Food taints and off flavours edited by m j saxby

1 Sensory evaluation of taints and off-flavours D.KILCAST 1.1 Introduction 1.2 Thresholds and their measurement 1.3 Sensory descriptions of taints and off-flavours 1.4 Principles of sens

Food Taints and OtT-Flavours

Food Taints and Off-Flavours

Second edition

Edited by

M J SAXBY

A fonner Manager Analytical Chemistry Section Leatherhead Food Research Association

Surrey

SPRINGER-SCIENCE+BUSINESS MEDIA, B.V

First edition 1993

Second edition 1996

© 1996 Springer Science+Business Media Dordrecht

Originally pubIished by Chapman & Hali in 1996

Softcover reprint of the hardcover 2nd edition 1996

Typeset in 1O/12pt Times by AFS lmage Setters Ltd, Glasgow

ISBN 978-1-4613-5899-2 ISBN 978-1-4615-2151-8 (eBook)

DOI 10.1007/978-1-4615-2151-8

Apart from any fair dealing for the purposes of research or private study, or criticism or review, as permitted under the UK Copyright Designs and

Patents Act, 1988, this publication may not be reproduced, stored, or

transmitted, in any form or by any means, without the prior pennission in writing of the publishers, or in the case of reprographic reproduction only in accordance with the terms of the licences issued by the Copyright Licensing Agency in the UK, or in accordance with the tenns of licences issued by the

appropriate Reproduction Rights Organization outside the UK Enquiries concerning reproduction outside the tenns stated here should be sent to the publishers at the Glasgow address printed on this page

The publisher makes no representation, express or implied, with regard to

the accuracy of the information contained in this book and cannot accept any legal responsibility or liability for any errors or omissions that may be made

A catalogue record for this book is available from the British Library

Library of Congress Catalog Card Number: 95-79979

~ted on acid-free paper, manufactured in accordance with

ANSI/NISO Z39.48-1992 (Permanence ofPaper)

139 Call Hall, Kansas State University, Manhattan,Kansas 66506-1600, USA

UCLA School of Public Health, Los Angeles,

CA 90024-1772, USASensory Analysis and Food Texture Section,Leatherhead Food Research Association, RandallsRoad, Leatherhead, Surrey KT22 7RY, UKSPK Consultancy Services, 48 Chiltern Crescent,Earley, Reading RG6 IAN, UK

TNO Division of Nutrition and Food Research,

PO Box 360, 3700 AJ Zeist, The NetherlandsCentral Laboratory-CIRSEE, Lyonnaise des Eaux etDumez, 38 Rue du President Wilson, 78230 Le Pecq,France

Fonnerly with Safeway Stores pic, Beddow Way,Aylesford, Maidstone, Kent ME20 7AT, UK

84 Pixham Lane, Dorking, Surrey, RH4 IPH, UKNeotronics Scientific Ltd, Western House,

2 Cambridge Road, Stansted Mountfitchet,Essex CM24 8BZ, UK

UCLA School of Public Health, Los Angeles,

CA 90024-1772, USASafeway Stores pic, Beddow Way, Aylesford,Maidstone, Kent ME20 7AT, UK

PIRA International, Randalls Road, Leatherhead,Surrey KT22 7RU, UK

In the past it is regrettable that many instances of serious taints in a food havenot been investigated, so that their cause, source and removal have remainedunresolved Sometimes this may have been caused by the lack of sufficientlysensitive analytical equipment required for the identification of the compound,but it is more often because the expertise needed to interpretall the clues has notbeen available This book provides the background information needed forpersonnel to recognise how food can become tainted, how to draw up guidelines

to prevent this contamination, and how to plan the steps that should be taken in theevent of an outbreak For the reader in academic and research institutions, thebook provides a very extensive literature survey as a basis for future research.The totally revised second edition of the book consists of ten chapters, eachwritten by specialist authors, known for expertise in their own subject It sets out

to present this unique subject in a way which will be understandable to academic,technical and commercial staff The importance of detecting the presence of ataint in a foodstuff by organoleptic methods, both as a preventive measure and inthe early stages of an investigation, is dealt with in the first chapter The secondchapter is an extensive survey of chemical compounds that can cause taints andlooks at their origin, while the following chapter deals with the critical subject oftheir analysis The opportunity has been taken for total revision of this chapter by

a new author The next three chapters deai with taints in water, dairy products andfats Two further chapters are written by experts in the packaging indsutry and inthe retailing sector The penultimate chapter deals with the incidence ofmicrobiologically formed off-flavours The final chapter is totally new and is asurvey of off-flavours in alcoholic beverages, with particular reference to beer

M.l.S.

1 Sensory evaluation of taints and off-flavours

D.KILCAST

1.1 Introduction

1.2 Thresholds and their measurement

1.3 Sensory descriptions of taints and off-flavours

1.4 Principles of sensory evaluation of food

1.4.1 Analytical tests

1.4.2 Hedonic tests

1.4.3 Requirements for a sensory evaluation operation

1.5 Sensory testing for taint

1.5.1 Test selection and modification

1.5.2 Diagnostic taint testing

1.5.3 Preventive (taint-transfer) testing

1.5.4 Sensory quality control taint testing

1.5.5 Storage and shelf-life testing

3 Analysis of taints and off-flavours

H MAARSE and H W GROSCH

25

263031

34

37 38

41

41 44 44

46 484953 55 56 60 61 626365 66 6672

72 73 73 79 82 83 83 83

4 Taste and odor problems in drinking water

I H (MEL) SUFFET, D KHIARI andJ.MALLEVIALLE

4.1 Introduction

4.2 Off-flavors in drinking water and their relation to drinking water standards

4.3 Classifying tastes and odors

4.4 The causes of tastes and odors in water supplies

4.5 Cause-and-effect relationships in drinking water taste and odor problems

4.6 Specific taste and odoHausing compounds

4.7 Treatment of specific odorous chemicals

4.8 Taste and odor treatment of off-odors by chlorination and chloramination

4.9 Taste and odor treatment of off-odor by ozonation

4.10 Taste and odor treatment by carbon adsorption

5.5.6 Moldy or musty flavor

5.5.7 Miscellaneous microbial flavors

139

139 139 141 141 142 142 143 144 144 145 146 148 148 148 149 150 151 151 151 152 155 155 158 160 162 163

7.3.2 Precautions to prevent odour and tainting from printed packaging 237 7.3.3 Methods for detennining levels of residual solvents 239 7.3.4 Some reponed odour and taint investigations 241

7.7 •Scalping' of food aromas and flavours by packaging 254 7.8 Analytical methods for isolating and identifying odorous and tainting substances 256

9.1.1 Off-flavours prefonned in the food

9.1.2 Off-flavours fonned as a result of cellular disruption

9.1.3 Off-flavours arising as a consequence of microbial deterioration

9.2 Milk and dairy products

9.3.5 Cucumber and lettuce

9.4 Wine and beer

308

309 309 310 310 311 311 311 311 312 312 314 315 315 316 316 317

317

321

1 Sensory evaluation of taints and otT-flavours

D KILCAST

1.1 Introduction

Throughout the world, food consumption patterns have formed from the need toensure the nutritional intake necessary for life People in developed countrieshave been fortunate to have sufficient supply and variety available to allowenjoyment of eating to have a dominant influence on food choice Many factorscan influence that enjoyment but the interaction of the food with the humansenses, and the consequent perception of organoleptic quality, are of primaryimportance

