Development of a sensory tool to assess overall liking for the fatty, salty and sweet sensations

Accepted Manuscript Development of a sensory tool to assess overall liking for the fatty, salty and sweet sensations Christine Urbano, Amélie Deglaire, Elodie Cartier-Lange, Virginie Herbreteau, Sylvie Cordelle, Pascal Schlich PII: DOI: Reference:

S0950-3293(15)00185-8 http://dx.doi.org/10.1016/j.foodqual.2015.08.003 FQAP 3012

To appear in:

Food Quality and Preference

Received Date: Revised Date: Accepted Date:

24 September 2014 12 June 2015 4 August 2015

Please cite this article as: Urbano, C., Deglaire, A., Cartier-Lange, E., Herbreteau, V., Cordelle, S., Schlich, P., Development of a sensory tool to assess overall liking for the fatty, salty and sweet sensations, Food Quality and Preference (2015), doi: http://dx.doi.org/10.1016/j.foodqual.2015.08.003

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Development of a sensory tool to assess overall liking for the fatty, salty and sweet sensations Christine Urbanoa, Amélie Deglairea,b, Elodie Cartier-Langea, Virginie Herbreteauc, Sylvie Cordellea, Pascal Schlicha

a

Centre des Sciences du Goût et de l’Alimentation, UMR6265 CNRS /UMR1324 INRA/Université

de Bourgogne, F-21000 Dijon. b

Unité

de

Recherche

en

Epidémiologie

Nutritionnelle,

UMR1125

INRA

/UMR557

INSERM/CNAM/Université Paris 13, F-93017 Bobigny. c

Réseau Mixte Technologique Sensorialis/Centre technique Actilait, 55 Boulevard d’Armorique, F-

35700 Rennes Corresponding author: Pascal Schlich

Abbreviations: Lpref: predicted optimal level Keywords: preference, liking, fat, sweet, salt Running title: sensory tool to assess salt, sweet and fat liking. Research highlights •

Development of sensory tests to assess an overall liking for salt, sweet and fat

•

Selection of the most relevant food models through a series of pretests

•

Feasibility and internal validity demonstrated for each sensation

•

Score computation based on quadratic regression

Abstract Understanding the origin of the overconsumption of too high levels of sucrose, sodium chloride and lipids in foods raises the question of the influence of the hedonics for these sensations. To better understand this relationship, a sensory tool that enables measurement of liking towards sweet, salty or fatty sensations is required. This liking towards a sensation has to be understood as an overall attractiveness of the sensation. Instruments already existing were unsatisfactory as including a limited number of foods not representative of the overall sensation. A set of hedonic tests, named PrefSens, was developed to measure an overall liking for fatty, salty or sweet sensations. Each test consisted in the rating on a 9-point hedonic scale of a food product at 5 different levels of lipids, sodium chloride or sucrose. To build the PrefSens test, a total of 144 food ranges were tested during the development step (n = 341 subjects) including various food categories, matrices and serving temperatures. Then, based on the technical feasibility and ability to discriminate, 32 food ranges were selected (10 for fatty, 10 for salty and 12 for sweet). The perceived intensity of fatty, salty or sweet “sensations” of these selected foods was evaluated by a trained panel (n= 12) using Spectrum® scales. During the application study, the hedonic evaluation of the selected foods was assessed in 6 tasting sessions by 567 subjects over 8 laboratories spread out in France. For each subject and each product, hedonic ratings (y) were fitted versus the level of fatty, sweet or salty (x) in a quadratic regression, from which the predicted optimal level (Lpref) was derived based on the x-coordinate of the parabola's maximum weighted by the correlation coefficient between observed and predicted data. The overall liking score for fatty, salty or sweet sensations was the average Lpref over the compounding food ranges. The sensory profile showed that within each sensation the mean perceived intensity significantly increased in a linear manner with the nutrient level. An exception was for the two lowest levels of lipids which were not significantly perceived differently. Concomitantly, the overall liking scores followed an inverted U-shape centered on the middle level of lipids, sodium chloride or sucrose. The distribution was close to normality. Internal validity and consistency of the items compounding

the sensations of fatty, salty or sweet were demonstrated. PrefSens is an internally valid and original tool that can be applied to a population sample in order to better understand the determinants of dietary behaviours.

