Creation of a food taste database using an in-home “taste” profile method

Food Quality and Preference 36 (2014) 70–80

Contents lists available at ScienceDirect

Food Quality and Preference journal homepage: www.elsevier.com/locate/foodqual

Creation of a food taste database using an in-home ‘‘taste’’ profile method Christophe Martin ⇑, Michel Visalli, Christine Lange, Pascal Schlich, Sylvie Issanchou CNRS, UMR6265 Centre des Sciences du Goût et de l’Alimentation, F-21000 Dijon, France INRA, UMR1324 Centre des Sciences du Goût et de l’Alimentation, F-21000 Dijon, France Université de Bourgogne, UMR Centre des Sciences du Goût et de l’Alimentation, F-21000 Dijon, France

a r t i c l e

i n f o

Article history: Received 31 July 2013 Received in revised form 17 March 2014 Accepted 17 March 2014 Available online 27 March 2014 Keywords: Basic tastes Fat sensation Database Sensory profile In-home method

a b s t r a c t The purpose of this study was to create a food ‘taste’ database using an innovative in-home profile method. The five basic tastes and fat sensation were studied. The proposed method consisted in an intensive training in laboratory (55 h, 5 months) immediately followed by an in-home measurements phase (8 months) during which 12 trained panelists had to evaluate the five tastes and fat sensation of the foods they typically consumed. The rating scales were inspired by scales used in the Spectrum™ method. During the in-home measurement phase, ratings were reported thanks to a web-based tool and each month the panelists returned to the laboratory for a 2-h retraining session. The results showed that the proposed method could lead to results of good quality compared to those obtained in laboratory. Over the in-home measurements period, 590 foods were described (average number of evaluations by food: 8.7). Six major classes of foods were identified on the basis of tastes and fat sensation perceptions enabling a ‘‘sensory’’ classification of foods to be proposed. The food contributors of high intensities were also highlighted, contributing to have an overview on the sensory sapid world we face in our diet. Linking this sensory database with other types of data on food opens new perspectives in nutrition and epidemiology. Ó 2014 Elsevier Ltd. All rights reserved.

1. Introduction Many food composition databases exist. Indeed, more than 35,000 European, North American foods and foods from other countries are now LanguaL™ indexed. But, to the best of our knowledge, no database provides information on the taste intensity perceived in foods. Building a database indicating the perceived intensity of the tastes for the foods which are commonly consumed in different countries would be greatly useful for public policy makers as well as for food industries. First, it is well-known that sensory characteristic, and in particular taste properties, are key drivers of food acceptability. Second, it is also known that some tastes, such as sweet and fat, could act as an early signal of calories and nutrients. A food taste database would provide an overview of the sensory landscape of diets, foods, foods within food groups and it would allow major food items or food groups, which are high-intensity vectors, to be determined. It would also allow studying to what extent ⇑ Corresponding author at: Centre des Sciences du Goût et de l’Alimentation, Centre de Recherches INRA, BP86510, 21065 Dijon Cedex, France. Tel.: +33 3 80 69 30 75; fax: +33 3 80 69 32 27. E-mail address: [email protected] (C. Martin). http://dx.doi.org/10.1016/j.foodqual.2014.03.005 0950-3293/Ó 2014 Elsevier Ltd. All rights reserved.

usual food groups based on culinary or nutritional are homogeneous from a sensory point of view. Combined with nutritional composition tables and food processing information, a food taste database could provide elements of understanding about relationship between food composition and processing characteristics (type of tastants, type of matrix, level of processing, etc.), and perceived taste intensity. A first attempt was made by Viskaal-van Dongen and collaborators on 50 foods in order to correlate taste intensities and composition in macronutrients (Viskaal-van Dongen, van den Berg, Vink, Kok, & de Graaf, 2012). A food taste database would also make possible the calculation of indices reflecting our exposure to tastes, in the same way as intakes for different nutrients are calculated. Identifying patterns of exposure to different tastes, or to foods with particular tastes in target populations (children, adults, elderly) could provide an additional factor, helping to understand the origin of food preferences and of excessive eating behaviours such as too sweet, too salty, or too fatty diets. The difficulty of producing such database lies mostly in the large number of food items available and by variations induced by several factors probably modifying the taste profile of food items, such as trademarks, product origins, degrees of maturity, and transformations made at home (recipes, cooking methods

71

C. Martin et al. / Food Quality and Preference 36 (2014) 70–80

and seasoning). An in-home sensory evaluation of food prepared by the panelists themselves would enable to collect taste profiles for a great number of foods regularly consumed in real conditions and to have an overview of the taste variability of our diet. The purpose of this study was to create a food taste database using an innovative in-home profile method and to have an overview on the sensory sapid world we face in our diet. The intensity of salty, sweet, acid and bitter tastes were scored using universal scales proposed within the Spectrum™ method (Muñoz & Civille, 1992) and newly developed scales for umami taste and fattiness. Fat sensation was added to the studied stimuli for two main reasons. Firstly, although this multimodal sensation is mainly based on characteristics of texture, odour, and appearance (Gaillard, Passilly-Degrace, & Besnard, 2008), it is possible that it relies in part on the detection of free fatty acids in the oral cavity, according to the same type of physiological mechanisms involved in the perception of basic tastes even if this issue is still being debated (Mattes, 2009). Secondly, a potential application of this study would be to correlate sensory and nutritional data for different foods and/or to correlate sensory exposure to socio-demographic data. Thus, it is important to take into account the fat sensation because of its role as an indicator of the fat content of foods and its role in the attractiveness of high-fat foods.

concentrate in performing a complex task (Bourdon T.I.B. test, Swets & Zeitlinger BV, Calisse, The Netherlands), and finally, to assess the intensity of the five tastes in three different foods. On the basis of their results, 16 candidates were selected. Over the 13 months of experimentation, four panelists stopped the study for personal reasons. The 12 remaining panelists (seven women and five men) completed the entire study. The age of the 12 panelists was between 28 and 67 years (mean age = 46 years). 2.2. Experimental conditions The laboratory sessions took place in the tasting rooms of the Chemosens platform (INRA Dijon), where assessments were conducted in individual booths equipped with FIZZÒ software (Biosystèmes, Couternon, France). The room temperature was controlled (21 ± 1 °C), and the samples were evaluated in daylight and in the white light of the rooms and booths. The ‘‘in-home’’ measurements were performed in a completely free environment. The panelists could make the assessments in their homes or during a catered meal, at a friend’s house, or in restaurants. The panelists all signed an informed consent form and were compensated for their participation (15 € per hour for training in the laboratory and 40 € per month for scoring at home). 2.3. Training in laboratory

2. Materials and methods

The panelists were trained during 55 h in 5 months. Given the ‘‘multi-product’’ mission of the panel, special attention was paid to the variety of food used during training. Indeed, over 180 different foods were evaluated. Several commercial foods were modified by adding food grade sapid substances (caffeine, sucrose, monosodium glutamate, citric acid, sodium chloride, differents fats) in order to modify the taste intensity and to obtain series of samples varying mainly by the intensity of one taste (e.g. mashed potatoes more or less salty sweet or bitter, fruits compotes more or less sweet or bitter, soft cheese more or less sweet, sour, bitter or fat, etc.). Items from all food groups, including mixed dishes, were presented. Some foods were selected because of the particularly high intensity of one of their sensations, others for the simultaneous occurrence of several sensations at a high level. Panelists learned to assess the tastes and fat sensation, regardless of the level of