When initially making a purchase of food, it is probable that many factors notrelated to the senses are operating: for example exposure through advertising; pos-itioning on the supermarket shelves; packaging and pricing Repeat purchase ofthatproduct, however, depends critically on how much its organoleptic quality isenjoyed Perception of that quality may also have been influenced by our expec-tations, which may be raised, for example, by strong brand image, by attractivepackaging and printing, and by premium pricing.If our expectations are not satis-fied, we will choose other products; if our expectations are violated, we may bedisposed to complain to the retailer or manufacturer Furthermore, the more impor-tant the characteristic that is violated, the stronger will be our complaint Whiledifferent foods can have different key sensory characteristics, flavour is generallythe most important sensory determinant of food choice There are many differenttypes of possible flavour defects but the presence of a flavour that is totally foreign

to that food, or taint, is particularly serious In order to understand how such taintscan arise, it is necessary to describe how the human senses are used, and howorganoleptic methods can be used for both detection and prevention purposes.The primary human senses involved in organoleptic assessment of food aresight, hearing, touch, taste and smell, although physiological responses are alsothought to have some importance The relative importance of the individualsenses will vary with factors such as the type of food, consumption situation andthe psychological state of the subject.Itis also important to note that the differentsensory modalities do not necessarily operate independently, and interactiveeffects are common Dark foods, for example, will be judged to have a strongerflavour than lighter foods, even when only the visual colour intensity differs Oursenses can often be misled to a large degree; an orange-coloured drink with astrawberry flavour will be judged by many people to be orange-flavoured

In practice, the sensory properties of food are generally grouped into three

M J Saxby (ed.), Food Taints and Off-Flavours

© Chapman & Hall 1996

2 FOOD TAINTS AND OFF-FLAVOURS

Flavour:

categories: (i) appearance; (ii) flavour; and (iii) texture The InternationalStandards Organisation has proposed the following working definitions

Appearance: all the visible attributes of a substance or object

There is a common tendency to equate appearance with colour, but this definitionstresses that other characteristics such as opacity, surface gloss and even form andshape can be of great importance Since the first contact we have with a food isusually visual, appearance is a characteristic that can determine subsequentperception and reaction

complex combination of the olfactory, gustatory and trigeminalsensations perceived during tasting

This definition recognises the interaction of: the sense of smell, perceived throughthe nose; the sense of taste, perceived on the tongue; and the pain response thatgives irritant effects and sensations such as 'cool' and 'hot'

Texture: all the mechanical, geometrical and surface attributes of a product

perceptible by means of mechanical, tactile and, whereappropriate, visual and auditory receptors

An important aspect of this definition is the recognition of the importance of othersensory modalities Indeed, one UK snack-food manufacturer has advertised itsmain product as the noisiest crisp!

Both variety and quality in our food are expected, and violation of consumerexpectations of food quality will result in not only a shift to other suppliers andfoods but lead also to complaints if the quality deterioration is substantial Theseconsequences are typical of taint problems, and their unpleasant nature can causesevere difficulties to the retailer, manufacturer, ingredient supplier, farmer,distributor, equipment supplier and even building contractor

These problems can include lost production, lost sales, lost consumerconfidence, damaged brand image, damaged commercial relationships betweensupplier, manufacturer and retailer, and expensive litigation proceedings Foodmanufacturers readily understand the financial implications of a day's defectiveproduction but not the more widespread implication of tainted production; forexample, the tainted production will not be reworkable, and the plant may sufferfrom extended shut-down while defective building materials are replaced Inaddition, identifying the source of a taint can be time-consuming and expensive.Examination of English language dictionaries for definitions of taint produceswords such as blemish, contamination, corruption, infection, pollution and defect.Even outside the specific context of food, the undesirable nature of taint is clear.Within the context of food, definitions become more precise The standarddefinition of taint (ISO, 1992) is a taste or odour foreign to the product Thisdefinition also distinguishes an off-flavour as an atypical flavour usually

SENSORY EVALUATION OF TAINTS AND OFF-FLAVOURS 3associated with deterioration These definitions are characterised by an importantdistinction from the dictionary definitions: food taints are perceived by the humansenses This does not diminish the undesirable nature of chemical contamination

of foods but focuses on those contaminants that can be perceived, particularly bytheir odour or flavour, and which can be perceived at extremely lowconcentrations, for example parts per million (lQ6),ppm, parts per billion(lQ9),ppb, or even parts per trillion (1012),ppt

An alternative means of distinguishing taints and off-flavours is as follows:

Taints: unpleasant odours or flavours imparted to food through

external sources

Off-flavours: unpleasant odours or flavours imparted to food through

internal deteriorative changeThese definitions have the useful practical value of discriminating taint arisingfrom external contamination, from taint arising from internal change Althoughthe two types of taint can render food equally unpleasant, this distinction is ofgreat assistance in identifying the cause of taint problems The difficulties inidentifying taint problems result from a number of sources

Consumer descriptions of taint are, with a few exceptions, notoriouslyunreliable, partly from a lack of any training in analytical descriptive methods butmainly from unfamiliarity with the chemical species responsible for taint Onenotable exception is taint resulting from chlorophenol contamination, which isreliably described as antiseptic, TCP or medicinal, this reliability being aconsequence of consumer familiarity with products characterised by thesesensations Equally importantly, as already indicated, taint can occur at all stages

of the food manufacture and supply chain, and from many different sources ateach stage Consequently, the detective work needed to identify the cause of taint-oriented consumer complaints can be quite different for taints and for off-flavours Examples of the numerous sources of taint are shown in Table 1.1, andtypical descriptions associated with chemical 'species' from these sources areshown in Table 1.2

If minimising the risk of taint is so important, then it is natural to seek ways ofpreventing taint occurrence In the more specific definition of the term 'off-flavour', this can, in principle, be achieved through consideration of ingredients,process, packaging;md storage factors, and is often reflected in product shelf-life.These factors can also be of great importance in taint problems occurring fromexternal sources but these external sources of risk also need to be identified Thecomplexity of the picture that emerges means that the use of chemical analyticalmethods can be compared to looking for a very small needle in a large field full ofhaystacks, a task that can be simplified by the use of appropriate sensory methods.Discussion of the uses of sensory test methods must be preceded, however, by anexamination of human sensitivity to chemical stimuli and of how tainting speciesare perceived

4 FOOD TAINTS AND OFF-FLAVOURS

Table 1.1 Common sources of taints and off-flavours

Wine corks Fumigants Water treatment systems Blue-green algae/actinomycetes Migration from process line components Effluent

Transpon containers Rooring materials Paints

Disinfectants Diesel fumes Insulation materials External chemical release Meat and fish spoilage Vanillin degradation Anaerobic fermentation Sorbic acid degradation Autoxidation reactions Enzymic reactions Precursor reactions

1.2 Thresholds and their measurement

Whether a chemical species can be perceived in a food depends on the chemicalstructure, its concentration, the type of food and the sensitivity of the humansubject Inorder to be perceived as a taint, the chemical need not be positivelyrecognised, but must be characterised as a deterioration in flavour quality Itshould be noted that chemical contamination does not necessarily produce thisdeterioration, and can even produce a perceived improvement in flavour

Characteristic food taints are often detectable at sub-ppm levels, even down toppt levels However, the concentrations at which chemical species can be detectedvaries considerably between individuals Figure 1.1 shows the frequencydistributions of sensitivity to some flavour stimuli (Blakeslee and Salmon, 1935).Not only are the different chemical species perceived at widely differentconcentrations but the range of sensitivities to a specific species also depends onits chemical nature An extreme in this range is exhibited by PTC(phenylthiocarbamide), with a bimodal distribution It has been reported(Teranishi, 1971) that for pyrazines there is a range of 108in odour thresholds.The term threshold, commonly defined as the concentration in a specifiedmedium that is detected by 50% of a specified population, is widely used indescribing sensory perception of stimuli but unfortunately is frequently misusedand misunderstood Thresholds indicate the level of stimulus that is sufficient to

SENSORY EVALUATION OF TAINTS AND OFF-FLAVOURS 5Table 1.2 Common chemical taints

Type Descriptions Examples of sources

Phenols Phenolic Rooring materials

Carbolic Potato deterioration

Vanillin degradation Halophenols Disinfectant Halogenation of phenols

Antiseptic Bleached paperlboard materials TCP Disinfectants

Phenolic Herbicides

Wood treatments Chemical discharges Haloanisoles Musty Microbial action on halophenols