1

Introduction The available nutritional guidelines that aim to prevent the health risk induced by

overconsumption of too high levels of sucrose, sodium chloride and lipids in foods (Hercberg, ChatYung, & Chauliac, 2008) are not well followed by the population. Sucrose, sodium chloride and lipids are known to largely contribute to food palatability, as recently discussed (Cornwell & McAlister, 2011). Understanding the origin of this overconsumption raises the question of sensory preferences for the sensations brought by these nutrients (Blundell & Finlayson, 2004). To better understand this relationship, a sensory tool that enables measurement of liking towards sweet, salty or fatty sensations is required. This liking towards a sensation has to be understood as an overall attractiveness of the sensation. Even some authors (Hayes, Sullivan & Duffy. 2010) used a diverse range of solid and liquid foods that varied in sodium to study liking for salty sensations, much of previous studies that assessed such liking have employed one or two simple food models presented with grading contents of sucrose, sodium chloride or lipids. Main food models were: soup, crackers, popcorn, tomato juice or hash browns for salty liking (Beauchamp, Bertino, Burke, & Engelman, 1990; Kanarek, Ryu, & Przypek, 1995; Kim & Lee, 2009, Bobowski, Rendall, & Vickers, 2015; Lucas, Ridell, et al, 2011 ); water solutions, fruits flavoured drinks or dairy products for sweet liking (Beauchamp et al., 1990; Hayes & Duffy, 2008; Monneuse, Bellisle, & Louis-Sylvestre, 1991); biscuits, popcorn or dairy products for fatty liking (Bowen et al., 2003; Engell, Bordi, Borja, Lambert, & Rolls, 1998; Hayes & Duffy, 2008; Kanarek et al., 1995). Results from these sensory tests based on a few simple foods are unlikely to represent liking for the sensation as experienced in various food models, including more complex food matrices. Such liking is thus unlikely to stand for the overall liking for the sensation of interest. A few studies have shown that the optimal concentrations of sucrose, sodium chloride or lipids differed with the food matrices, probably due to a different taste perception according to the food matrix (Drewnowski, 1988; Drewnowski & Schwartz, 1990). In the study of Bertino et al. (1983), the optimal sucrose content was different between a solid and a liquid stimuli

with values being 15% in water vs. 24-30% in cookies, but it was related to a similar level of perceived sweetness (approximately 5 out of a 9-point scale). When Mela (1990) plotted the liking scores for 8 different food matrices (with different fat contents) against the perceived fat level, he observed the expected inverted U-shape ( Pangborn, 1988; Stone & Pangborn, 1990) unlike that observed when plotted against the fat content. Similarly, Hayes & Duffy (2008) observed that, although all their studied groups of subjects liked highly sweet and creamy sensations (in liking by sensation models), the fat and sugar levels for hedonic optima varied (in liking by concentration models). This suggests that the overall liking for a sensation, as measured in different matrices, may be better determined according to the taste perception rather than directly according to the nutrient content. Our purpose was to evaluate the liking of fatty, salty and sweet sensations, through a large number of foods being representative of each overall sensation. Instruments already existing were unsatisfactory for this purpose, as they did not assess an overall liking score for the sensation of sweet, salty or fatty. Thus, the present study aimed at developing such a tool that we called PrefSens. Firstly, a development study was conducted over a large series of food models sought to be representative of the usual food repertory of the targeted population and of the various food contexts. The food models and the grading contents of sodium chloride, sucrose or lipids were selected according to their feasibility and their discriminating ability. The perception associated with each level of nutrients within each food product was assessed by a trained panel. Secondly, an application study was conducted over different laboratory tests spread out in the country. The internal validity and consistency of the overall score were explored using the statistical approach of exploratory factor analysis and by calculating the Cronbach’s alpha coefficients. The distribution of the overall liking scores was examined. PrefSens was developed in the framework of a larger project, called EpiPref, aimed at developing tools for measuring exposure to and liking for fatty, salty and sweet sensations in foods in both consumer panels (several hundreds of subjects) and large human cohorts (several thousands of

subjects).