The in-home taste profile method experimented in this study consisted in an intensive training in laboratory, and an in-home measurements phase during which the panelists had to evaluate the tastes and fat sensation of the foods they typically consumed. A general description of the in-home ‘‘taste’’ profile method is presented in Fig. 1. 2.1. Panelists Thirty-two subjects with previous experience in sensory profiling were invited to participate in a selection session. The selection criteria were based on the ability of the subjects to identify the five basic tastes, to classify simple sapid solutions according to the intensity of the perceived tastes, to use a linear scale, to

Training phase Month

1

2

3

4

Measurement phase 5

6

7

8

9

10

11

12

13

Laboratory Panellist recruiting Choice of the food references (scales) Panel training (rating) Evaluation of the panel performances st

1 evaluation of the 14 control products Monitoring sessions Home In-home sensory evaluation nd

2 evaluation of the 14 control products Providing food references

Fig. 1. General description of the in-home ‘‘taste’’ profile method. Twelve panelists with previous experience in sensory profiling were recruited through a 1-h selection session. Five months (55 h) were dedicated to the development of the rating scales (choice of the food references) and to the training in the use of these scales. The panelists’ performances (discriminatory power, agreement within the group, and repeatability) were assessed approximately every four sessions in the second half of the training. A set of 14 foods (control products) was evaluated first in laboratory, at the end of the training phase, and then at home at different times of the in-home measurement phase (from the 2nd to 8th month). The comparison of the results obtained in both conditions gave an idea of the quality of the in-home measurements. The in-home measurements phase lasted 8 months during which the panelists had to characterise the taste and fat sensation profile of the food they typically consumed and to regularly report their ratings thanks to a web-based tool. During the measuring phase, each month, the panelists returned to the laboratory for a 2-h monitoring session. These sessions allowed taking stock of the situation and retraining the panelists with the sapid reference solutions of the Spectrum™ scales and enabled to provide the panelists with food references for the upcoming month.

72

C. Martin et al. / Food Quality and Preference 36 (2014) 70–80

intensity and complexity of the food evaluated. Several training sessions were devoted to the definition of an evaluation procedure of heterogeneous foods and mixed dishes. Initially, group discussion sessions helped to define these two categories of foods. Heterogeneous foods were defined as foods whose sensory characteristics vary from one part to another of the sample evaluated but cannot be evaluated separately (physically impossible) or are not typically consumed independently. For example, a fruit cake covered with a layer of whipped cream was considered as a heterogeneous food. In this case, the panelists were instructed to evaluate the fruit cake as an indivisible whole and to base their assessment on a representative portion composed of the different components of the cake. Mixed dishes were defined as a combination of components sometimes eaten separately, prepared according to an identified and relatively constant recipe. In general, the method of preparing a mixed dish has the effect of modifying the sensory characteristics of the different components in the dish. For example, a cassoulet, a typicall dish from the South-west of France, slow-cooked, containing meat (typically pork sausages, goose, and duck), pork skin and white beans, was considered as a mixed dish. It is indeed possible to separate the various components constituting this dish; however, the sensory characteristics of these components are unique to their combination and likely different from the same foods prepared separately. In this case, the panelists were instructed to base their assessment of the intensity of each sensation on the most intense component of the dish (the saltiness intensity based on a piece of sausage, the sweetness intensity based on beans, etc.). This evaluation method was chosen because under natural conditions, the bites are rarely composed of all the components in the dish but rather of alternating pieces of food constituting the dish. Thus, we considered that the consumption of a mixed dish is therefore equivalent to being alternately exposed to the most intense components of the dish. The usual performance criteria for a sensory profile jury (discriminatory power, explanatory power agreement within the group, and repeatability) were considered and monitored throughout the training (AFNOR, 2009). These performance tests were conducted approximately every four sessions in the latter half of the training. While not reported in this paper, the results demonstrated that the performances of the panelists were very good early in the second third of the training. 2.4. Rating scales Rating scales developed for this study were based on scales used in the Spectrum™ method (Muñoz & Civille, 1992). Subjects evaluated the taste intensity of a food item according to four reference solutions for each taste. These reference solutions contained increasing concentrations of sucrose for sweetness, sodium chloride for saltiness, monosodium glutamate (MSG) for umami taste, citric acid for sourness and caffeine for bitterness, dissolved in EvianÒ mineral water. For convenience, all responses of the panelists were converted into intensity scores from 0 to 10 (instead of 0– 15 in the Spectrum™ method), but the positions (percentage of the original scale) and the concentrations of the reference solutions were conserved. The comparison with previous results obtained with 0–15 scale remains possible by multiplying the scores obtained in this work by 1.5. The reference solutions help the panelists to rate intensity on the scales and permit the ratings collected to be standardised. Indeed, the intensity rating is always made relative to these references. The sapid solutions illustrating the 100% level of the sweet, sour, salty and bitter scales were used during the first training sessions. These solutions were then gradually dropped as they were unanimously considered as too intense compared to intensities generally perceived in foods typically consumed by the panelists. The umami rating scale does not exist in

the Spectrum™ method. Thus, a scale previously developed in our laboratory was used (Martin, Tavares, Schwartz, Nicklaus, & Issanchou, 2009). MSG concentrations are given in Table 1. To develop the fat sensation scale, we initially attempted to use a range of emulsions (oil/water) more or less concentrated in vegetal oil. However, this range proved to be unsuitable in illustrating the different levels of intensity of fat sensation for all types of products, in particular for solid products. Thus, the different levels of intensity of fat sensation were only illustrated by commercial food references as in the Spectrum™ method for all sensory characteristics, except tastes. During the in-home measurements phase it was not possible to periodically provide panelists with these reference solutions for practical and hygienic reasons. Thus, it was decided to provide panelists with food reference. The selection of food references was performed in three steps. First, a preliminary list of foods that could illustrate the different levels of the rating scales was established through discussion with the panelists. Foods commonly consumed by all the panelists and with a dominant taste were preferentially selected. We hypothesised that these foods would be easily memorised as references. Then, these foods were evaluated using the Spectrum™ taste scales (with sapid references). For fat sensation a linear scale (‘‘Not perceived’’ to ‘‘Very intense’’) was used. Products whose scores exhibited the greatest variability between panelists were eliminated from the preliminary list. Moreover, when two products had approximately the same average intensity, only the most frequently consumed and/or most representative of the sensation in question was retained. Then, the remaining products were assessed again (two replicates). Results allowed us to further reduce the list by eliminating foods that had a 95% confidence interval greater than 1.0 on a scale of 10 (0.5 on each side of the mean). Finally, 29 products were selected to illustrate 33 reference positions on six rating scales (Table 1). At the end of the laboratory training period, we verified that the panelists could give reliable ratings without the sapid reference solutions. Six food products (sparkling mineral water, cooked cereal mixture, sweetened biscuit, iced tea soda, coconut milk, vanilla custard) were evaluated once with the help of reference sapid solutions. The panelists were not told that they would have to later reassess these products. Eight days after, the same six food products were evaluated again without using reference solutions. No significant difference (t test, p < 0.05) was observed between the results obtained with and without the sapid solutions, and we concluded that the panelists were sufficiently trained to perform the evaluations without the reference solutions at home. However, panelists will be retrained once a month at lab with these references. During the second half of the training in laboratory, these food references were used as well as the solution references to illustrate different levels of intensity on the scales (Table 1). The objective was that the panelists would gradually base their ratings primarily on food references, as they would be asked to do later during the in-home measurements phase. A monthly distribution of these reference products, for home free consumption, helped to consolidate the memorisation of their sensory characteristics for the duration of the study.