Earthy Wine corks

Bleached paperlboard materials Chlorobenzenes Chemical Drain cleaners

Woody Toilet deodorants Sulphur compounds Catty Solvent/food reaction (mesityl oxide)

Cabbage Fumigation by methyl bromide

(dimethyl sulphide) Geosminand Earthy Algal and mould growth

methylisobomeol Musty

Muddy Guaiacol Smokey Microbial degradation of vanillin

Phenolic Hydrocarbons Plastic/petrol Residual styrene monomer in

polystyrene Chemical Board surface coatings

Microbial degradation of sorbic acid Microbial degradation of

cinnamaldehyde Oil spillages at sea Aerylates Plastic UV -cured inks and vamishes

Acrid

trigger perception (Moskowitz, 1983) but, contrary to common usage, a number

of thresholds can be defined, none of which is invariant ISO 5492 (ISO, 1992)gives the following definitions for thresholds:

Detection threshold: the minimum value of a sensory stimulus needed to

give rise to a sensation

Recognition threshold: minimum value of a sensory stimulus permitting

identification of the sensation perceived

Difference threshold: value of the smallest perceptible difference in the

physical intensity of a stimulus

Terminal threshold: minimum value of an intense sensory stimulus

above which no difference in intensity can beperceived

When dealing with taints we are generally concerned with detection thresholds,

ozW :::J

>-@ If

FOOD TAINTS AND OFF-FLAVOURS

Value/range (ppb)

4.5 0.19-30.0 4.5-400 0.2-10 150-300

Sources: Fazzalari (1978) and Pangborn (1981)

SENSORY EVALUATION OF TAINTS AND OFF-FLAVOURS

Table 1.4 Influence of practice on individual taste recognition thresholds

Sucrose Citric acid Sodium chloride Caffeine

2800 33 126 183 Threshold values given in J,lg/kg (ppb)

See Table 1.9 for codes

Souce: Maarse et al (1988)

Table 1.6 Aavour thresholds of 6-COC, 2,3,6-TrCA and 2,3,4,6-TeCA in different media Medium

6-COC

Material 2,3,6-TrCA 2,3,4,6-TeCA Tea

Red wine

Whisky

Blancmange

0.03 2

0.016 100 500

15 X 10'6

25 29 Threshold values given in J,lg/kg (ppb)

-Figure 1.2 Theoretical relationships between threshold and rate of increase of perceived intensity

with stimulus concentration Redrawn and adapted from Pangborn (1981).

shown in Table 1.4 Examination of the threshold data in Table 1.3 shows that themedium in which the stimulus is present has a substantial effect on the measuredthresholds Table 1.5 shows flavour detection threshold data, taken from Maarse

et al (1988), on some chloroanisoles in different media Even greater dependency

of detection thresholds on the medium was found by Jewell (1976) (Table 1.6).The influence of familiarity would be expected to be less important in measuringdetection thresholds than in measuring recognition thresholds

A further problem associated with the use of threshold data is the incorrectassumption that threshold concentration and potency are synonymous, since theperceived intensities of different chemical stimuli do not necessarily increase atthe same rate with concentration Figure 1.2 shows a schematic diagram (fromPangborn, 1981), showing theoretical curves for different stimuli Stimuli A and

B have equal thresholds but dissimilar intensities at higher concentrations,possibly concentrations that may be used in products Stimuli B and C havedifferent thresholds but equal intensities at higher concentrations Pangborn alsoindicated that these curves could also represent the same stimulus perceived bydifferent subjects

The wide range of human thresholds to chemical stimuli is a major reason forthe difficultiesin preventing food taints and in positively identifying the causes oftaints Figure 1.3 shows threshold data from studies of Zoeteman and Piet (1973)

on taints in drinking water The figure shows normalised cumulative distributions

of concentrations of l,3,5-trimethylbenzene, geosmin, dimethyldisulphide and chlorophenol, all differing widely in chemical structure and mean threshold.Examination of the curve shows that the threshold concentration difference

2-SENSORY EVALUATION OF TAINTS AND OFF-FLAVOURS 9

20

80 100

01

C 60t;

•

1io., 40

Concentration ratio to threshold

Figure 1.3 Cumulative distribution of sensitivities of 120 people to four different tainting stances (I,3,5-trimethylbenzene; dimethyldisulphide; 2-chlorophenol; geosmin) in water Data from Zoeteman and Piet (1973), and redrawn and adapted from Williams and Aitken (1983) with

sub-pennission from Ellis Horwood Limited, Chichester.

between the 10% most sensitive and 10% least sensitive is over 200-fold, andbetween the 5% most and 5% least sensitive nearly 2000-fold The difference atthe I% most and I % least sensitive rises to approximately 106•Since I in every

100 consumers complaining to a supermarket that food has, for example, anantiseptic taint, would be regarded as unacceptably high, it can be seen that thefood industry must ensure that any tainting species is not only below the meanthreshold level but also below the levels that this small but highly importantproportion of consumers can detect Effectively, therefore, the industry mustideally aim for zero levels of tainting species

When using sensory methods for taint identification and prevention, it istherefore important to utilise human subjects who are known to represent, as far aspossible, those highly sensitive consumers, through selection criteria based onthreshold measurement The methodologies used are similar for measurementboth of individual sensitivity and of mean thresholds

Any test procedure must recognise the potentially wide range of individualsensitivities, and ensure that a stimulus range is presented that is sufficiently wide

to bracket that range In addition, the stimulus should be presented in a medium assimilar as possible to that which will be encountered in subsequent use.If littleinformation on the likely threshold concentration is available, a wide concentra-tion range can be spanned using a logarithmic presentation sequence, in which thestimulus concentration increases in a constant ratio An arithmetic presentationsequence, in which the stimulus concentration increases by a constant added

10 FOOD TAINTS AND OFF-FLAVOURS

Table 1.7 Comparison of logarithmic

and arithmetic presentation sequences Logarithmic

I 4 16 64 128 512 (multiply by 4)

Arithmetic I 5 9 13

17

21 (add 4)

increment, covers a smaller concentration range but offers the possibility of moreprecise threshold data (Table 1.7) In practice, logarithmic sequences are usuallymore appropriate when a wide range of sensitivities is anticipated Detaileddescriptions and critiques of threshold measurement methods can be found in

Brown et al (1978), Pangborn (1981) and Moskowitz (1983).

The coded stimuli are presented to the subjects, who note the samples for whichthey can detect a stimulus or recognise a stimulus, depending on the purpose ofthe test In tests in which the stimuli are presented in increasing order ofconcentration, there exists the risk of 'error of anticipation', i.e subjectsincorrectly anticipate the presence of a stimulus Conversely, if a decreasing order

of concentration is presented, subjects may exhibit the 'error of habituation', i.e atendency to continue to report false positives These errors can, in principle,beaveraged out using a randomised or balanced order of presentation Many taintstimuli, however, are characterised by extremely persistent flavour carry-over,and a high level of such a stimulus can desensitise the palate and rendersubsequent judgements highly unreliable Ascending presentation orders aretherefore commonly used but with modifications to minimise the 'error ofanticipation' One of the simplest ways of achieving this is to include randomblank samples in the presentation sequence A more sophisticated method is touse the 'choose 1 ofn'procedure, wherenis usually two or three samples at eachlevel For two stimuli at each level, one sample contains the stimulus and the other

is a blank The sequences are shown in Table 1.8

It is important to note that, regardless of the method used, care shouldbetaken

to minimise the risk of subject fatigue an carry-over effects by limiting the

Table1.8 Sample presentation sequences

1.2.3 =increasing stimulus concentration

SENSORY EVALUAnON OF TAINTS AND OFF-FLAVOURS 11number of stimuli presented in any given session and by allowing sufficient timebetween sample tastings.