2 2.1

Methods and participants Development study

2.1.1 Food products and range levels Pretests were conducted over a period of one and half years, over the course of six pretesting campaigns with groups of 43 to 63 participants. Food products were selected to represent the usual foods consumed by the French population, based on different matrix types and different consumption temperatures. Each product had to be “homemade” (not commercially prepared) and easily reproducible. There was no crossvariation, e.g. no concomitant variation of fat and sodium chloride in one product, to avoid as much as possible the interactions between the sensations. For fat, we selected fatty-salty and fatty-sweet products. The ingredients for the fatty “sensation” were sunflower oil, pork fat or whipping cream; for the salty one, it was NaCl (sodium chloride) and for the sweet one it was sucrose, intensified in some samples by a non-nutritivesweetener (aspartame). It was assumed that for most individuals, the hedonic ratings for a food product plotted against the level of salty, fatty or sweet sensations should follow an inverted U-shape curve (Pangborn, 1970; Pangborn, 1988; Stone & Pangborn, 1990). Thus, the optimal level of sweet, salty or fatty sensations, which can be estimated as the curve maximum, can be assessed for most individuals and for each food product if the tested levels cover the U-curve abscissa. For each food product, we tested several ranges of 5 levels of salty, sweet or fatty sensations. We hypothesised that the distribution of the maximally optimal level of sweet, salty or fatty over our population sample would approximate a normal distribution. Based on this, we determined the different levels of lipids, sodium chloride or sucrose. Firstly, the medium level (level 0, L0), which was initially based on the content usually met in the basic commercial products or common recipes, was adjusted to conform to the preferences of approximately 50% of subjects. From this L0 level

was derived the four other levels, by decreasing it (levels L-1 and L-2) or increasing it (levels L+1 and L+2). We checked that over the population sample, the distribution of liking for a food product at different levels of lipids, sodium chloride or sucrose contents approximated a normal distribution. 2.1.2 Hedonic evaluation The products were blind tasted under red light. For each food product, the participants received 5 samples at the same time, each sample corresponding to one level of the food range. The participants had to taste and swallow each sample following a balanced order based on a William Latin square and to rate their subsequent hedonic feeling on a 9-point scale, with anchors “I dislike very much” on the left (coded as 1) and “I like very much” on the right (coded as 9). The subjects were required to eat each sample entirely. We conducted pre-tests in order to derive the right portion sizes allowing product discrimination and small enough to be eaten entirely. For salty, a sufficient portion size to permit a discrimination of liking among the different levels was found to be about 10 g (one or two bites). For fatty and for sweet, the sufficient portion size had to be larger (25-30 g). In order to construct a representative set of tests for each “sensation”, products were served cold (e.g. soft white cheese and milkshake served at about 4°C), warm (e.g. mashed potatoes and vegetable soup, served at about 45°C), or at room temperature (e.g. fish terrine and apple sauce, served at about 22°C), corresponding to the temperatures at which these products are habitually consumed. 2.1.3 Sensory profile A panel of 12 subjects, 5 men and 7 women, aged from 29 to 68 years was trained to assess the perceived intensity of fatty, sweet and salty sensations using Spectrum® scales in the food products that were selected. The anchored points of the Spectrum® scales are presented in Appendix 1. The minimum and maximum possible ratings were 0 and 10, respectively. For each sample, the perceived intensity of two of the “sensations” was evaluated: fat and either salt or sweet, depending of the product.

2.1.4 Statistical analysis The statistical analyses were done using SAS 9.2. Differences were considered significant when p < 0.05. For each food product, the agreement between the expected normal distribution of the hedonic ratings and the observed distribution over all the participants was assessed calculating a kappa coefficient. The expected distribution was as follows: 10%, 20%, 40%, 20%, 10%. Perceived intensity ratings were submitted to a two-way analysis of variance (ANOVA) with subject and nutrient level (5 categories) as factors for each product range. Then, for each of the 3 “sensations” (fatty, salty and sweet) a unique ANOVA was run mixing all product ranges according to the following model: perceived intensity = subject + level + product(level). In this model, level was tested against product nested within level. If level was significant, this meant that the same level differences held for most product ranges. Adjusted level means were compared two by two and a linear trend between them was tested using an ad-hoc contrast. 2.2

Application study

2.2.1 Administration and participants The application study was conducted in 8 sensory laboratories spread out in France (Agen, Caen, Dijon Lyon, Paris, Rennes, Strasbourg, Surgères). Within each laboratory, the sensory testing occurred over six sessions for most at weekly intervals. Each session was conducted at lunch time. In overall, it took place over a 6-month period. Participants were recruited either through the Nutrinet-Santé web-study (Hercberg et al., 2010) or were recruited by the laboratories. All participants gave written informed consent prior to taking part in the study. All procedures were approved by the Ile-de-France III Ethics Committee of the Tarnier-Cochin hospital (n° 2010-A00182-37) and by the French Data Protection Authority (Commission Nationale Informatique et Libertés, n° 1148039). Socio-demographic and anthropometric data were collected at the first testing session.