2.5. In-home measurements The in-home measurements phase immediately followed the training phase in laboratory and lasted 8 months during which the panelists had to characterize the taste and fat sensation profile of the food they typically consumed. Each panelist had to evaluate at least 75 food items per month. The panelists were asked to vary

73

C. Martin et al. / Food Quality and Preference 36 (2014) 70–80 Table 1 Nature and position of the food and sapid references on intensity rating scales. Sensation

Food references

Solution references

Food

% Scale

Solution

% Scale

Sweet

White sandwich bread HarrisÒ (plain) Sourdough bun PasquierÒ Actimel DanoneÒ Chamallows HariboÒ Nutella spread FerreroÒ Sweetened condensed milk NestléÒ

11.5 23.9 32.6 59.8 68.0 73.3

Sucrose 20 g
L 1 (R1) Sucrose 50 g
L 1 (R2) Sucrose 100 g
L 1 (R3)

13.33a 33.33a 66.67a

Salty

Cracotte (crispbread) LuÒ Vache qui rit (Processed cheese) BelÒ Cracker Tuc original LuÒ Salted butter (crystals of salt from Guérande) Paysan bretonÒ Soy sauce KikkomanÒ

18.4 32.1 48.0 61.2 81.8

NaCl 2.00 g
L NaCl 3.50 g
L NaCl 5.00 g
L

Petit suisse GervaisÒ Dehydrated tomato soup RoycoÒ Yogurt Velouté DanoneÒ Tomato Ketchup HeinzÒ Lemon juice (fresh and plain)

20.9 24.2 33.7 56.4 93.2

Citric acid 0.50 g
L Citric acid 0.80 g
L Citric acid 1.50 g
L

Mineral water VolvicÒ Camembert LepetitÒ Marmelade of bitter orange Bonne mamanÒ Black chocolate 70% cocoa Lindt ExcellenceÒ Beer (alcohol-free) BucklerÒ Black chocolate 90% cocoa Lindt ExcellenceÒ

4.7 24.6 32.3 38.6 52.6 60.2

Caffeine 0.50 g
L Caffeine 0.80 g
L Caffeine 1.50 g
L

Pâté HénaffÒ Strasbourg sausage Knacki HertaÒ Surimi stickCorayaÒ Soy sauce KikkomanÒ

15.5 22.9 41.7 61.6

MSG 1.20 g
L MSG 3.00 g
L MSG 7.00 g
L

Single cream 15% fat PâturagesÒ Strasbourg sausage Knacki HertaÒ Vache qui rit (Processed cheese) BelÒ Camembert LepetitÒ White chocolate Galak NestléÒ Mascarpone FioriniÒ Unsalted butter PrésidentÒ

13.3 24.6 34.3 45.6 60.0 72.0 90.0

– – –

Sour

Bitter

Umami

Fat sensation

1 1 1

1 1 1

16.67a 33.33a 56.67a

(R1) (R2) (R3)

1 1 1

1 1 1

(R1) (R2) (R3)

(R1) (R2) (R3)

(R1) (R2) (R3)

13.33a 33.33a 66.67a

13.33a 33.33a 66.67a

13.33b 33.33b 66.67b – – –

Sucrose, caffeine, citric acid, sodium chloride (NaCl), and monosodium glutamate (MSG) were all of pharmacopoeia quality, and dissolved in EvianÒ water. a Muñoz and Civille (1992). b Martin et al. (2009).

the foods evaluated as much as possible, but were allowed to rate the same product up to six times throughout the study. A tasting notebook was developed to allow the panelists to note down the results of their assessments during the tasting. The front page of the notebook reminded the panelists of the scoring instructions. On each subsequent page, the panelists were asked to write the date, time, and place of evaluation; the name of the food tasted; and the intensity score for the five tastes and for the fat sensation (see Fig. 2). The rating scales were the same as those used during the last phase of training in laboratory. These scales included the Spectrum™ mark references and the names of the food references. Each page also included additional fields where the panelists were asked to record informations about the products evaluated (method of preparation or cooking, seasoning, ingredients, and particular features). The panelists had to regularly report their ratings thanks to a web-based tool developed in PHP (Hypertext Preprocessor) and JavaScript. The data were stored in a MySQL database and were available in real time. This Internet form allowed panelists to copy out all the informations written in the tasting notebook. To standardise the data collected, the names of foods were selected on a dynamic predefined list. Each month, the panelists returned to the laboratory for a 2-h monitoring session. These sessions allowed taking stock of the situation and retraining the panelists with the sapid reference solutions of the Spectrum™ scales. The tests were primarily intended

to verify that the panelists were able to assign scores as close as possible to the expected scores for all controlled sapid solutions and foods for which a sensory profile was established during training. The monitoring visits also enabled to provide the panelists with food references for the upcoming month. The panelists were asked to consume these food references as often as possible and to refer to these references when they were in doubt about the intensity corresponding to the label placed on the scale. The panelists were also asked to bring back the notebook to the laboratory each month during the monitoring sessions, and they were given a blank notebook in exchange. 2.6. Control of the quality of the in-home measurements To control the quality of the in-home measurements, a set of 14 foods (control products) was selected and evaluated (2 replicates) at the end of the training in laboratory. The panelists were not told they would later have to evaluate these products. We considered the results of this evaluation as the reference. These control products were then evaluated, at home, at different times of the inhome measurement phase (from the 2nd to 8th month). Control products were either commercial foods eaten without preparation (cakes, puddings, fruit juices, prepared foods, etc.) or foods that required a mixing, heating, and/or rehydration operation, to be made by the panelists. In this case, the panelists were asked to strictly follow the manufacturer’s recipe on the packaging and to evaluate

74

C. Martin et al. / Food Quality and Preference 36 (2014) 70–80

Date: ........ / ......... / 2010

Time: ......... h.........