The importance of these factors will depend on the nature of both the stimulusand subject response but no more than four to eight stimuli should be presented in

a single session, and about 1-2 minutes should be allowed between stimuli.If thestimuli are presented in foods, palate cleansers should also be specified and used

As indicated previously, the detection or recognition threshold is commonlydefined as the concentration that is detected or recognised by 50% of a specifiedpopulation, or more precisely as the concentration detected or recognised on 50%

of occasions by each of a specified group of individuals (Land, 1989) Althoughgroup thresholds are quoted frequently, individual thresholds are of morepractical value when interpreting responses from a given group of subjects.There are other methods for threshold assessment that are used for chemical

stimuli but these are rarely used for food taints Ad libitum mixing, for example, in

which the subject adjusts the concentration of a comparison stimulus to apparentequality with a standard (Pangborn, 1981) is only possible with mixable materialsand would be seriously compromised by carry-over effects Methods based onsignal detection procedures, in which responses to a blank: or a stimulus areclassified as 'hit', 'miss', 'false alarm' or 'correct rejection', require a largenumber(ca.200) of measurements per subject (Green and Swets, 1966)

A short-cut signal detection method, the R-index, is less demanding butnevertheless requires about 20 judgements per subject (O'Mahoney, 1979, 1986)

1.3 Sensory descriptions of taints and off-flavours

The problems associated with the verbalisation of perceived sensory phenomenaare often underestimated A common problem encountered when carrying outconsumer acceptance testing is that whilst it is relatively easy to elicit reliableinformation on level of liking,it is difficult to elicit reliable information on thesensory properties that contribute to these levels A major concern in quantitativedescriptive testing is the selection of individuals who are capable of verbalisingsensory perceptions, especially in the presence of many other sensory stimuli.The difficulties in obtaining reliable and consistent descriptions for chemicaltainting species were well illustrated by Thomson (1984) Boar taint is anunpleasant odour sometimes detected when fat from uncastrated male pigs (boars)

is heated The taint was reported (Hansonet al., 1980) to be caused by 16-en-3-one (androstenone), 3 methylindole (skatole) and indole, constituting36%,33% and 7% of the boar taint, respectively The sensory descriptions mostcommonly associated with androstenone are urinous, animal or sweaty (Griffithsand Patterson, 1970; Amoore, 1977); with skatole, mothballs, faecal or musty;and with indole, mothballs, musty/earthy or paint (Harper et aI., 1968) In asurvey of untrained subjects, Thomson (1984) found that only 25% of the termsdescribed androstenone as sweaty, animal or urinous, and about the same

5-androst-12 FOOD TAINTS AND OFF-FLA YOURS

proportion were descriptions such as solvent-like In total, 32 subjects used 22different descriptive tenns Similarly, only 28% and 44% of the tenns used todescribe skatole and indole, respectively, were mothballs or camphor In total, 40different tenns were used to describe skatole and 44 tenns to describe indole It isalso interesting to note that faecal was only used 4 times to describe skatole andnot at all to describe indole, despite the ready association of these tenns and

chemical species in the literature (Whitfield et al., 1982).

It is generally recognised that description of sensory tenns is more readilyachieved when subjects are familiar with those stimuli as a result of pastexperience This is most easily seen in the case of chlorophenols, since exposure

to oral mouthwashes (e.g TCP) results in a ready production of tenns such asTCP, medicinal and antiseptic However, even within a given class of chemicalcompounds, it should be noted that relatively minor changes in chemical structurecan give rise to substantially different perceptions Table 1.9 shows data on theodour description of chloroanisoles, together with the detection thresholdsmeasured in water for a selection of these materials (Griffiths, 1974) Relativelysmall structural differences can result in major changes, not only in odourthresholds but also in the sensory descriptions Although chloroanisoles as a classare often described as musty, other descriptive tenns are often equally appropriateTable 1.9 Odour description (% of terms used) and odour detection thresholds for chloroanisoles Stimulus Musty Medicinal Solvent! Sweet! Detection

and and alcoholic fruity threshold in

SENSORY EVALUATION OF TAINTS AND OFF-FLAVOURS 13and, in the case of compounds such as 2,3,5-TrCA, 3,4,5-TrCA and 3,4-DCA,medicinal and sweet/fruity descriptors are more appropriate Mustiness was moreapparent in chloroanisoles with higher degrees of chlorination and withchlorination in the 2,6 positions Later work (Griffiths and Fenwick, (977)reported the sensory changes produced by replacing the chlorine substituents bymethyl groups The 2,6 compounds retained the musty odour character butthresholds were higher.

Considerable confusion can arise from one curious characteristic of taintingchemicals: the fact that whilst they may generate an unpleasant response in a widerange of foods, they can be important characteristic flavour components in theirnatural food environment This has long been recognised in the whisky industry,where heavily peated whiskies can contain up to 50 ppm total phenols, includingsubstituted guaiacols (Paterson and Piggott, (989)

A similar situation also occurs in the case of the algal metabolite geosmin,which, while a known tainting hazard in water supplies, is a characteristic flavourcomponent of beet

More recent investigations have revealed a potentially important contributionfrom bromophenols to the flavour of fish (Boyle et a/., 1992, (993) Thesecompounds have been known for some time as a source of taint in seafood(Whitfieldet a/., 1988, (992), and some isomers are known to have thresholdseven lower than those of the chlorophenols For example, 2,6-dibromophenol has

a threshold of 0.0005 ppb in water.It has now been proposed that very low levels

of these compounds in saltwater fish contribute to an attractive flavour profile that

is not present in freshwater fish At higher levels, however, their flavourcharacteristics are perceived as taint

A further difficulty lies in the appearance of increasing natural levels of taintingmaterials in the environment The sources of these are often unknown but thepresence of background levels of chlorophenols may result from environmentalcontamination from incineration processes

Environmental concerns are also introducing the risk of another novel source oftaint, that from recycled packaging materials Certain flavour materials can beabsorbed by the packaging material during storage, and these materials can then

betransferred into another food type following recycling of the material This riskhas caused particular concern in the carbonated soft drinks industry, where, forexample, it has been shown that terpenes can be absorbed into PET andpolycarbonate bottles (Neilsen, (994)

Searching the literature for reliable descriptor sets for chemical stimuli is atedious task but one that should not be confined to the food literature Forexample, chemical companies with an interest in pollution control have attempted

to characterise odours associated with the petrochemical industry (Hellman andSmall, 1973, (974) In a study of 101 petrochemicals, odour thresholds, odourqualities and hedonic characteristics were recorded (Hellman and Small, (974).Although the reported descriptors were limited to one or two per compound, thehedonic rating (pleasant/neutral/unpleasant) is potentially useful However, with

14 FOOD TAINTS AND OFF-FLAVOURS

Table 1.19 Possible chemical compounds related to specific

2,6- Dichloroanisole 2,3,6-Trichloroanisole 2,4,6-Trichloroanisole 2,3,4,6-Tetrachloroanisole Pentachloroanisole 2,4,6-Tribromoanisole Geosmin

2-Methylisobomeol I-Octen-3-01 Octa-I ,3-diene

IX- Terpineol 4,4,6-Trimethyl-l ,3-dioxan Heptan-2-one

Trans,trans-hepta-2,4-dienal Trans-I,3-pentadiene 2-(2-Pentenyl) furan Benzothiazole Methyl acrylate Methyl methacrylate Trans-2-nonenal Oct-l-en-3-one Trans-I,3-pentadiene Styrene

only a few exceptions, the compounds studied were not those commonlyimplicated in food taints A consequence of the non-unique nature of sensorydescriptors is that inspection of the indexes of taint data sets can produce arelatively large number of compounds characterised by a specific descriptor

In Table 1.10 data on the possible compounds that may be associated with threecommon sensory descriptors are shown (These data are taken from the Index ofChemical Taints, available to members of the Contaminants Working Group atthe Leatherhead Food Research Association.) Each descriptor may be associatedwith distinctly different classes of chemical structures, adding to the difficultiesexperienced by the chemical analyst in attempting a positive identification Inaddition, the descriptor for a particular compound may change with concentration

and the medium As the concentration of trans-2-nonenal in water increases from

0.2 ppb to 1000 ppb, for example, the descriptors change from plastic to woody to

fatty to cucumber (Parliment et al., 1973) Even the relatively easily identified

chlorophenols are frequently described as plastic when present in orange juice atthreshold concentrations (Kilcast, Crawford and Marchant, unpublished results)

An additional difficulty in using such information is that a specific description andchemical compound may be a result of either external contamination (taint) orinternal degradation (off-flavour) Diagnosis could clearly be hindered if an

SENSORY EVALUATION OF TAINTS AND OFF-FLAVOURS 15incorrect assumption is made For example, oct-l-en-3-one can be present fromautoxidation of fats (Hammond and Seals, 1972; Swoboda and Peers, 1977) orfrom plastics containing the plasticiser di-isooctylphthalate.