2.2.2 Food products The application study was conducted on the product ranges that were selected from the pretests. Each laboratory produced their own food products following the protocols developed in the supervising laboratory of Dijon and reviewed by the other laboratories. 2.2.3 Hedonic evaluation The protocol of hedonic evaluation and the sample portions that were served were as described above (cf. 2.1.2). The total amount consumed by each participant was approximately 500 g with 600 kcal which is equivalent to a small meal. For each subject, the presentation orders within each sensation was balanced according to a William Latin square design. 2.2.4 Statistical analysis 2.2.4.1 Predicted optimal level of fat, salt or sweet within a product range For each subject and each product, a quadratic regression was computed for the hedonic ratings (y) plotted against the level of fattiness, sweetness or saltiness (x) and the eigenvalue of this regression was determined. Based on this regression, the predicted optimal level (Lpref) of fat, sweetness or saltiness and the corresponding correlation coefficient (R) between the observed and predicted data values were derived as follows. A negative eigenvalue indicated that the regression curve was an inverted parabola. If the maximum of this inverted parabola occurred within the abscissa range (-2 to + 2), its x-coordinate corresponded to Lpref. If the maximum of the regression curve was expected to be outside of the abscissa range (> +2 or < -2), Lpref was assigned to the extreme value of the abscissa (+2 or -2, respectively) being the closer to the abscissa of the maximum of this inverted parabola. A regression eigenvalue equal to 0 meant that the regression curve was 'flat', showing that the subject did not have an optimal level. Lpref was thus considered as missing for this subject and product. A positive regression eigenvalue indicated that the curve was a non-inverted parabola with a minimum. If this minimum occurred within the abscissa range (-2 to + 2), this indicated that the subject preferred both extreme levels and Lpref was considered as aberrant and was thus missing for this subject and product. If the minimum of the regression curve was

expected to be outside of the abscissa range (> +2 or < -2), the predicted optimal level Lpref was assumed to take the opposite extreme value of the scale abscissa (-2 or +2, respectively). 2.2.4.2 Factorial analysis An exploratory factor analysis was run over each sensation on its compounding Lpref values to ensure there was only one underlying dimension (factor) per “sensation”. The analysis was performed using squared multiple correlations as prior communality estimates. Factors were extracted based on the maximum likelihood method. Cronbach’s alpha coefficient was then computed for each sensation to evaluate the internal consistency among food products. A Cronbach coefficient higher than 0.7 indicated a good internal consistency whereas a coefficient higher than 0.9 indicated redundancy (Hatcher, 1994; Nunnally, 1978). 2.2.4.3 Computation of overall liking scores For each subject, 3 liking scores (sweetness, saltiness and fat) were computed as the Rweighted average of the Lpref values of the corresponding food products. Some Lpref could be missing, because they could not be determined (as described above) or because of an untested product (e.g. a subject not eating a particular product for personal reasons). If the Lpref values of a subject were missing for more than 50-60% of the food products within a sensation, i.e. for more than 5 products for the fatty or the saltiness and 7 products for the sweetness, the corresponding liking score was not calculated and was considered as missing for this subject. 2.2.4.4 Normality of the score distribution Normality of the score distributions was assessed based on the Skewness and Kurtosis indices. Strong violation of normality was indicated by coefficients higher than 2 for Skewness and 7 for Kurtosis in absolute value (Curran, West, & Finch, 1996; Kline, 2005). 2.2.4.5 Analysis of variance For each product range, hedonic ratings were submitted to a two-way ANOVA with subject and fatty, salty or sweet level as factors. Then over each of the 3 sensations (fatty, salty and sweet),

a unique ANOVA was run mixing all product ranges according to the following model: hedonic ratings = subject + level + product(level). In this model, level was tested against product nested within level. If level was significant, this meant that the same level differences held for most product ranges. Adjusted level means were compared two by two. Overall liking scores were submitted to an ANOVA to assess whether there was an effect of the test location (8 different cities). The model was adjusted for the participant characteristics (gender, age class, level of education, weight status). 3 3.1