Context:

Food : ..................................................................................................................... White sandwich bread

Sweet

Sourdough bun

Pet suisse

Sour R1

Pâté Hénaff

Camembert

R3

Yogurt Velouté

Ketchup

Signle cream 15%

Knacki

R2

Nutella Condensed spread milk

Soy sauce

R3

Beer 0%

Black chocolat 90%

R2

R3

Soya sauce

Surimi

Vache qui rit

Other occasion

Lemon juice

R2

R1

Fat

R2

Knacki

Invited

Salted buer

Tuc Original

Black hocolat Marmelade 70%

Volvic

R1

Umami

Soupe de tomate

Canteen / Restaurant

R3

Vache qui rit

Cracoe

R1

Bier

Chamallow

R2

R1

Salty

Acmel

Home

R3

Camembert

White chocolat

Mascarpone

Unsalted buer

Comments : Fig. 2. Example of one page of the notebook given to the panelist for rating the perceived taste intensities for the in-home measurements.

the products without changing the seasoning. The quality control test consisted in comparing the results obtained in laboratory and at home.

2.7. Data analysis The data (intensity scores ranging from 0 to 10) were processed using SASÒ software (version 9.1.3, SAS Institute, Cary, NC, USA). For each control product, the assumption of equal average scores for each of the sensations in both conditions (laboratory/non-laboratory) was tested using a paired Student t test (threshold set at 5%). The equality of variances tests were performed using the Folded F method (TTEST procedure), with a 5% threshold. The agreement within the panel about the classification of products according to the intensity of the six tastes was investigated using Kendall’s W. Analyses of variance were performed using the GLM procedure (threshold set at 5%). Multiple comparisons of the means were performed using the LsMeans function (threshold set at 5%). Pearson’s correlation coefficients were calculated to study the relationships between two variables (CORR procedure). The multidimensional representation selected to present the data is a biplot (superimposition of observation components and variables loadings multiplied by a constant) from the principal component analysis (PCA) performed on the covariance matrix (Gabriel, 1971). The VARCLUS procedure was used for the classification of variables (centroid option). The threshold selected for the contrast tests was 10%.

3. Results 3.1. Quality of the in-home measurements The quality control test consisted in comparing the results obtained first in laboratory and then at home, for a set of 14 products. The in-home measurements occurred 3–8 months after the laboratory evaluation; we can thus rule out the hypothesis that the panelists would have memorized the intensity scores assigned at the first evaluation. The results demonstrated that for 13 of the 14 control products and for all the studied sensations, the mean scores of intensity did not differ significantly (p > 0.05) from one assessment to the other (Table 2). An extract of the results obtained in both conditions (laboratory vs. home) for four control products is given in Fig. 3. A single product (vanilla custard) had different average scores for sweetness from one condition to another (p < 0.0001). No significant difference was observed for the other five studied sensations. Unlike the other 13 products, this control product was not precisely the same in the two evaluation conditions (same registered trade name but different volume packaging). Indeed, the nature of the ingredients varied slightly from one product to the other. It was therefore impossible to ensure that both custards would have the same sweet intensity. Given this doubt, we neglected the difference observed for sweetness. The tests of equality of variance, performed in order to compare the dispersion of scores in both assessment conditions, demonstrated that in 79% of cases, the variance of the scores was not

Table 2 Comparison of average scores of intensities obtained for the 14 control products, in both conditions (home/laboratory). Food

Condition

Barquette LUÒ (strawberry)

Mediterranean cereal, TipiakÒ Vanilla custard Danette, Danone

Ò

Pink grapefruit juice, PamprylÒ Coconut milk, Suzi WanÒ Ò

Drink with tea extract, peach flavour, Ice Tea Chocolate bar, Mars

Ò

Chocolate custard, Mont blancÒ Napolitain cake, LUÒ Sparkling mineral water, Perrier

Ò

Vermicelli soup, KnorrÒ Dried sausage Bâton de berger, Justin BridouÒ Ò

Tabbouleh, Garbit

Salty

Sour

Bitter

Umami

Fat

p Value

Mean

p Value

Mean

p Value

Mean

p Value

Mean

p Value

Mean

p Value

4.73 5.24 1.62 1.57 0.78 0.74 4.12 5.55 2.13 2.08 2.33 1.41 3.38 3.19 5.60 6.37 4.34 3.85 4.93 5.83 0.11 0.05 0.39 0.53 0.29 0.28 1.19 0.91

0.29

0.53 0.25 2.09 2.64 2.52 2.55 0.09 0.21 0.18 0.03 0.39 0.13 0.08 0.07 0.54 0.43 0.15 0.13 0.57 0.68 1.40 1.38 3.36 3.87 4.39 5.14 2.42 2.90

0.17

1.25 0.85 1.98 2.48 0.69 0.52 0.34 0.35 3.55 4.77 0.90 0.66 1.70 1.61 0.43 0.25 0.63 0.27 0.34 0.38 1.63 1.99 0.60 0.86 0.74 0.47 1.36 1.46

0.32

0.35 0.13 0.49 0.73 0.61 0.36 0.23 0.19 2.81 3.16 1.18 0.54 1.01 0.78 0.25 0.11 1.17 0.84 0.38 0.43 0.82 0.93 0.31 0.60 0.53 0.32 0.74 0.51

0.11

0.16 0.10 0.57 0.64 1.64 1.41 0.28 0.33 0.22 0.10 0.48 0.23 0.28 0.07 0.32 0.13 0.32 0.12 0.26 0.28 0.14 0.03 1.54 1.98 1.57 1.30 1.09 1.83

0.60

1.53 1.00 1.52 1.53 2.03 1.32 2.58 3.74 0.13 0.08 2.90 2.20 0.23 0.03 3.89 4.23 3.23 2.58 2.80 3.37 0.10 0.02 1.35 1.77 4.09 4.15 2.23 2.57

0.25

0.87 0.85 <0.0001 0.92 0.13 0.72 0.06 0.12 0.10 0.54 0.53 0.94 0.13

0.30 0.95 0.38 0.19 0.11 0.94 0.69 0.87 0.70 0.96 0.21 0.20 0.27

0.19 0.50 0.95 0.16 0.61 0.86 0.36 0.15 0.86 0.67 0.40 0.38 0.82

0.33 0.14 0.81 0.46 0.17 0.48 0.27 0.22 0.81 0.78 0.26 0.38 0.37

0.77 0.62 0.81 0.49 0.35 0.26 0.36 0.32 0.93 0.28 0.23 0.48 0.08

0.97 0.09 0.08 0.56 0.34 0.08 0.72 0.19 0.43 0.11 0.24

C. Martin et al. / Food Quality and Preference 36 (2014) 70–80

Grated carrot with sauce, C. LégerÒ

Home Laboratory Home Laboratory Home Laboratory Home Laboratory Home Laboratory Home Laboratory Home Laboratory Home Laboratory Home Laboratory Home Laboratory Home Laboratory Home Laboratory Home Laboratory Home Laboratory

Sweet Mean

0.94 0.41

p Value: p value paired Student’s test.

75

76

C. Martin et al. / Food Quality and Preference 36 (2014) 70–80

Fig. 3. An example of the results obtained in both conditions (laboratory vs. home) for four control products. The number of days between the two evaluations is indicated under the name of the food. The complete results (14 control products) are presented in Table 2.