Descriptive information is clearly essential in diagnosing the source of taintsand off-flavours but descriptive information commonly available is limited and,

in general, incomplete More comprehensive sets of descriptions from largernumbers of human subjects would ease diagnosis, if available in suitable form

1.4 Principles of sensory evaluation of food

Many methods can be used for the sensory analysis of foods but some care isneeded in selecting appropriate methods for taint assessment and, of equalimportance, it should be recognised that many of these methods requiremodification both in use and in interpretation

Sensory testing methods can be grouped into two broad classes:(i)analyticaltests; and (ii) affective (or hedonic) tests (IFf, 1981) Although each classsometimes contains the same test procedure, the purpose of the test is quitedifferent In analytical tests, which can be subdivided into discriminative anddescriptive categories, the senses of human subjects are used to provideinformation on characteristics of the food The panel of human subjects isconsequently used as the equivalent of an analytical instrument, and steps aretaken to minimise the various forms of bias that can influence their performance,and to reduce the effect of natural biological variation Affective tests (hereafterreferred to as hedonic tests), in contrast, are used to assess the effect of the food onhuman response, normally in terms of preference or acceptability Natural humanreactions are therefore required, which reflect biases normally encountered andalso biological variation that can result from both physiological and psychologicalsources A major practical distinction between the tests that fall into each category

is that analytical tests utilise small numbers of carefully selected and trainedassessors, whereas hedonic tests use relatively large numbers of untrainedassessors, the members being chosen totryto reflect the likely response of a largerpopulation It is normally important not to seek hedonic information from trained

or, conversely, analytical information from untrained consumers The uniqueproblems of taint, however, sometimes require a relaxation of this otherwisestringent requirement, as discussed in section 1.5

The following sections describe the available sensory test methods only briefly;further details can be found in various standards (e.g ISO, 1985) and referencetexts (e.g Stone and Sidel, 1985; Piggott, 1988)

1.4.1 Analytical tests

1.4.1.1 Discriminative, or difference tests. These tests are used to determinewhether or not a sensory difference exists between two samples, and also as a

16 FOOD TAINTS AND OFF-FLAVOURS

means of measuring thresholds and sensitivity, as described in section 1.2 Threetypes of difference tests are used most frequently: (i) paired comparison;(ii) triangular; and (iii) duo-trio Two other tests, two out of five and 'A' or 'not A'are used less commonly, and will not be described here A short-cut signal detec-tion test (R-index) has been reported and may have future applications

Paired comparison test. Two coded samples are presented either sequentially

or simultaneously in a balanced presentation order (i.e AB and BA) There aretwo variations of the test In the simple difference variant the panellists are asked

if there is a difference between the two samples, having been previously informedthat there mayor may not be a difference In the directional difference variant, thepanellists are asked to choose the sample with the greater or lesser amount of aspecified characteristic The panellists may be allowed to record a 'no-difference'response, or they may be asked to make a choice (forced-choice procedure) Theforced-choice procedure is more correct statistically but no-difference responsescan provide a useful source of information The probability of a selection beingmade purely by chance is 0.5, and responses are analysed in terms of statisticalsignificance levels calculated using the binomial distribution This hastraditionally been carried out using statistical tables calculated at fixed levels ofsignificance (e.g I% and 5%) but calculation of exact significance levels (Lewinsand Wilson, 1985) should be used where possible (see section 1.5.1) Numbers ofassessors recommended by ISO 6658 (ISO, 1985) are shown in Table 1.11

Triangular test. Three coded samples are presented to the panellists, two of

Table 1.11 Minimum recommended numbers of assessors (ISO 1985)

Experts· Selected· Assessors·

2

I I 5 5

20 15

5 I 5 5

5

5

302520 10 20 20

100 100

50 (2 samples)

100 (> 2 samples)

50 (2 samples)

100 (> 2 samples)

SENSORY EVALUATION OF TAINTS AND OFF-FLAVOURS 17which are identical, using all possible sample permutations, i.e.

forced-of some foods Statistical significance values can again be calculated from thebinomial distribution No-difference responses can either be ignored and thenumber of judgements reduced accordingly or, alternatively, one-third of the no-difference responses can be added to the correct responses

Duo-trio test. The panellists are presented with a sample that is identified as astandard, followed by two coded samples, one of which is the same as thestandard and the other different They are asked to identify the sample that is thesame as the standard The sample presented as the standard may be the same foreach panellist (fixed reference standard) or may be balanced between the twotypes The test has the same power as the paired comparison test and is analysedusing the same procedures Presentation of an identified reference reduces theneed for expert or selected assessors, and use of a fixed reference standard canreduce problems when flavour carry-over associated with the test sample isanticipated - of potential importance in taint testing An additional advantage ofthe duo-trio test is that if there are difficulties in preparing identical portions of afood sample (e.g pizza), the panellists can be asked to identify the sample that ismost similar to the reference

Difference from control. This is sometimes used when a control is available.The panellists are presented with an identified control and a range of test samples.They are asked to rate the samples on suitable scales anchored by the points 'notdifferent from control' to 'very different from control' This type of test is morecommonly classified as 'similarity/dissimilarity scaling' and is analysed usingmultidimensional scaling methods (Schiffman and Beeker, 1986)

R-index test. This method is a relatively recent development (O'Mahoney,

1979, 1986), and is a short-cut signal-detection method The test samples arecompared against a previously presented standard, and rated in one of fourcategories For difference testing these categories are: (i) standard; (ii) perhapsstandard; (iii) perhaps not standard; and (iv) not standard The test can also becarried out as a recognition test, in which case the categories are: (i) standard

18 FOOD TAINTS AND OFF-FLAVOURS

recognised; (ii) perhaps standard recognised; (iii) perhaps standard notrecognised; and (iv) standard not recognised The results are expressed in terms ofR-indices, which represent probability values of correct discrimination or correctidentification The method is claimed to give some quantification of magnitude ofdifference but its use has not been widely reported in the literature, although an

application to taint testing has been described (Linssen et al., 1991).

1.4.1.2 Descriptive tests.

Classification This method is used to sort items into a pre-defined number of

categories, using a few expert or selected assessors The categories used inclassification are nominal only

Ranking Ranking tests are used to sort several samples into order of intensity

of a specific characteristic Rank totals are calculated over all panellists, anddifferences are interpreted by statistical tests, e.g the Friedman test The tests areoften carried out as a screening test prior to more precise assessments Nomagnitude information is produced and different sets of results cannot becompared

Rating These tests involve classification into categories in the form of an

ordered scale The categories need to be defined clearly and understood bypanellists Unlike ranking tests, estimates of magnitude of attributes are produced.The scales can take various forms, for example graphic or descriptive, andunipolar or bipolar It should not be assumed that the magnitude of differencesbetween adjacent categories is the same throughout the scale

Scoring Scoring is a form of rating using a numerical scale, in which the

scores bear meaningful mathematical relationships Coded samples are presentedsimultaneously or sequentially to panellists in a balanced presentation order Thescales can be category scales, unstructured line scales (or graphic scales) withverbal anchors (Figure 1.4), or in a form of scaling known as magnitudeestimation, ratio scales

Analysis of scored data from two samples can be carried out on two samplesusing simple Hests or, when more than two samples are being assessed, analysis

of variance and multiple comparison tests Checks should be made to ensure thatthe data are normally distributed; if not, non-parametric tests should be used

Simple descriptive tests. Purely qualitative in nature, these tests require eachpanellist to assess one or more samples independently to identify and describe the

not

Figure 1.4 Example of an unstructured line scale.

