 15.65COMPONENT LOADINGSPeak # Factor 1 Factor 2 Factor 37 0.850 0.180 0.0742 0.845 -0.391 -0.2145 0.787 0.013 0.1924 0.698 0.496 -0.3046 0.520 0.044 0.4303 0.105 -0.736 0.5821 -0.169 0.736 0.626Table 3. Results of PCA (a= eigenvalues and % area explainedand b= factor scores of peak areas) of the sevennormalized peak obtained from GC chromatograms forall samples.ab80210-1Figure 28. Plot of PCA factor scores (Principal Components),derived from GC results for four representativelots: A= Tommy Atkins ripened on the tree, B=Tommy Atkins ripened in storage, C= Haden ripenedon the tree and D= Haden ripened in storage.81representing the individual and mixed sample were plotted foreach of the four groups. This plot, as in most 3-dimensionalgraphs, is difficult to visualize. Although the Haden mangosamples (C and D) are reasonably well separated into theirgroups, the Tommy Atkins samples (A and B) could not be welldifferentiated. Similar results were obtained when usingvarious combinations of different lots. Therefore, no clearclassification was noted using the PCA program.2. Linear Discriminant Analysis LDA was performed on the seven normalized peak areas forthe same four lots (28 samples) used in PCA and compared toLDA performed on the three factor scores obtained from firstrunning PCA. Table 4 shows the results of the predictedvalues for both the principal component and individual peakarea methods. Group 1(or A in the graph), group 2 (or B inthe graph), Group 3 (or C in the graph) and Group 4 (or D inthe graph) were assigned respectively to Tommy Atkins ripenedon the tree and in storage and Haden ripened on the tree andin storage respectively. For the individual variables (thepeak areas), three samples were misclassified from group 1 andone sample in group 3 resulting in 14% error ofclassification. Using the principal components, two sampleswere misclassified in group 1, three in group 3 and 1 in group4, yielding an error of classification of 21%. Therefore, forthe predictive equations and graphical representation of the82individual variables 1^2^3^4^TOTAL1^4^1^1^1^72 0 7 0 0 73^0^0^6^1^74 0 0 0 7 7TOTAL 4^8^7^9^28using principal components1^2^3^4^TOTAL1^5^1^1^0^72 0 7 0 0 73^1^0^4^2^74 0 1 0 6 7TOTAL^6^9^5^8^28Table 4. Actual group membership (rows) vs. predicted(columns) for LDA of GC results using bothindividual and the first 3 principal components fromPCA. Group 1= Tommy Atkins ripened on the tree,Group 2= Tommy Atkins ripened in storage, Group 3=Haden ripened on the tree and Group 4= Haden ripenedin storage.83data (Figure 29) the individual component results were used.As in PCA, a 3-dimensional plot of the first three factors (orcanonical variables) was used to try and classify the samples.In LDA, the groups were more distinguishable than in PCA andonly two samples appear to be misclassified (sample A, TommyAtkins lot 2 ripened on the tree individual 1, misclassifiedinto group B and sample C,^Haden lot 1 ripened on the treeindividual 3, misclassified into group D).