Table 3 Comparison of the Fproduct obtained in both conditions (home/laboratory). Home

Sweetness Saltiness Sourness Bitterness Umami Fat

Laboratory

Fprod

p Value

Fprod

p Value

56.40 44.28 9.71 16.71 13.39 22.73

<0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001

95.99 58.91 22.17 18.76 15.98 18.48

<0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001

significantly different from one condition to the other. When variances differed significantly, nine out of ten times the variance of the scores obtained at home was highest. The observed differences in variance were not observed for a particular product or sensation. The results of the variance analysis with two factors (food, panelist) performed for each taste demonstrated that, although still highly significant (p < 0.0001), the FProd values are on average 41% higher for the laboratory data (Table 3). The discriminatory power of the panel, although very satisfactory in both cases, was slightly higher for the laboratory measurement. Furthermore, no significant drift ratings were observed over time. Given these results, we considered that the in-home taste profile method allowed data of good quality to be obtained. 3.2. Taste database 3.2.1. General description of the data During all the in-home measurement phase (8 months), the 12 panelists provided 8175 taste and fat sensation profiles. The term ‘‘profile’’ is used to denote the result of the evaluation of one food by one panelist. These data were checked manually, which led to the elimination of 37% of the data, mainly because of doubts about the product names, revealed by the mismatch between the names of the food selected and the informations entered in the comment field. Food name confusions or approximations were mainly observed in the first part of the in-home measurement phase and became less frequent with increasing experience of the subjects. A specific training at the end of the training phase in laboratory would limit this problem. After verification, the final data table contained the results of 5173 profiles of 590 foods (Appendix A). The term ‘‘food’’ takes into account the mode of preparation or seasoning of a food item. For example, coffee and sweetened coffee are

considered two different food items because they differ in terms of tastes. On average, each panelist made 423 assessments (min = 174, max = 747) of 174 different foods (min = 94, max = 261). For each family of foods, on average, twenty foods were characterised. However, high variability was observed. Thus, the milk family included only one food, while the family of vegetables included 68 different foods. The ten food groups containing the most items were the following, in descending order: vegetables, fruits, cheeses, mixed dishes, breads, crackers and cereals, pastries and cakes, fish, cold cuts, starchy vegetables (tubers, pulses, etc.), and alcoholic beverages. The average number of evaluations by food item was 8.7 (min = 1, max = 77). Each food was evaluated on average by 3.5 different panelists (min = 1, max = 12). The ten most frequently evaluated foods were the following, in descending order: pork (grilled or roasted), melon, green salad with vinaigrette, fries or potatoes, apple, yogurt with fruit flavours, Comté cheese, fruit tarts, pears, and camembert cheese. For all the studied sensations, the intensity scores collected are mainly in the first third of the scale (Table 4). Depending on the tastes, between 77.2% and 98.6% of the values are below or equal to the R2 Spectrum™ reference (33.3% of the scale). This observation is consistent with results obtained by Viskaal-van Dongen et al. (2012) who also used the SpectrumÒ scales. For the bitter and umami tastes, approximately 80% of the values lie below the R1 Spectrum™ reference (13.3% of the scale). The means, calculated on all products for each taste, range from 0.74 (umami) to 1.99 (salty). The parameters of the distribution of the intensity scores for each of the studied sensations are presented in Table 4. Biplot representations of the principal component analysis (PCA) performed on the covariance matrix are presented in Fig. 4. The three most discriminative tastes were saltiness, fattiness, and

Table 4 Parameters of the distribution of the intensity scores. Parameter

Sweet

Salty

Sour

Bitter

Umami

Fat

Mean Standard deviation 100% (max) 75% Q3 50% (median) 25% Q1 0% (min)

1.66 1.57 7.40 2.81 0.95 0.50 0.00

1.87 1.45 5.70 3.01 2.20 0.23 0.00

1.31 1.10 9.25 1.73 1.10 0.58 0.00

0.82 0.76 6.00 1.05 0.65 0.34 0.00

0.79 0.81 4.50 1.27 0.57 0.13 0.00

2.14 1.41 8.37 2.99 2.06 1.10 0.00

77

C. Martin et al. / Food Quality and Preference 36 (2014) 70–80

7 2

2

2

2 22 Sweet

2

7

3

5

2 2

4

6

62

Fat

1

3

Sour 5

1

2 22 2 2 2 2 3 1 2 2 2 222 22 11 2 2 1 1 2 2 222 22 2 21 11 1 1 2 2 2222 2 2 4 1 2 2 1 1 2 1 1 11 11111 22 222 6 1 1 1111 1 1 2 22 22222222 2 6 1 111 11 11 1 1 2 22 421 22 2 2222222222 222 1 111 11111111111 1 2 1 5 2 1 1 1 111 1 1 22222 2 2 4 2 22 22222 1 11111 11111111111111151 Salty 2 4 2222 2 2 6 51 11 11 11 111111111 11111 11 1111111 5 2 111 22 22 2 0 5 16Umami 16 11111111111 61 111151111111 11 1 1 3 3 2 2 222 2 2 22222 6 1 1 1 2 6 1 1 1 1 1 5 1111111 6111111 5 -5 -43 323-3 -22 23 -1 6360 61 6115 2 1111 31 4 5 2 1 2 5 2 4 1 1 1 2 2 3 33332232 55511 1115 2 44 36 6 551661611615111 11 2 4-161 6 611 61 61115115511111 2 3222 423 3 33334 6 55155511 115 5 65 333 3333 2 3 2 44 66 6Bitter 1 6 5 6 5 6 5 5 5 4 2 4 564166 1555655 55 1 3 3 33 3 3 5 433 5 565 65 55 3 5 65 3 33 3 44Sour 5 -2 3 6 5 5 3 3 33 4443 4 5 3 3 3 4 3 -3 4 3 3 3

5 4

3

Dim3: 12.8%

2

Dim2: 22.6%

3

6 2 2

-6

8

2

3

-4 -5

Dim 1: 49.8%

-6

-5

32

3

2 3 3

3 33

334 3

5 5

5 5

55 4 Bitter 55 2

5

Fat

1

5

1

1 11

1 1 5 5 1 1 1 11 22 3 5 5 5 5 42 1 5 2555 555155 11 11 1 1 1 3 3 2 5 111 1 111 5Umami 11Salty 1 1 22 2 2232 32 223 3 222 1 5 33 32 23 2 23 3 2 6 55 5 551 1111111 111111111 1 2 23 3 24 2 23 22 11115111 111 11 56 565 615 11111 1 1 1 2 2 1 2 2 2 22332322322322232 1 1 1 5 1 1 4 44 55 2 155115 15 111 1 11111111 1 11611111 0 1111 32 22224222 2 11 1 111511111 11111 1 1 2322 2232223 32 111111 23 4 223 4442 4 56 616 61 15111 1515 2-222 2 -1 -4 2 -3 2 2 2111 11111131 4 5 44 4 6 06 5161 6511111111111111111 2 2 223 2 22222222322 11 11 1 11 11 11111 1 6 444 6 6 6 5 1 1 1 1 1 1 1 1 2 2 22222 6 2 1 1 3 4 22 2 -1 66651 1 6 1 111 1111 2 2 2222222 1 3 5 11 2 2 222 1 4232 42566266 6 666166 116 61116 1 2 1 2 2 2 6 6 166111 1 2 2 22 222 2 16 6 6 -26