SENSORY EVALUATION OF TAINTS AND OFF-FLAVOURS 19sensory attributes Following the independent assessments, the results are collated

by the panel leader and a list of descriptive terms is produced based on frequency

of use The order of appearance of individual attributes is also frequentlyrecorded An additional step that is frequently carried out is for the panel leaderand the panel to discuss the descriptive terms and to generate a list of terms that isagreeable to all panellists In order to reach agreement, several discussions may beneeded, and it is important that the panellists can also agree on a definition of eachdescriptive term

Sensory profile tests Profile tests are a means of quantifying the sensory

attributes of foods that have been established using descriptive tests Two mainclasses of tests are used Consensus profiling, based on a procedure termedquantitative descriptive analysis (QDA), uses the agreed list of attributesgenerated by the simple descriptive tests already described Following agreement

of attributes and their definitions, suitable scoring scales (usually unstructuredline scales) are constructed with appropriate anchors, and the panellists are trained

in scoring reproducibly the intensity of the chosen attributes using a training set ofsamples Training can be lengthy for foods characterised by large attributenumbers Once satisfactory reproducibility has been achieved, the test samplesare scored in replicated tests using appropriate statistical designs The data areanalysed statistically using appropriate methods, for example analysis of varianceand multiple difference testing, but multivariate analysis methods such asprincipal components analysis are increasingly common Graphical methods areoften used to display results

The second profile method is free choice profiling, in which the assessors usethe individual attribute list generated as the first stage of a simple qualitative test.Each assessor constructs an individual profile based on his own attributes, and aconsensus profile is constructed mathematically using a technique known asgeneralised Procrustes analysis This procedure reduces panel training timesconsiderably, but the consensus data are difficult to interpret reliably

1.4.2 Hedonic tests

Hedonic tests are used to measure liking, usually in terms of acceptability orpreference Although employing some of the mechanics of analytical tests,hedonic tests demand the use of sufficient numbers of respondents who are,ideally, typical of a larger consumer population In addition, the tests may need to

be carried out in an appropriate consumption situation, e.g at the respondent'shome or in a restaurant The most common tests used are: (i) paired comparisontests (usually paired preference); (ii) ranking tests; or (iii) scoring tests usinghedonic category scales

A consequence of the major conceptual difference between analytical andhedonic tests is that the two types of test should not be carried out with the sameassessors or respondents A notable exception to this general rule, however, lies in

20 FOOD TAINTS AND OFF-FLAVOURS

the adaptation and application of sensory methods in testing for taint, as will bedescribed subsequently

1.4.3 Requirements for a sensory evaluation operation

Everyone has some sense of taste, and for this reason the requirements for a quality sensory evaluation operation are often severely underestimated Simplyallowing people to taste products in an uncontrolled manner is subject to manyforms of biases and errors and will not produce reliable data A successful oper-ation needs visible management support and a high profile Analytical sensorytesting uses the sensory panel as an analytical instrument, and the aim must be toachieve optimum levels of accuracy and precision, while remembering that theindividual components of the instrument (the panellists) are not electronic compo-nents but biological entities Detailed requirements depend on the specific objec-tives and practical constraints, however, several general features can be identified.(i) Clearly defined objectives Sensory data are often misdirected andmisinterpreted if the objectives of the operation are unclear These objectiveswill usually highlight associated requirements

high-(ii) A suitable environment Great care is usually taken to house sensitiveanalytical instrumentation in the correct environment; the same considerationmust be given to human panellists to allow them to concentrate on their taskswith minimum external distractions Important aspects include:

- a dedicated room

- a means of preventing panellist interaction, either by using individualbooths (permanent or temporary) or by imposition of a rigid discipline

- freedom from external noise and odours

- uniform lighting and temperature

(iv) Suitable panellists The type and number of panellists depends on the nature

of the test but several basic screening requirements can be identified forchecking that all panellists to be used for analytical tests are suitable.Panellists must be able to:

- detect and recognise the four basic tastes

- detect and recognise common odorants and flavours, especially those ofdirect interest

- discriminate between different levels of the stimuli of interest

SENSORY EVALUATION OF TAINTS AND OFF-FLAVOURS 21

- verbalise and to communicate perceptions

- work in a team

In addition, it is important to appoint a panel leader who has the expertise notonly to plan, analyse and interpret sensory tests but also to manage andmotivate the panellists

(v) Data validation Quantitative sensory data are usually subject to much morevariation than data from analytical instruments, and careful experimentaldesign is needed to minimise this variability Correct experimental designwill permit a wide range of statistical analysis methods This requires:

• minimising biases arising from panellist interaction

• ensuring balanced sample presentation order

• avoidance of samples with extreme characteristics

• use of random number codes

• limiting sample numbers to reduce fatigue

Data analysis and presentation methods include:

(i) Univariate statistical methods, such as analysis of variance, t-tests, multiplecomparison tests, non-parametric tests;

(ii) Multivariate methods, such as principal component analysis;

(iii) Data presentation as bar charts, star diagrams and product maps

Many of the most important tests are now available within common spreadsheetpackages for personal computers

1.5 Sensory testing for taint

1.5.1 Test selection and modification

Any attempt to select sensory testing procedures for taint testing encounters afundamental problem Since a taint that could spell commercial disaster may only

be detectable by a few per cent of consumers, can sensory tests that, for practicalreasons, using only small numbers of panellists, be designed to guard against thisoccurrence? Although no procedure of practical value can guarantee that a taintwill be detected, steps can be taken to minimise the risk of not identifying a taintstimulus The most important of these are the following

(i) Where the identity of the tainting species to be tested for is known, usepanellists who are known to be sensitive to that species As explained in section1.2, threshold measurements can be used as a guide to sensitivity but it is alsouseful to measure sensitivity at the higher concentrations that may be present inpractice Unfortunately, it cannot be assumed that a panellist sensitive to onespecific tainting species will also be sensitive to other tainting species Inprinciple, therefore, a taint panel would need to be selected on the basis of high

22 FOOD TAINTS AND OFF-FLAVOURS

sensitivity to several different species In general this is not feasible, particularlywhen panels are to be used to guard against unknown taints, and a compromisemust be found by using panellists who are known to be generally sensitive andreliable in their judgements

(ii) If a high-sensitivity panel is unattainable, use as many panellists as possible

in the hope of having someone present who is sensitive to the taint.In section 1.4,recommended minimum numbers of panellists (assessors) of various levels ofexpertise have been given (Table 1.11) When testing for unknown taints, it isimportant that at least these minimum numbers are used, and increased if possible.For example, a triangular test should not use less than 15 panellists, and for keytests at least 30 should be used

At this stage, a note of caution is necessary Practical operating constraintsoften limit the availability of personnel for forming sensory panels Frequently,therefore, a small number of panellists is used to make the replicate judgements;for example a triangular test can use a panel of five to carry out three tests This isallowable in conventional difference testing, since the statistical analysis is,strictly, based on the number of independent judgements and not on the number ofpanellists This procedure is, however, of no value for taint testing; if the fivepanellists used are insensitive to a taint, then no amount of replication willincrease the chances of detecting that taint

(iii) Use of a high-sensitivity test procedure Difference tests are generallymore suitable than profile-type tests, being more rapid while not requiringintensive training In addition, a difference test against an appropriate untaintedcontrol is a relatively easy task for the panellist The publication of taint testsusing the R-index method (Linssen et al., 1991) demonstrates some potentialvalue but this is relatively unproven