Results Development study

3.1.1 Participants Overall, 341 subjects participated in pre-testing. There were 46% male and 54% female of ages between 18 and 73 years (mean 45.2 years, SD 13.3). 3.1.2 Product selection There were 69 different food products that were tested including 22 products for fat, 20 for saltiness and 27 for sweetness as detailed in Table 1. For each food product, several ranges of the 5 nutrient levels were tested resulting in a total of 144 food ranges experienced. Based on the technical feasibility of each food product and on the distribution of the hedonic ratings per nutrient level that should approach a normal distribution, food products and ranges were selected resulting in a total of 32 product ranges including 10 products for fat, 10 for saltiness and 12 for sweetness. The numbers of products that were cold, temperate and hot were respectively for fat 2, 3 and 5, for saltiness 1, 2 and 7 and for sweetness 4, 6 and 2. The contents of added lipids, sodium chloride or sugars for each level of each food product are given in Table 2. 3.1.3 Perceived intensity of the food products according to the nutrient level The sensory profile was conducted on the 32 selected products. Over each “sensation”, the nutrient level impacted significantly on the perceived intensity of the corresponding sensation over all the compounding foods (Figure 1). Similarly, for each product, there was a significant effect of

the nutrient level on the fatty, salty or sweet perception (Table 3). The perceived intensity of fat, saltiness or sweetness increased significantly in a linear manner with the nutrient level for each product (Table 3). An exception was for the fat product zucchini with white sauce, for which there was no significant difference of fat perception among the 5 levels (p = 0.63) with values ranging from 3.4 to 3.9 from L-2 to L+2. Thus, this last product was not selected for the application study. Over all the products compounding the sensation of fatty, sweet or salty, the perceived intensity also increased significantly in a linear manner with the nutrient level, except for the levels of fat -2 and -1 that were not significantly different (Figure 1). 3.2

Application study

3.2.1 Participants There were in total 567 participants scattered into the 8 sensory laboratories by groups of 32 to 146. There were 56% of women and 44% of men, the mean age was 49.4 years (SD 14.4; 18-81 years range) and the mean body mass index was 24.7 kg/m² (SD 4.5). Data from one participant were removed because of a large number of missing or aberrational data. 3.2.2 Estimation of the predicted optimal level (Lpref) per product For each product, there was a significant effect (p < 0.0001) of the lipids, sodium chloride or sucrose level contents on the hedonic ratings (Table 4). The mean hedonic ratings for each level of fat, saltiness or sweetness followed an inverted U-shape for all the products, as shown by the eigenvalues of the quadratic regression (Table 4) and as visualized on Figure 1. For each sensation, there was a significant effect of the fatty, salty or sweet level on the mean hedonic ratings and the latter significantly differed among each fatty, salty or sweet level (Figure 1). Lpref could be determined for most products as the liking-level curves concurred with the expected inverted U-shape parabola (67% of the products for fatty sensation, 78% for sweet, 84% for salty) or in the minority the curves were non-inverted U-shape parabola with the minimum outside of the abscissa range (10% of the products for fatty sensation, 9% for sweet, 6% for salty). For the complementary products, Lpref could not be determined as the liking-level curve either

followed a non-inverted U-shape parabola with its minimum within the abscissa range or was flat. The proportion of subjects for each shape of curves is presented in Table 5. The medians of the number of products actually computable by subject were 7 out of 9, 9 out of 10 and 11 out of 12 for fat, salt and sweet respectively. Consequently, overall liking scores could be determined for virtually all subjects (96% of subjects for fatty, 98% for salty and 99% for sweet). 3.2.3 Overall liking scores: internal validity and distribution According to the exploratory factor analysis conducted on the Lpref of the compounding foods, a single factor solution appeared as the most satisfactory solution for each sensation. All of the food products had a salient loading (> 0.3; Table 6) within each sensation. Soft white cheese had a loading close to salient (0.28) and was thus kept for the score computation. Cronbach’s alpha ranged between 0.8 and 0.9 for salty and sweet, and reached 0.6 for fatty. The liking score distributions (Figure 2) showed that the scores fell into the normative ranges as predicted by the hypothesis. Some of the participants preferred the extreme levels, some the mid ranges (-1, +1) and most the mean level (0). This is confirmed by the Skewness and Kurtosis indices that were close to 0, thus indicating a distribution close to normality (Table 7). There was no effect of the test location on the overall liking scores for fatty sensation (p = 0.26), salty (p = 0.12) and sweet (p = 0.69).