^Although Table 4shows 4 samples being incorrectly grouped, the other twosamples have very similar probability of being in theirrightful group or in the misclassified one.The predictive equations (below) can be used to assign anunknown sample to a predetermined group given the seven peakareas (ie the sample belongs to the group whose equationyields the greatest positive value).Predictive Equations:GROUP 1= 1.5A1+3.6A2-3.7A3+1.8A4+41.8A5+7.0A6+6.5A7-15.0GROUP 2= 2.4A1-3.4A2-0.4A3+0.2A4+55.3A5+28.4A6-6.5A7-10.4GROUP 3= 3.1A1+5.1A2+1.7A3+1.2A4+2.8A5-9.0A6+10.2A7-8.9GROUP 4= 9.5A1+8.1A2-1.7A3+12.4A4-21.4A5-29.3A6+10.210-9.8where Al-A7 are the peak areas3. Principal Component Analysis- Similarity The first three factors from PCA and the % variance theyeach explain (Table 3b) were subjected to PCA-SIM. Thearbitrary magnification factor (M) was chosen as 10.84-Q\u00E2\u0080\u00A2V-GGV-V-P--V- G.70-e\u00E2\u0080\u00A2--e...)_.. cr.-^ -.c) O^ 4:3Ne\"t.. J'\u00E2\u0096\u00A0^,AMAX THEN AMAX=X(I)1170 IF X(I)BMAX THEN BMAX=Y(I)1190 IF Y(I)1 THEN SCREEN 2:CLS:KEY OFF2030 KEY 20,CHR$(&H8)+CHR$(46):KEY (20) ON2040 ON KEY (20) GOSUB 34602050 IF KLR.P=0 THEN KLR.P=152060 OUT 985,KLR.P2070^XINC.P=XINC:YINC.P=YINC2080 XRANGE.P=XMAX-XMIN:YRANGE.P=YMAX-YMIN2090 IF XINC<=0 THEN XINC.P=10^(INT(LOG(XRANGE.P*.66)/LOG(10)))2100 IF YINC<=0 THEN YINC.P=10^(INT(LOG(YRANGE.P*.66)/LOG(10)))2110^XMIN.P=XINC.P*INT(XMIN/XINC.P):XMAX.P=XINC.P*(INT((XMAX/XINC.P)+1))2120^YMAX.P=YINC.P*INT((YMAX/YINC.P)+1):YMIN.P=YINC.P*(INT(YMIN/YINC.P))2130 IF XLIN=1 THEN XMAX.P=LOG(XMAX)/LOG(10):XMIN.P=LOG(XMIN)/LOG(10)2140 IF YLIN=1 THEN YMAX.P=LOG(YMAX)/LOG(10):YMIN.P=LOG(YMIN)/LOG(10)2150 XRANGE.P=XMAX.P-XMIN.P:YRANGE.P=YMAX.P-YMIN.P2160 DX=SIZE*XRANGE.P/100!:DY=SIZE*YRANGE.P/100!2170^XT.P=XRANGE.P*(9!/XLEN):YT.P=YRANGE.P*(7!/YLEN)2180^TICX=.03*XRANGE.P:TICY=.04*YRANGE.P2190 XTRA=(XT.P-XRANGE.P)*9/XT.P:YTRA=(YT.P -YRANGE.P)*7/YT.P2200^LBD.X=XMIN.P-(1!*XT.P/9)2210^LBD.Y=YMIN.P-(1!*YT.P/7)2220^UBD.X=XMAX.P+((XTRA-1!)*XT.P/9):UBD.Y=YMAX.P+((YTRA-1!)