When selecting high-sensitivity difference tests, potential problems of flavourcarry-over must be addressed The triangular test is statistically powerful, with a1/3 chance of identifying the odd sample by chance but a balanced presentationorder will involve presenting two tainted samples to all the panellists.Examination of triangular taint test data has shown that in presentationscontaining two tainted samples (TIC, TCT, CTI) identification of the odd sample

is less easily achieved than in permutations containing two control samples (CCT,CTC, TCC) (Wilson and Kilcast, 1984) Limiting the permutations to thosecontaining two control samples may introduce unacceptable bias into the test, andshould only be used in selection or threshold measurement procedures

Paired comparison tests and duo-trio tests, although less powerful, have theadvantage that only one taint sample is presented to each panellist The former test

is not appropriate for taint testing but the constant reference duo-trio test (with thecontrol sample as reference) is becoming used with increasing frequency Themain disadvantage of the duo-trio test, however, is that the lower statistical powerrequires the use of more panellists than the triangular test, with consequentpractical problems Other sensory tests have shown few applications in tainttesting

SENSOR Y EVALUAnON OF TAINTS AND OFF-FLA YOURS 23(iv) Maximise the infonnation content of the test Some purists in the field ofsensory analysis (e.g Stone and Sidel, 1985) maintain that identification of adifference is the only infonnation that should be elicited from panellists, thereason being that any attempt to elicit other infonnation will require differentmental processes that may invalidate the test As a minimum requirement,descriptive infonnation on the nature of any identified difference must berecorded.

At the Leatherhead Food Research Association, two other types of infonnationare elicited Firstly, since taints are by definition disliked, preference infonnation

is recorded As indicated previously, this is a unique exception to the general rulethat hedonic and analytical tests must not be mixed The preference infonnation isnot interpreted as a likely measure of consumer response but is used purely as adirectional indicator in conjunction with descriptive infonnation Secondly,panellists are asked to rate how confident they were in their choice of the oddsample on a 3-point category scale (An example of a typical triangular testquestionnaire is shown in Figure 1.5.) Confidence levels weighted toward one end

of the scale or the other can help resolve indetenninate results by indicating towhat extent panellists may be guessing Such a scale may be fonnalised byassigning scores to the scale points

An important point to note when using such ancillary data, however, is thatthese data are only valid from panellists who have identified correctly the oddsample Data from panellists who have made incorrect identifications are invalidand must not be used

(v) Use statistical tests that are appropriate to taint testing A fundamentalproblem is apparent here, as discussed by O'Mahoney (1982, 1986).Conventional hypothesis testing involves testing the experimental data against anull hypothesis(Ho)that no trend, or difference, exists in the data A probabilityvalue that represents a difference occurring by chance is calculated.If this value islow, it is unlikely that the null hypothesis is true, and the alternative hypothesis(HI)is accepted, which states that a difference is present On the other hand, ahigh value indicates that the result could have occurred by chance, and the nullhypothesis is not rejected A probability value of 0.05 (5% significance) in adifference test can then be interpreted as saying that a difference appears to existbut with a 5%(lin 20) probability that the result could have been due to chance

If we require more assurance that we really have found a difference, a lowersignificance level of I%could be used, giving a 1 in 100 probability of a chanceresult

Unfortunately, the greater the assurance of a real difference that is sought, thegreater the risk of not identifying a real difference that is present (Type II error)

By increasing the significance level to 10%, 15% or even 20%, the risk of notidentifying a real difference diminishes but the risk of incorrectly identifying adifference (Type I error) increases The choice of an appropriate cut-off pointdepends on the required level of risk; even I%would be too high a risk in medicalexperiments, and values of 0.1 % or 0.01 % may be more appropriate In sensory

24 FOOD TAINTS AND OFF-FLAVOURS

c you ct.I8Ct any diffwnnce?

If 10, which is the odd umple?

How confident , you in your choice of 1IIRPIe?

Abeo!utely _e §

Flirly_e Not-vsure

Figure 1.5 Example of triangular test questionnaire.

testing, however, and in particular in taint testing, the consequences of incorrectlysaying a difference exists are relatively minor, compared with the consequences

of not identifying a difference and allowing tainted product to reach consumers.Consequently, levels of up to 20% should be used to minimise this risk butaccepting that by using a 20% cut-off there will be an expectation that overall I in

5 will be incorrect

It should be noted that, when interpreting probability levels, there is littlepractical difference between probabilities of 4.9% and 5.1 % but that if a rigid cut-off of 5% were used, different interpretations would result Consequently, it ispreferable to calculate exact significance values and use common sense in theirinterpretation

SENSORY EVALUATION OF TAINTS AND OFF-FLAVOURS 25O'Mahoney (1982) has made the observation that, for the purposes of sensoryevaluation, there is a need for statistical tests designed to reject the alternativehypothesis (HI) rather than the nun hypothesis(Ho).

(vi) Regardless of the results of statistical tests, take careful note of minorityjudgements, particularly from panellists of established reliability, and re-test foradded assurance Such procedures have been built into guidelines produced fortaint testing with pesticides (MAFF, undated)

1.5.2 Diagnostic taint testing

Unfortunately for the food industry, taint problems continue to be so widespreadthat considerable effort and expense continue to be necessary for diagnosing taintproblems These problems frequently involve insurance claims or litigation and,

in such cases, correct sensory (and also chemical analysis) procedures must beadhered to rigorously

The first indication of a taint problem is usually through consumer complaints

of sensory quality One consequence of the commonly low level of taint detection

is that the complaints may come in at a low rate over an extended period, andrecognition of a taint problem may not be immediate In addition, investigating asensory quality complaint arising from a unit item is undesirable as the safety of

an item removed from its packaging may be questionable In such cases,examination should be restricted to odour and, if feasible, chemical composition.Examination of batches of suspect product should be carried out as a means ofinvestigation but, again, care must be taken to guard against possible safetyproblems The suspect product to be tested should be from the same batch coding

as the complaint material, and as far as possible should have gone through thesame distribution channels In addition, suitable control material of similar ageshould be available In circumstances in which the complaint pattern suggestsnon-uniform distribution within a production batch, testing can be carried out to asuitable statistical sampling plan; such testing can often prove prohibitively time-consuming and expensive Where possible, logical considerations can be used toselect samples, for example those units on a pallet that are immediately adjacent

to the wood

Consumer descriptions of most taints cannot be relied upon as a means offocusing chemical analysis investigations, and sensory testing of suspect batchesshould be carried out to generate reliable descriptive information However, ashas been discussed in section 1.3, care is needed in relating descriptions topossible chemical species

If the presence of a taint in complaint batches can be established, efforts must

be made as quickly as possible to isolate affected product and to identify thesource of the taint Sensory testing can be used to investigate whether the problem

is associated with a single transport container, production run, ingredients batch

or packaging material batch.If the problem appears to be continuing over a period

of time, however, possible sources such as new building materials, process line

26 FOOD TAINTS AND OFF-FLAVOURS

components or water-borne contamination must be examined If ingredients(including water supply) are suspected as continuing sources of taint, small testbatches of product can be prepared and compared against appropriate controls.Materials suspected as sources of taint can be tested using taint transfer tests, asdescribed in the next section

Particular care must be taken when gathering evidence and setting up testprocedures if, as must frequently be assumed, insurance claims or, even moreimportantly, litigation are likely Companies supplying tainted materials may facelitigation by their customers and, in tum, may enter into litigation against theirown suppliers.It is frequently advantageous to contract the testing work out to athird-party organisation in order to establish impartiality in generating data to beused as evidence Care should, however, be taken to establish the scientificcredentials and expertise of such organisations Several suggestions can be madewhen initiating such investigations if the time scales and costs of litigation are to

be minimised

(i) Act quickly to identify the nature and source of the taint

(ii) Use both sensory and chemical analysis to establish both the occurrenceand identity of the taint - do not rely on one type of information only.(iii) Store both suspect and control samples (under deep freeze if necessary)for future testing

(iv) Carry out sensory testing following the general guidelines given in ISO

6658 and use as many assessors (preferably sensitive) as possible.(v) Extract as much information from the tests as possible but do notcompromise the test quality