22 2

6

3

33

3

33 43 3

2 32 2 224 Sweet 2

6

-3 -4

Dim 1: 49.8%

Fig. 4. Biplot representations from covariance PCA of 590 foods means (planes 1–2 and 1–3).

sweetness. As expected, sweetness and saltiness were negatively correlated (r = 0.65, p < 0.0001). The fat sensation appeared to be related more closely to saltiness (r = 0.53, p < 0.0001) than to sweetness (r = 0.10, p = 0.0116). The other tastes (umami, bitter and sour) were less discriminating. Umami was positively correlated with saltiness (r = 0.66, p < 0.0001). 3.2.2. Six major classes of foods A classification based on the average scores of intensity obtained for the six studied sensations suggests to classify the 590 foods into six homogeneous subsets with contrasting general sensory profiles (proportion of explained variation: 78%). The results of the contrast tests, comparing for each taste the mean of each group to the general average, enabled to specify the main characteristics of these groups (Table 5). The food items composing these six classes are presented in Appendix A. Class 1 (43% of the data) included both more salty/umami/fatty and less sour/bitter/sweet foods, compared to the average. Food items in this class belonged to very different groups: 91% of all cheeses, 83% of all mixed dishes, 94% of all cold cuts, and 73% of

all fish (often grilled, in gravy, or breaded). These four categories accounted for 57% of foods in Class 1. Food items in Class 2 were sweeter than the average, and less intense for other tastes. This food class contained 90% of all pastries, 48% of all fruits (fruits with added powdered sugar, some compotes, jams, etc.), 56% of all breads (eaten with jam, honey, or a spread), and all the fresh desserts (mainly custards and mousses). These four categories accounted for 66% of the food items in Class 2. Class 3 included sweeter than average foods, which, unlike the food in Class 2, were also more bitter and sour than average. The intensities of other perceived tastes were below on average. In total, 52% of all fruits (almost all fresh fruits and fruit juices) and 60% of all alcoholic drinks (aperitifs made of wine, flavoured beers, some sweet wines) were in this class. These two categories accounted for 81% of the food items in Class 3. Food items in Class 4 were mainly bitter. Other tastes were present at a level below the average. In total, 54% of all unsweetened hot drinks were in this class (coffee, tea), along with 25% of all alcoholic beverages (mostly beer) and only 6% of all vegetables (broccoli, cauliflower, and green beans). These three categories accounted for 67% of foods in Class 4. Foods in

Table 5 Main characteristics of the six classes of foods.

The signs ‘‘+’’/‘‘ ’’ indicate that the class average is significantly higher/lower than the grand mean. The double horizontal bars delimit groups of tastes that co-vary (hierarchical clustering).

78

C. Martin et al. / Food Quality and Preference 36 (2014) 70–80

Class 5 were characterised by more intense saltiness, umami, sourness, and bitterness than average. These food items were less sweet and less fat than average. This class included 47% of all vegetables (eaten cold with vinaigrette) and 24% of all fish (usually flavoured with lemon). These two categories accounted for 71% of foods in this class. Finally, Class 6 consisted mainly of salty foods. The intensities of other sensations were below average. In total, 19% of all vegetables, 22% of all breads, and 29% all starchy vegetables (mainly potatoes) are in this class. These three categories accounted for 65% of the food items in Class 6. The major food groups commonly recognised are based on the origin or nature of the products (vegetables, fruits, meats, fish, etc.), their use (desserts, snacks, breakfast cereals, etc.), or on nutrient criteria (starches, fats, etc.). The comparison between these usual food groups and the six classes defined in the present study according to sensory criteria showed that some food groups are very homogeneous. This is particularly the case for offal, processed foods based on fish, eggs and egg-based foods, pizzas and quiches, meat, poultry, cold cuts, cheeses, and mixed dishes. Eighty-three to 100% of foods belonging to these food groups were categorised in the sensory Class 1. Sweet cookies, pastries, chocolates, fresh or frozen desserts, milk, and soft cheeses and yogurts, were also very homogeneous. Indeed, 87–100% of foods belonging to these food groups were categorised in the sensory Class 2. Some groups were divided into two sensory classes. This was the case for crackers and chips, which were distributed in equal proportions in sensory Classes 1 and 6. These salty foods differed in fat content and the presence or absence of ingredients providing umami taste. Hot drinks (coffee, tea, herbal teas, milk chocolate) were divided almost equally between sensory Classes 2 and 4, depending on the addition of sugar or not. The fruit family was splitted in sensory Classes 2 (dried fruits, canned fruits, some fruit juices, bananas, dates, figs, some compotes, fruit jams) and 3 (other fresh fruit, fruit juice, some compotes). Snacks and sandwiches were distributed unequally between sensory Classes 1 (75%) and 6 (25%), depending on the ingredients used. The other food groups were even more heterogeneous and distributed in more than two different sensory classes (3–6). 3.2.3. Food contributors of high intensities For each of the sensations studied, the examination of food items in the last decile (distribution of the intensities) allowed the most intense foods to be determined, among the 590 characterised foods. These foods are detailed in Appendix A. For sweetness, the highest intensities (>4.2/10) were observed in pastries, jams and fruit compotes, fruit juices, dried fruits, chocolates, fresh or frozen desserts, sweets, soft drinks, cookies, cereal for breakfast, and alcoholic aperitif beverages. The last decile also included food items for which sweetness was provided by the addition of jam, honey, or spread. The saltiest foods (>3.7/10) were mainly cheeses, mixed dishes, condiments, cold cuts, fish or processed foods based on fish, aperitifs, grain products, and snacks. The last decile also included many foods for which saltiness was provided by the addition of mayonnaise, dressing, salt, or salted butter. The most sour foods (>2.5/10) were mainly processed fruits (compotes, juices, jams), alcoholic beverages, and vegetables eaten cold as a salad, for which sourness was intensified by seasoning (vinaigrette, lemon juice, vinegar, and marinade). The most bitter foods (>1.8/10) were mainly fresh fruits, fruit juices, and hot drinks (coffee and tea). Other foods in the last decile included alcoholic beverages, vegetables, and dark chocolate. The most umami foods (>1.8/10) were fish, mixed dishes with meat or fish, and cold cuts. Other foods in the last decile were mostly fish-based foods (surimi, terrines) and shellfish. For the fat sensation, 39% of foods included in the last decile (intensity > 4.0 out of 10) were cheeses, and 12% were cold cuts. Other foods in the last decile belonged to

diverse groups: mixed dishes, pastries, chocolates, and fresh desserts, among others, and sometimes, these high intensities were due to the addition of fatty substances (mayonnaise, butter, spread, and chocolate).