*YT.P/7)symbol type (0=none,1=open sq,2=fill sq,3=open tri4=fill tri,5=open cir,6=fill cir7=open diamond,8=filled diamond,9=XSymbol size in % of axes lengthline type (0=none,1=solid,2=dashed,3=dotted,4=regressiois the number of data pointsarrays that contain the x and y data pointsx and y axis length in inchesx any y minimum valuesx and y axis maximum valuesflag iur linear(=0) or Log(=1) axisunit increment on each axis (valid only for linear)1372230^IF MORE<>1 THEN WINDOW (LBD.X,LBD.Y) - (UBD.X,UBD.Y)2240^IF MORE<>1 THEN LINE (XMIN.P,YMIN.P) - (XMAX.P,YMAX.P),1,B2250 XLOW.P=XMIN.P-LBD.X:YLOW.P=YMIN.P -LBD.Y2260 XHI.P=XT.P-XRANGE.P-XLOW.P:YHI.P=YT.P-YRANGE.P-YLOW.P2270^XP.P=.00159*(UBD.X-LBD_X)2280^IF MORE<>1 THEN LINE (XMIN.P+XP.P,YMIN.P)-(XMAX.P+XP.P,YMAX.P),1,B2290 STYLE=&HEFFF:IF LTYPE=0 THEN STYLE=&H02300 IF LTYPE=2 THEN STYLE=&HFOFO2310 IF LTYPE=3 THEN STYLE=&HCOCO2320 IF LTYPE=4 THEN STYLE=&H02330 IF MORE=1 THEN 31702340 '2350^'label axes2360 '2370 XPOS.P=((XLOW.P+(XRANGE.P/2!))*80!/XT.P)-(LEN(XLABS)/2)2380 LOCATE 25,XPOS.P:PRINT XLAB$;2390 YPOS.P=25!-(25!*((YLOW.P+(YRANGE.P/2!))/YT.P))-(LEN(YLABS)/2!)2400 FOR I=1 TO LEN(YLAB$):YT$=MID$(YLABS,I,1):LOCATE I+YPOS.P,3:PRINT YT$;:NEXTI2 ,41u '2420 ' Print label on figure in specified corner2430 '2440 '2450 IF CORNER=0 THEN GOTO 25902460 MAXLEN=0:IF LEN(LAB3$)>MAXLEN THEN MAXLEN=LEN(LAB3$)2470 IF LEN(LAB2$)>MAXLEN THEN MAXLEN=LEN(LAB2$)+12480 IF LEN(LAB1$)>MAXLEN THEN MAXLEN=LEN(LAB1$)+12490 IF CORNER=1 OR CORNER=2 THEN XPOS.P=((XLOW.P/XT.P)*80!)+32500 IF CORNER=3 OR CORNER=4 THEN XPOS.P=(((XLOW.P+XRANGE.P)/XT.P)*80!)-MAXLEN2510 IF CORNER=2 OR CORNER=4 THEN YPOS.P=((YHI.P/YT.P)*26)+22520 IF CORNER=1 OR CORNER=3 THEN YPOS.P=(((YHI.P+YRANGE.P)/YT.P)*26!)-4!2530 LOCATE YPOS.P,XPOS.P:PRINT LAB1$;CN2540 LOCATE YPOS.P+1,XPOS.P:PRINT LAB2$;2550 LOCATE YPOS.P+2,XPOS.P:PRINT LAB3$;2560 '2570 ' tic marks and numbers on linear x axis2580 '2590 IF XLIN=1 THEN 27102600^FOR XTIC=XMIN.P TO XMAX.P STEP XINC.P2610 LINE (XTIC,YMIN.P)-(XTIC,YMIN.P+TICY),12620 LINE (XTIC+XP.P,YMIN.P)-(XTIC+XP.P,YMIN.P+TICY),12630^LINE (XTIC,YMAX.P-TICY)-(XTIC,YMAX.P),12640 LINE (XTIC+XP.P,YMAX.P-TICY)-(XTIC+XP.P,YMAX.P),12650 XPOS.P=MXLOW.P+(XTIC-XMIN.P))/XT.P)*80!)-(LEN(STR$(XTIC))/2)2660 LOCATE 23,XPOS.P:PRINT USING \"#.##\";XTIC;2670^NEXT XTIC2680 '2690 '^tic marks and numbers on linear y axis2700 '2710 IF YLIN=1 THEN 28502720^FOR YTIC=YMIN.P TO YMAX.P STEP YINC.P2730 LINE (XMIN.P,YTIC)-(XMIN.P+TICX,YTIC),12740 LINE (XMAX.P-TICX,YTIC)-(XMAX.P,YTIC),12750^YPOS.P=((YHI.P+(YMAX.P-YTIC))/YT.P)*26!2760 XPOS.P=6-(LEN(STR$(YTIC))/2)2770 IF YPOS.P>25 OR YPOS.P<1 THEN BEEP:GOTO 28002780^IF XPOS.P>80 OR XPOS.P<1 THEN BEEP:GOTO 28002790 LOCATE YPOS.P,XPOS.P:PRINT USING \"##.#\";YTIC2800^NEXT YTIC2810 '1382820 '^tic marks and numbers on log x axis2830 '2840 '2850 IF XLIN=0 THEN 30102860^FOR CYC=-5 TO 52870 FOR LTIC=1 TO 102880 XTIC=LTIC*(10^CYC)2890^LXTIC=LOG(XTIC)/LOG(10)2900 IF LXTIC<=XMIN.P OR LXTIC>=XMAX.P THEN 29502910 LINE (LXTIC,YMIN.P)-(LXTIC,YMIN.P+TICY),12920^LINE (LXTIC+XP.P,YMIN.P)-(LXTIC+XP.P,YMIN.P+TICY),12930 LINE (LXTIC,YMAX.P-TICY)-(LXTIC,YMAX.P),12940 LINE (LXTIC+XP.P,YMAX.P-TICY)-(LXTIC+XP.P,YMAX.P),12950^NEXT LTIC2960^IF LXTIC>=XMIN.P AND LXTIC<=XMAX.P THEN LOCATE 23,(((XLOW.P+(LXTIC-XMIN.P))/XT.P)*80!)