(vi) Have the tests carried out and interpreted on a double-blind basis,especially if the tests are to be subcontracted

(vii) Ensure that the names and addresses of panellists are held, as tation of sensory data in a court of law may require the presence of theindividual panellists as witnesses

presen-1.5.3 Preventive (taint-transfer) testing

Preventive testing is a powerful, but frequently misapplied, means of limitingproblems arising from the introduction of new materials and changes inenvironmental conditions The tests seek to expose food or food simulants topotential taint sources in an exposure situation that is severe but not unrealistic.Severity factors of up to ten times are usually used but higher factors can be usedfor critical applications An outline protocol for such tests is shown in Figure 1.6.The design of the exposure system varies considerably depending on the nature

of the test For example, taint testing of pesticide residues requires a full-scalefield trial with rigidly defined crop growing, pesticide application and cropsampling procedures (MAFF, undated) When testing packaging systems, themodel system may need to simulate either direct contact or remote exposure,

SENSORY EVALUATION OF TAINTS AND OFF-FLAVOURS 27

Design model exposure system

Select food or food simulant

Expose to taint source for appropriate time

Test against control

in high sensitivity test

Minimum risk interpretation Figure 1.6 Experimental protocol for preventive (taint-transfer) testing.

while when testing process-line components, factors such as product residencetime and product temperature mustbeconsidered

Factors tobeconsidered when designing model systems for testing materialssuch as flooring, paints and packaging materials include the following:

• ratio of the volume or surface area of the material to the volume of the vessel

• ratio of the volume or surface area of the material to the volume or surfacearea of food/food simulant

• stage of exposure (e.g stage during curing of a flooring material at whichexposure is to start)

• length of exposure

• temperature and humidity at exposure

28 FOOD TAINTS AND OFF-FLAVOURS

• exposure method (e.g direct contact or vapour phase transfer)

• exposure lighting conditions (especially when rancidity development mayoccur)

• ventilated or unventilated exposure system

• temperature and length of storage of food/food simulant between exposureand testing

• sensory test procedure and interpretation

Choice of appropriate foods/food simulants is an important consideration, withtwo alternative approaches In situations in which a specific ingredient or product

is known to be at risk, the test can be designed around it Where the purpose of thetest is more general, however, simple foods or food simulants are often used.Solvent of adsorptive properties are probably the most important considerationswhen selecting appropriate general simulants Oils and fats will tend to absorbwater-insoluble tainting species, and materials such as butter are known to besensitive to taint transfer High surface area powders with hydrophiliccharacteristics have also been found to be sensitive to taint transfer, and tend toabsorb water-soluble taints Use of such materials will simulate a large proportion

of the solvent and absorptive characteristics of real foods An additionalrequirement for suitable simulants, however, is that they should be relativelybland to enable easy detection, and also of reasonably high palatability This latterconsideration unfortunately renders some simulants recommended for packagingmigration tests (EC, 1985), for example 3% acetic acid, unsuitable for taint-transfer testing Still mineral water can be used to simulate aqueous liquids, and8% ethanol in water to simulate alcoholic drinks In the author's laboratory,however, the characteristic ethanol flavour has been found to be rather unpleasant,and a bland vodka is now used, diluted down to 8% ethanol Some suitablematerials for general-purpose use are given in Table 1.12

Figure 1.7 shows a schematic diagram of a simple model experiment used at theLeatherhead Food Research Association to carry out taint-transfer testing on

Table l.1Z Foods/food simulants for taint-transfer testing

Sugar

Cornflour

BiscuilS Milk

Comments Bland variely (e.g milk) preferred Mixed prior to sensory teslingorouter surfaces only tested for severe leSI High surface area preferred (e.g icing sugar)

Tesl as 5% solulion Test as blancmange formulation (can gel textural variation) High-fat (e.g shonbread) Full cream For shon-term exposure leslS only or rancidity problems can imerfere

SENSORY EVALUATION OF TAINTS AND OFF-FLAVOURS 29

lid containing _ -ial

Figure 1.7 System employed at the Leatherhead Food Research Association for testing the tainting

potential of building materials.

flooring materials and paints The use of a jam jar lid and a 3-litre beaker give asurface:volume ratio of I: 100; if required this can be changed by using differentlid areas and beaker volumes The beaker can be raised on supports to giveunforced ventilation if required, or a pump can be used to simulate the effect offorced ventilation

In practice, sophisticated procedures such as those described here are oftenimpractical for quality control procedures, and are adapted a short-cut screeningtests

A number of standard procedures for taint transfer testing have been published,mainly aimed at food packaging materials (BSI, 1964; OICC, 1964; DIN, 1983;ASTM, 1988) The British Standard and the American Standard deal with tainttransfer from packaging films in general, and the OICC standard ('Robinson test')deals specifically with taint transfer to cocoa and chocolate products, although it

is frequently used for other products The German DIN standard also refers tofood packaging but contains much useful information for setting up tests on othermaterials All the published methods are, however, deficient in their use ofsensory testing methods, and the more correct and more sensitive proceduresdescribed here and based on ISO 6658 should be used where possible A list ofISO standards in sensory analysis is shown in Table 1.13

In seeking to maintain high food quality and to minimise the risk of taintproblems, Marks & Spencer has developed Codes of Practice referring to the use

of packaging films, plastic and paints (Goldenberg and Matheson, 1975) Theseguidelines stress the importance of testing by the packaging supplier beforedispatch, and by the food manufacturer before use This important principle is,unfortunately, rarely recognised by the food industry in general Foodmanufacturers frequently rely on suppliers to provide some general form ofcertification or test evidence that a material is free from taint but the material isseldom tested under the conditions in which it will be used Information provided

by suppliers can be regarded as useful screening information but users mustprotect themselves by re-testing under more realistic and rigorous conditions

30 FOOD TAINTS AND OFF-FLAVOURS

Table 1.13 Relevant standards in sensory analysis

Title Glossary of terms relating to sensory analysis Specification for wine-tasting glass Introduction and general guide to methodology Paired comparison test

Triangular test Flavour profile methods 'A' - 'Not A' test Ranking Investigating sensitivity of taste Duo-trio test

Initiation and training of assessors in the detection and recognition of odours

Guide to design of test rooms for sensory analysis of food Assessors for sensory analysis Part I Guide to the selection, training and monitoring of selected assessors

Assessors for sensory analysis Part 2 Guide to the selection, training and monitoring of experts

1.5.4 Sensory quality control taint testing

The different test procedures described here need to be used in order to ensurehigh sensitivity In a quality control (QC) environment, however, the use ofrelatively large numbers of panellists is usually impractical, and the requiredthroughput often mitigates against tests that can only compare one sample against

a control In such circumstances, it is often necessary to use short-cut screeningprocedures as part of a more general QC system, and to have in place an alertingsystem and a back-up system for examining suspect foods or materials Effortshave been made to introduce alternatives to difference testing in order to improvethe ease of use of sensory testing for QC purposes but difficulties have beenencountered in setting up appropriate scoring systems One such system has been

described for the evaluation of flavours from packaging materials (Thompson et

aI., 1994) A set of 28 descriptive aroma and flavour terms was established,together with chemical reference standards, and intensities are quantified on 9-point category scales Although designed primarily for evaluating the aroma ofpackaging materials and the flavour of water exposed to packaging materials, theauthors claim that the system can be extended to foods

The importance of having effective systems in place for guarding against taint

is gradually being reflected both in recommendations for good qualitymanagement in the food industry and in legislation Many companies are nowseeking accreditation under ISO 9000IEN 29000, and guidelines prepared for thefood industry (Leatherhead Food Research Association, 1989) have highlightedthe need for systems to prevent the occurrence of taint Perhaps more importantlyfor the future, companies may need to demonstrate that they are taking measures

to prevent taint as a defence under 'due diligence' clauses In practice, it is likelythat satisfactory measures will depend on the resources available to a company