4. Discussion 4.1. In-home taste profile The results of this work demonstrated that the in-home taste profile method allowed data of good quality to be obtained. This conclusion was based on the comparison of the results obtained for a set of 14 products evaluated first in laboratory, and then at home several months later. To the best of our knowledge, this is the first study that clearly demonstrates this possibility. Nogueira-Terrones and collaborators proposed a method to educate and train a panel to conduct a sensory profile remotely via Internet (Nogueira-Terrones, Tinet, Curt, Trystram, & Hossenlopp, 2006). After applying this method to four products (dry sausages), the authors found some disagreement between the results of the Internet and reference panels. They also observed a relatively poorer performance for the Internet panel. However, the method used by these authors was very different from the method used in the present study, making a comparison hasardous. The in-home taste profile allowed data concerning 590 foods typically consumed by the 12 trained panelists to be collected over 8 months. Even if several variants of a given food were considered as different foods (e.g. coffee with or without sugar), the variability for a given variant was sometimes higher than those usually observed for data obtained using a conventional sensory profile in laboratory. This variability could be partially explained by the fact that a same food name covered a multitude of variations (recipe, brands, maturity, seasoning, etc.). Given the good performances of the panel for the 14 well-known control products, it is likely that the variability of scores obtained during this study reflects a real sensory variability, even though in some cases, the large standard error observed for some foods was due to an insufficient number of assessments. Considering the inability to control the exact nature of the foods evaluated by the panelists in their homes and the impossibility to imagine a priori an exhaustive list of all variants for each food, we chose to accept this variability as part of an inhome measurement. The approach consisting in characterizing as many foods as possible using a classical sensory profile in laboratory would have probably led to more accurate measurements. However, this approach would have required a difficult selection of a representative prototype for each food studied and the generalisation of the sensory characteristics of these prototypes to the foods they represent would have caused inaccuracies. For a lot of products, several variants were considered according to the addition of sapid substances. However, this distinction appeared to not always be sufficient. For example, we observed that the range of scores of sweetness intensity for herbal tea and flavoured tea were respectively 2.5/10 and 3.6/10, probably depending on the amount of sugar added, which was not specified by the panelists in this study. Considering ingredient addition in a binary way is therefore most likely not entirely satisfactory. Measuring the amount of sapid substances added to accommodate the tastes of a dish could be useful to draw general rules concerning the sensory input from these additions. However, this measurement would raise the question of the unit to be used to quantify these additions of various types. Furthermore, this additional step would complicate the task of the panelists and could influence the value of the intensity scores assigned. Extending the taste database developed in the present study to other sensory dimensions (odour, flavour, texture, trigeminal

C. Martin et al. / Food Quality and Preference 36 (2014) 70–80

sensations) using an in-home sensory profile method would be an interesting objective but more difficult to achieve for several reasons. If the goal is to provide a comprehensive description of the sensory characteristics of a wide range of foods, then a first difficulty lies in the almost unlimited number of texture descriptors and especially of potentially interesting odours/flavours. The adaptation of Spectrum™ type scales for an in-home use is the second difficulty. The taste Spectrum™ scales are accompanied by reference solutions, prepared with identified chemicals available all over the world. These reference solutions could be considered as universal references of taste intensity allowing choice of equivalent food references adapted to an in-home use, as those used in this study. Unlike, the texture and aroma Spectrum™ scales are supplemented by food references which are not always available all over the world. If the original references are not available in a given country, appropriate local references with the same intensity of a given attribute to the original ones should be selected. Moreover, concerning flavours, the main difficulty is about teaching panelist to translate the intensity of one flavour to the intensity of a different flavour, in order to be sure that the different scales are iso-intense. 4.2. Food taste database The database developed in the present study includes the taste and fat sensation profiles of 590 foods. This database presents a good compatibility with some other types of data concerning foods. For example, for the main usual food groups (meat, fruits, vegetables, cheeses, cold cuts, desserts, yogurt and soft cheeses, cookies, hot drinks, and alcoholic beverages), 71–94% of the foods characterized in the present study were also included in the table of nutrient composition of the SU.VI.MAX study (Hercberg, 2005). Because of the different objectives of the two lists, nutrition in one case and taste in the other, the precision levels of food names may differ slightly. For example, ‘‘pike dumplings’’ and ‘‘poultry dumplings’’ are differentiated in the SU.VI.MAX list but grouped in our list because these foods showed few differences in terms of tastes and fat sensation. We observed that 77–99% (depending on the studied tastes) of the intensity scores were included in the first third of the scales. This observation is consistent with the results of a previous study of Viskaal-van Dongen et al. (2012) who used the Spectrum™ method to obtain an anchored rating of the taste intensity of sweetness, saltiness, savouriness, sourness and bitterness for 50 food items. A major advantage of Spectrum™ scales is to permit the intensity of extreme food to be estimated. However, the present study showed that a large majority of foods frequently consumed were not so intense, compared to the intensities of Spectrum™ reference taste solutions. It would be interesting to develop and to test new scales using the R3 reference solution (66% of Spectrum™ scales) as a maximum intensity. Even if such scales would not be fitted to a small number of very intense foods, they could possibly enable a greater precision for most of everyday foods to be obtained. The variation range of intensity scores was larger for sweetness, saltiness, and the fat sensation. This result could be partly explained by the availability of many substances (ingredients, condiments) for adjusting the intensity of these three sensations. For the other tastes, the variation ranges were smaller, especially for bitterness and umami taste, which were not perceived in approximately one third of the profiles and for which approximately 80% of the intensity scores were included in the first third of the scales. Bitterness is a taste expected in certain foods (beer, coffee, some sodas). However, this taste is moderately sought-after, and adjusting its intensity is most often performed in the direction of a reduction, by adding sugar for example. Moreover, apart from some