-1:PRINT XTIC;2970^NEXT CYC2980 '2990 '^tic marks and numbers on log y axis3000 '3010 IF YLIN=0 THEN 31403020^FOR CYC=-5 TO 53030 FOR LTIC=1 TO 103040 YTIC=LTIC*(10^CYC)3050^LYTIC=LOG(YTIC)/LOG(10)3060 IF LYTIC<=YMIN.P OR LYTIC>=YMAX.P THEN 30903070 LINE (XMIN.P,LYTIC)-(XMIN.P+TICX,LYTIC),13080^LINE (XMAX.P-TICX,LYTIC)-(XMAX.P,LYTIC),13090 NEXT LTIC3100 YPOS.P=((YHI.P+(YMAX.P-LYTIC))/YT.P)*26!3110 XPOS.P=6-((LEN(STR$(YTIC))/2!))3120 IF LYTIC>=YMIN.P AND LYTIC<=YMAX.P AND YPOS.P>=1 THEN LOCATE YPOS.P,XPOS.P:PRINT YTIC;3130^NEXT CYC3140 '3150 ' now plot data on axes3160 '3170 SX=0:SY=0:SSX=0:SXY=03180 FOR I=1 TO NPTS3190^Xl(I)=X(I):IF XLIN=1 THEN X1(I)=LOG(X(I))/LOG(10)3200^Yl(I)=Y(I) :IF YLIN=1 THEN Y1(I)=LOG(Y(I))/LOG(10)3210^IF I>1 THEN LINE(Xl(I-1),Y1(I-1))-(Xl(I),Y1(I)),1\u00E2\u0080\u009ESTYLE3220^IF I>1 THEN LINE (Xl(I-1)+XP.P,Y1(I-1))-(Xl(I)+XP.P,Y1(I)),1\u00E2\u0080\u009ESTYLE3230^IF SYM=1 THEN LINE (X1(I)-DX,Y1(I)-DY)-(X1(I)+DX,Y1(I)+DY),1,B3240^IF SYM=1 OR SYM=2 THEN LINE(Xl(I)-DX+XP.P,Y1(I)-DY)-(X1(I)+DX+XP.P,Y1(I)+DY) , 1, B3250^IF SYM=2 THEN LINE(X1(I)-DX+XP.P,Y1(I)-DY)-(X1(I)+DX+XP.P,Y1(I)+DY),1,BF3260^IF SYM=3 OR SYM=4 THEN LINE (X1(I)-DX,Y1(I)-DY)-(X1(I)+DX,Y1(I)-DY),1:LINE (X1(I),Y1(I)+DY)-(X1(I)-DX,Y1(I)-DY),1:LINE (Xl(I),Y1(I)+DY)-(X1(I)+DX,Y1(I)-DY),13270^IF SYM=3 OR SYM=4 THEN LINE (X1(I)+XP.P,Y1(I)+DY)-(X1(I)+XP.P+DX,Y1(I)-DY),1:LINE (X1(I)+XP.P,Y1(I)+DY)-(X1(I)+XP.P-DX,Y1(I)-DY),13280 IF SYM=4 THEN PAINT (Xl(I)+2*XP.P,Y1(I)),13290^IF SYM=5 OR SYM=6 THEN CIRCLE (X1(I),Y1(I)),DX:CIRCLE (X1(I)+XP.P,Y1(I))DX3300 IF SYM=6 THEN PAINT (X1(I)+2*XP.P,Y1(I)),13310^IF SYM=9 THEN LINE (X1(I)-DX,Y1(I)-DY)-(X1(I)+DX,Y1(I)+DY),1:LINE (X1(I)+DX, Y1 (I) -DY) - (Xl (I) -DX, Y1 (I) +DY) , 13320^IF SYM=7 OR SYM=8 THEN LINE (Xl(I),Y1(I)+DY)-(Xl(I)+DX,Y1(I)),1:LINE -(X1(I),Y1(I)-DY),1:LINE -(X1(I)-DX,Y1(I)),1:LINE -(Xl(I),Y1(I)+DY),11393330 IF SYM=8 THEN PAINT (X1(I)+2*XP.P,Y1(I)),13340^SY=SY+Y1(I):SX=SX+X1(I):SSX=SSX+(Xl(I)^2):SXY=SXY+(X1(I)*Y1(I))3350 NEXT I3360 IF LTYPE<>4 THEN RETURN3370 '3380 ' Regression line plotted3390 :A=((NPTS*SXY)-(SX*SY))/((NPTS*SSX)-(SX*SX))3400 B=(SY/NPTS)-(A*SX/NPTS)3410 YMTN.P=(A*XMIN.P)+B:YMAX.P=(A*XMAX.P)+B3420 LINE (XMIN.P,YMIN.P)-(XMAX.P,YMAX.P),13430 LINE (XMIN.P+XP.P,YMIN.P)-(XMAX.P+XP.P,YMAX.P),134403450 RETURN3460 '3470 ' key trap of Alt-C to change color3480 '3490 KLR.P=(KLR.P+1) MOD 128:IF KLR.P MOD 8=0 OR KLR.P MOD 16=0 THEN KLR.P=KLR.P+13500 OUT 