4

Discussion The present study demonstrated a good feasibility and internal validity of the PrefSens

sensory tool developed to measure the overall liking scores for the salty, sweet and fatty sensations. This approach provided an overall liking score based on the hedonic evaluation of 9 to 12 foods per sensation, each food being tasted and rated with 5 grading levels of fatty, salty or sweet sensations. This study demonstrated that it was feasible to combine hedonic data for different food matrices derived from the optimal levels (-2,-1, 0, 1, 2) of the perceived sensation. This tool allowed the

subjects to be discriminated according to their liking with a distribution close to normality. To date, such complete sensory tool has never been developed. 4.1

Feasibility and reproducibility

Food products were chosen in order to be the most representative of those habitually consumed for the sensations of interest (fatty, salty or sweet) by the French population. Despite the large number of possibilities, technical constraints restricted our choice. In particular, it was necessary to use ingredients available anywhere in France and, as much as possible, with a similar composition independently of its purchase location and of the season. In addition, food preparation protocols had to be easily reproduced in the other collaborating laboratories. Some products did not fit these criteria, in particular meats which can present large inter-individual variation in their nutritional content. In addition, hedonic data for such product can be biased by a cooking time that is not corresponding to the subject habits. Meats therefore were not used. To minimize this problem with fish, canned fish was employed. Regarding vegetables, variability between products was limited by using frozen vegetables from a national brand to provide to the consumer a stable product along the year. The food production and the hedonic tests were conducted in 8 different laboratories spread out in the country. There was no apparent problem of food production. It was not possible to strictly evaluate the reproducibility among laboratories as subjects differed. Nevertheless the overall liking scores of fatty, salty or sweet sensations did not significantly differ among laboratories, suggesting that population sampling was similar across laboratories in terms of sensory characteristics. 4.2

Selection of food products and of their levels of added lipids, sodium chloride or sucrose

A large number of products were tested and selected based on their technical feasibility and their potential to discriminate subjects for their hedonic ratings. We hypothesized that for each food product, an inverted U-shape parabola should be found between the average hedonic ratings and the corresponding levels of added lipids, sodium chloride or sucrose (Pangborn, 1970; Pangborn, 1988; Stone & Pangborn, 1990). For some products, there was no liking discrimination regardless of the

nutrient level. For instance, chocolate mousse was highly liked regardless of the sucrose content and thus was not selected. Other products, for instance tomato juice, were also well appreciated even without compound addition and thus could not be selected. For each food product, the quantities of added nutrients (lipids, sodium chloride or sucrose) were adjusted in order to approach a normal distribution of hedonic ratings centered on the level 0. Over a sensation, a perceived level of the sensation of fat, salt or sweet could refer to a different quantity of added nutrients. For instance, the fat perception was not significantly different between leeks and cake (data not shown) at level L2 whereas there was 13% of added lipids in leeks and 37% in cake. Similarly, the sweet perception was not significantly different between tea and cake at level L0 whereas there was 5% of added sucrose in tea and 18% in cake. On the contrary, for a similar content of added sodium chloride in carrot puree and zucchini + white sauce (1.5-1.6%), the salty perception was significantly higher in zucchini + white sauce (6.8 out of 10) than in carrot puree (5.6) (data not shown). These results highlight the importance of the food matrix on the perception. For this reason, the different contents of the nutrients were grouped into levels that were significantly different from each other in terms of perception, such as undertaken by Mela (1990). Regarding the nutrients, we used sodium chloride and sucrose for the salty and sweet tastes as they are the reference molecules recognized by their corresponding taste-receptor cells on the tongue (Chandrashekar, Hoon, Ryba, & Zuker, 2006; Lindemann, 2001). As a high concentration of sucrose impacted on the texture of baked products such as cakes, a mixture of sucrose and nonnutritive sweetener were used for the L+1 and L+2 level in this food model. The non-nutritive sweetener was added in a limited quantity to minimize the introduction of off-tastes potentially brought by the non-nutritive sweetener. Unlike for the sweet and salty tastes, the stimuli as well as the sensations elicited by fats are complex (Mela & Marshall, 1991). Fat is perceived mainly through textural cues (1990) (Rolls, 2008; Verhagen, Kadohisa, & Rolls, 2004) but also possibly through gustatory or olfactory cues (Khan & Besnard, 2008; Laugerette et al., 2005; Schiffman, Graham, Sattely-Miller, & Warwick, 1998). Fat may arise from a wide range of ingredients from

animal origin (cream, butter, lard) or vegetal origin (oil, margarine) and each of these ingredients elicits different sensations (Mela & Marshall, 1991; Running, 2011) because of different physicochemical properties resulting in different fat structure and thus in different textural cues, but also because of different associated gustatory sensations (Hayes & Keast, 2011). In order to represent the variety of fat, we used as vegetal source sunflower oil or mayonnaise in 3 products out of 10, and as animal source dairy fat (whipped cream) in 4 products and pork lard in 3 products. These ingredients were used as much as possible in their classical recipes. 4.3