79

foods used to bring a touch of bitterness to a dish or drink, very few natural food substances enhance the bitterness of a food product. Umami is not specifically sought-after in the French culture, and monosodium glutamate is rarely used at home to season food. Therefore, the umami taste is mainly perceived in foods that naturally contain compounds responsible for this taste. However, with few exceptions, the natural content of these compounds in food does not lead to very high intensities. According to the results of this study, six classes of food were proposed, showing contrasted sensory profiles. Some food groups are homogeneous (taste and fat sensation) because they correspond fairly well with either one of these six classes established on sensory criteria, despite the many recipes or cooking methods possible. However, for some other groups we observed a greater heterogeneity. The vegetable group is the most disparate. Before seasoning, raw or cooked vegetables do not have very prominent tastes or fat sensation, which could explain why the methods of preparation and seasoning are very influential on these characteristics. The study of the main characteristics of the six classes of food enabled the observation of combinations of the most common sensations. This classification is consistent with results obtained by Viskaal-van Dongen et al. (2012) with a different panel and on different products. Even if saltiness may be encountered on its own in foods (bread, vegetables, etc.), it is mostly associated with umami and the fat sensation (cheese, cold cuts, meat, mixed dishes, etc.). Sourness and bitterness form a fairly constant duo, sometimes associated with saltiness and umami (vegetables with dressing, fish with lemon, etc.) and sometimes with sweetness (fruit and fruit juices, alcoholic beverages, etc.). This duo, however, is rarely present when the fat sensation is intense. Although sweetness is sometimes associated with sourness and bitterness, it is also often encountered alone (sweet cookies, pastries, jams, etc.). While sweetness and saltiness are rarely observed together, some foods evaluated in this study exhibit both characteristics at once (salted butter caramel, bread or crackers with jam or spread, etc.). Even if the nutritional content of the studied foods were not known, it seems that most of the time the high intensities were consistent with the amount of sapid or fat compounds, taking into account any seasonings. This relationship is particularly evident for sweetness and saltiness. Thus, sweeter foods were rich in simple carbohydrates or sweeteners, and the more salty items were rich in sodium. The inverse relationship is not always true. For example, soups and sandwiches are among the major food contributors of sodium in adults (AFSSA, 2002) but are not part of the saltiest foods perceived. The relationship between the composition of organic acids and sourness is also quite clear. Indeed, the main vectors of acidity are foods rich in citric, lactic, or acetic acid (fruits, beers, wines, fermented products, dressing, pickled foods, etc.). The more bitter foods were cheeses, some fruit or fruit juice, and unsweetened hot beverages such as tea or coffee. Except for endive, dandelion, and asparagus, few vegetables are among our list of very bitter foods. Presumably the work on varietal selection (Drewnowski & Gomez-Carneros, 2000) helped to reduce the bitterness of some vegetables, and the method of preparation of these foods can reduce this taste. Conversely, cheeses are rarely cited as an example of bitterness, although many of them exhibit a relatively strong bitterness. The high intensities of fat sensation and salty taste usually perceived in cheeses may explain that bitter taste appeared not dominant. This suggests that to be identified as bitter, a food product must be perceived as bitter and not very intense for the other tastes. The results on umami taste confirm that foods with the most intense umami taste are rich in animal protein (meat-, fish, or egg-based products). As expected, the fat sensation is higher in high-fat foods of various types (cheeses, cold cuts, pastries, etc.) or in food consumed with substances high in fat (mayonnaise, spreads, etc.). However, we note also that the addition of jam or

80

C. Martin et al. / Food Quality and Preference 36 (2014) 70–80

honey increases the intensity of the ‘‘fat sensation’’. This result reinforces the idea that the fat sensation is a multidimensional sensation based largely on texture (Mattes, 2009).

fat sensation) to be proposed. Linking this sensory database with other types of data on food opens new perspectives in nutrition and epidemiology.

4.3. Potential use of the food taste database

Acknowledgements

The table of ‘‘tastes’’ is interesting in itself because it provides a first image of the sensory (taste and fat sensation) world that we face in our diet. The richness of this table lies in the possibility of its cross-correlation with other types of data concerning food. Combined with compositional data, e.g., from the ANSES-Ciqual data bank (ANSES/Ciqual, 2008) or the Oqali observatory (Menard et al., 2011), the taste database would allow, in the spirit of the study by Viskaal-van Dongen and collaborators (Viskaal-van Dongen et al., 2012) an investigation into how the tastes and fat sensation may (consciously or not) inform us about the nutritional content of foods and thus guide food choices and food intake. The sensory impact of key nutrients could be studied, and it would also be possible to detect foods or food types whose the nutritional content can be accurately predict by our senses or, instead, foods that can fool our senses and provide nutritional intake in disagreement with the perceived tastes. Combined with consumption data, this database could be used to calculate indices of exposure to the different tastes and fat sensation, just as it is possible to calculate nutrient intakes, taking into account the composition and frequency of consumption of the types of food eaten during a given period. These indices could provide an additional characterisation to be considered in the study and understanding of the origin of food preferences or the study of excessive eating behaviours (too sweet, salty, or fatty diets). The food taste database could be exploited in the context of epidemiological studies based on dietary surveys conducted on a large scale, such as the Nutrinet study (Hercberg et al., 2010). Contrasting groups of consumers, based on the hypo- or hyper-consumption of foods with an aversive sensory characteristic (such as bitterness) could be identified, and the impact of genetic, physiological, psychological, and sociodemographic factors on aversion or attraction to these foods could be determined.

The authors would like to acknowledge Françoise Durey for her assistance during all the study. The project was carried out with the financial support of the French National Agency of Research within the ALIA 2008 Program (EpiPref Project, ANR-08-ALIA-06), of the Regional Council of Burgundy and of the FEDER (European Funding for Regional Economical Development).

5. Conclusion This study demonstrates the feasibility of an in-home sensory profile applied to taste and fat sensation. The proposed method constitutes the first contribution of this work. The second contribution of this work is an initial food taste database describing the intensity of tastes and the fat sensation perceived in hundreds of foods. This table could be supplemented and the data refined because our method is in principle reproducible. These data enabled a new classification of foods based on sensory criteria (tastes and

Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.foodqual. 2014.03.005. References AFNOR. (2009). Sensory analysis – Performance measurement for panel(s) conducting conventional sensory profiles – Analyse sensorielle. XP V09-503. La plaine SaintDenis, France: Agence Française de NORmalisation (AFNOR). AFSSA. (2002). Rapport du groupe de travail sur le sel. Agence Française de Sécurité Sanitaire des Aliments (AFSSA) (p. 86). ANSES/Ciqual (2008). The ANSES/Ciqual food composition data bank. , 31 July 2013. Drewnowski, A., & Gomez-Carneros, C. (2000). Bitter taste, phytonutrients, and the consumer: A review. The American Journal of Clinical Nutrition, 72, 1424–1435. Gaillard, D., Passilly-Degrace, P., & Besnard, P. (2008). Molecular mechanisms of fat preference and overeating. Annals of the New York Academy of Sciences, 1141, 163–175. Hercberg, S. (2005). Tables de composition des aliments SU.VI.MAX. Paris, France: Economica. Hercberg, S., Castetbon, K., Czernichow, S., Malon, A., Mejean, C., Kesse, E., et al. (2010). The Nutrinet-Sante Study: A web-based prospective study on the relationship between nutrition and health and determinants of dietary patterns and nutritional status. BMC Public Health, 10. http://dx.doi.org/10.1186/14712458-10-242. Martin, C., Tavares, J., Schwartz, C., Nicklaus, S., & Issanchou, S. (2009). The taste of infants’ diet: From formula milks to baby fruits & vegetables. In 8th Pangborn Sensory Science Symposium, Florence, Italy, July 26–30, 2009. Mattes, R. D. (2009). Is there a fatty acid taste? Annual Review of Nutrition, 29, 305–327. Menard, C., Dumas, C., Goglia, R., Spiteri, M., Gillot, N., Combris, P., et al. (2011). OQALI: A French database on processed foods. Journal of Food Composition and Analysis, 24, 744–749. Muñoz, A. M., & Civille, G. V. (1992). The spectrum descriptive analysis method. In R. C. Hootman (Ed.), Descripting analysis testing (pp. 22–34). Philadelphia: ASTM. Nogueira-Terrones, H., Tinet, C., Curt, C., Trystram, G., & Hossenlopp, J. (2006). Using the internet for descriptive sensory analysis: Formation, training and follow-up of a taste-test panel over the Web. Journal of Sensory Studies, 21, 180–202. Viskaal-van Dongen, M., van den Berg, M. C., Vink, N., Kok, F. J., & de Graaf, C. (2012). Taste–nutrient relationships in commonly consumed foods. British Journal of Nutrition, 108, 140–147.