985,KLR.P3510 RETURN3520 OPEN \"com1:9600,s,7,1,rs,cs65535,ds,CD\" AS #13530 PRINT #1, \"IN;SP1;IP1750,1100,8500,7550;\"3540 PRINT #1, \"SCXMIN,XMAX,YMIN,YMAX;\"3550 PRINT #1 \"PUXMIN,YMIN, PDXMAX,YMIN,XMAX,YMAX,XMIN,YMAX,XMIN,YMIN PU\"3560 PRINT #1,\"SI.2,.3;TL1.5,0\"3570 I=03580 FOR XTIC=XMIN TO XMAX STEP XINC3590 PRINT #1,\"PA\";X(I),\",0\":XT:\"3600 PRINT #1,\"CP-2,-1;LB\";X(I),+CHR$(3)3610 I=I+1:NEXT XTIC140Apendix 2. Explanation of symbols for each mango sample.Tommy Atkins Samples^ Haden Samples tallmil^ hllmiltallmi2 hllml2tallmi3 hllmi3tallmi4 hllmi4tallmi5^ hllmi5tallmml hllmmltallmm2 hllmm2taligil hllgiltallgi2^ hllgi2tallgi3 hllgi3tallgi4 hllgi4tallgi5 hllgi5tallgml^ hllgmltallgm2 hllgm2tal2mil hl2miltal2mi2 hl2mi2tal2mi3^ hl2mi3tal2mi4 hl2mi4tal2mi5 hllmi5tal2mml hl2mmltal2mm2^ hl2mm2tal2gil hl2giltal2gi2 hl2gi2tal2gi3 hl2gi3tal2g14^ hl2gi4tal2gi5 hl2gi5tal2gml hl2gmltal2gm2 hl2gm2tal3mil^ hl3miltal3mi2 hl3mi2tal3mi3 hl3mi3tal3mi4 hl3mi4tal3mi5^ hl3mi5tal3mml hl3mmltal3mm2 hl3mm2tal3gil hl3giltal3gi2^ hl3gi2tal3gi3 hl3gi3tal3gi4 hl3gi4tal3gi5 hl3gi5tal3gml^ hl3gmltal3gm2 hl3gm2tal4mil hl4miltal4mi2 hl4mi2tal4mi3^ hl4mi3tal4mi4 hl4mi4tal4mi5 hl4mi5tal4mml hl4mmltal4mm2^ hl4mm2tal4gil hl4giltal4gi2 hl4gi2tal4gi3 h14913tal4gi4^ hl4gi4tal4gi5 hl4gi5tal4gml hllgmltal4gm2 hl4gm211-14 = lot numbersi= individual samplesm= first m= ripened on the treem= second m= mixed samplesg= ripened in storage141Apendix 3. Sample calculation for the Box plotIntensity: treatment #1Raw data^Raw data reorganized in ascending order3.90 1.95 - whisker6.97 2319.98^3.132.31 3.616.14 3.904.69^4.07 - lower hinge5.21 4.126.13 4.647.05^4.694.07 5.21 Median5.54 5.543.13^6.136.51 6.141.95 6.14738^6.51 - upper hinge6.14 6.978.07 7.054.64^7.384.12 8.073.61 9.98 whiskerEquations used:Median location = # of observations + 1 = 20 + 1 = 10.52^2Median = # at median location (if odd #, use the average of thetwo middle #'s)= 5.21 + 5.54 = 5.382Equations cont.Quartile location = # of observations/ 2 + 1 = 20/2 +1 = 5.52^2Lower hinge = # at quartile location (from top) = 5.5=6 = 4.07Upper hinge = # at quartile location (from bottom) = 6.51Whiskers = high and low range of observations = 1.95 & 9.98"@en . "Thesis/Dissertation"@en . "1992-05"@en . "10.14288/1.0086412"@en . "eng"@en . "Food Science"@en . "Vancouver : University of British Columbia Library"@en . "University of British Columbia"@en . "For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use."@en . "Graduate"@en . "Classification of the characteristics of two mango cultivars harvested at different stages of maturity using gas chromatography and sensory data"@en . "Text"@en . "http://hdl.handle.net/2429/1831"@en .