Internal validity and distribution of the overall liking score

Based on the exploratory factorial analysis, we could demonstrate that the overall liking score for a “sensation” was unidimensional. Cronbach’s alpha confirmed that there was a good internal consistency among items of the sensation, especially for salty and sweet sensations, and to a lesser extent for fat. This lower degree of consistency for fat must be due to the complexity of the sensation and the various ingredients for fat. To our knowledge, no previous study has checked the internal validity of their sensory tool, as undertaken here. The distribution of the overall liking score of each sensation was close to normality, as shown by the Skewness and Kurtosis indices, which were lower than 2 and 7 respectively (Curran et al., 1996; Kline, 2005). 4.4

Limits of the tool

Although PrefSens has been developed for a French population, the selected foods were not specific to the French food culture and seem to be adapted for other western countries. Nevertheless, the use of pork lard may be a limitation for people from a culture prohibiting the consumption of pork. This may need to be adapted depending on the population sample. In addition, the present range levels of added lipids, sodium chloride or sucrose may not match the optimal levels of subjects from other food cultures. These levels may need to be adapted to recover a distribution of the liking-level curve close to normality. To summarize, PrefSens could be used in western countries with slight adaptations, but would require substantial ones with other types of food cultures, such as the Asiatic

one. PrefSens included a lower number of fatty-sweet foods compared to fatty-salty foods, as it was difficult to select more fatty-sweet foods. However, in a usual meal, we encounter more fatty-salty foods as components of the main meal and eventually of the starter and the cheese plate, while the fatty-sweet foods are eaten only for desserts. This is supported by the lower dietary intake of categories of foods that are fatty-sweet compared to those that are fatty-salty (Lafay, 2009). The main constraint of PrefSens is the heavy work load required for sample preparation and distribution, which results in a rather costly and time-consuming tool. The authors investigated the extent to which it was possible to reduce the number of food products per sensation. Although methodology and full results will be proposed for publication later, the bottom line of our findings was that 4 out of 12 for sweet (Apple sauce, Strawberry syrup, Cake, Chocolate custard), and 4 out of 10 for salty sensations (Mashed potatoes, Zucchini + white sauce, Pasta + Bolognese, Carrot puree) were enough to reproduce reasonably well the full tests. It was not possible to reduce the

number of 10 fatty food products without a substantial loss of information. 4.5

Perspectives

PrefSens could be used in nutritional studies for contrasting the liking distribution from a specific population (e.g. diabetic patients, elderly) to that of the general population (Figure 2). A food industry could easily compare actual sucrose, sodium chloride and lipids contents of their products to the PrefSens recipes (Annex 1) in which the L0 level would be the most appreciated in average. A service provider in sensory analysis should be interested in administering the PrefSens to the members of its panels for being able later to balance their sampling over their preference towards sweet, salty and fatty sensations. As the PrefSens sensory tool cannot be used when the studied population is too large, a questionnaire (PrefQuest) was also developed in the framework of the EpiPref larger project mentioned in introduction in order to measure liking for sweet, salty or fatty sensations with no tasting (Deglaire et al., 2012). This questionnaire has been validated internally. However, PrefQuest

collects declared liking which are expected to be influenced by external factors other than those influencing the direct sensory liking as measured in PrefSens, i.e. based on actual food tasting. A comparison between PrefQuest and PrefSens will be the subject of another publication.

5

Conclusion

PrefSens is an original tool that will be useful to better understand the determinants of dietary behaviours. To this aim, PrefSens can be applied to a population sample (French or western countries) for which the dietary intakes are well characterised. PrefSens can also be used for a follow-up of a cohort in order to evaluate whether there is a causal relationship between liking and dietary intakes, and indirectly with body mass indices. PrefSens has been carefully developed and grouped a variety of foods in order to be representative of the overall liking. Internal consistency of the overall liking score has been demonstrated for each sensation of fatty, sweet or salty and this has never been undertaken in other studies. Overall, PrefSens is a good basis for researchers from other countries to measure likings for salty, sweet and/or fatty sensations by means of a sensory tool.

Acknowledgements This study was supported by the French National Research Agency (Agence Nationale de la Recherche) in the context of the 2008 Programme de Recherche ‘Alimentation et Industries Alimentaires’ (ANR-08-ALIA-006), by Centre d'Etudes et de Documentation du Sucre (CEDUS) and the competitive cluster VITAGORA. We thank all the scientists who helped to carry out the Nutrinet-Santé study and Christophe Martin who trained the sensory panel.

References Beauchamp, G. K., Bertino, M., Burke, D., & Engelman, K. (1990). Experimental sodium depletion and salt taste in normal human volunteers. American Journal of Clinical Nutrition, 51(5), 881-889. Bertino, M., Beauchamp, G. K., & Jen, K. C. (1983). Rated taste perception in two cultural groups. Chemical Senses, 8(1), 3-15. Blundell, J. E., & Finlayson, G. (2004). Is susceptibility to weight gain characterized by homeostatic or hedonic risk factors for overconsumption? Physiology & Behavior, 82(1), 2125. Bobowski, N., Rendahl, A., Vickers, Z. (2015). A longitudinal comparison of two salt reduction strategies: Acceptability of a low sodium food depends on the consumer. Food Quality and Preference, 40(B), 270-278. Bowen, D., Green, P., Vizenor, N., Vu, C., Kreuter, P., & Rolls, B. (2003). Effects of fat content on fat hedonics: cognition or taste? Physiology & Behavior, 78(2), 247-253. Chandrashekar, J., Hoon, M. A., Ryba, N. J., & Zuker, C. S. (2006). The receptors and cells for mammalian taste. Nature, 444(7117), 288-294. Cornwell, T. B., & McAlister, A. R. (2011). Alternative thinking about starting points of obesity. Development of child taste preferences. Appetite, 56(2), 428-439. Curran, P. J., West, S. G., & Finch, J. F. (1996). The robustness of test statistics to nonnormality and specification error in confirmatory factor analysis. Psychological Methods, 1(1), 16-29. Deglaire, A., Méjean, C., Castetbon, K., Kesse-Guyot, E., Urbano, C., Hercberg, S., et al. (2012). Development of a questionnaire to assay recalled liking for salt, sweet and fat. Food Quality and Preference, 23, 110-124. Drewnowski, A., Shrager, E. E., Lipsky, C., Stellar, E., Greenwood, M.R.C. (1989). Sugar and fat: Sensory and hedonic evaluation of liquid and solid foods. Physiology & Behavior, 45(1), 177-183. Drewnowski, A., Schwartz, M. (1990). Invisible fats: Sensory assessment of sugar/fat mixtures. Appetite, 14(3), 203-217. Engell, D., Bordi, P., Borja, M., Lambert, C., & Rolls, B. (1998). Effects of information about fat content on food preferences in pre-adolescent children. Appetite, 30(3), 269-282. Hatcher, L. (1994). A step-by-step approach to using SAS® for factor analysis and structural equation modeling. Cary, NC: SAS Institute Inc. Hayes, J. E., & Duffy, V. B. (2008). Oral sensory phenotype identifies level of sugar and fat required for maximal liking. Physiology & Behavior, 95(1-2), 77-87. Hayes, J. E., Keast R. S. J. (2011). Two decades of supertasting: Where do we stand? Physiology & Behavior, 104(5), 1072-1074.Hayes, J. E., Sullivan, B. S., Duffy V. B. (2010). Explaining variability in sodium intake through oral sensory phenotype, salt sensation and liking. Physiology & Behavior, 100(4), 369-380. Hayes, J. E., Keast, R. S. J. (2011). Two decades of supertasting: Where do we stand? Physiology & Behavior, 104, 1072–1074. Hercberg, S., Chat-Yung, S., & Chauliac, M. (2008). The French National Nutrition and Health Program: 2001-2006-2010. International Journal of Public Health, 53(2), 68-77. Hercberg, S., Castetbon, K., Czernichow, S., Malon, A., Mejean, C., Kesse, E., et al. (2010). The Nutrinet-Santé 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(242), 242. Kanarek, R. B., Ryu, M., & Przypek, J. (1995). Preferences for foods with varying levels of salt and fat differ as a function of dietary restraint and exercise but not menstrual cycle. Physiology & Behavior, 57(5), 821-826. Kim, G. H., & Lee, H. M. (2009). Frequent consumption of certain fast foods may be associated with an enhanced preference for salt taste. Journal of Human Nutrition and Dietetics, 22(5),

475-480. Kline, R. B. (2005). Principles and practice of structural equation modeling (2nd ed.). New York: The Guilford Press. Lafay, L. (2009). Étude Individuelle Nationale des Consommations Alimentaires 2 (INCA 2) 20062007: Agence Française de Sécurité Sanitaire des AlimentsLindemann, B. (2001). Receptors and transduction in taste. Nature, 413(6852), 219-225. Lucas, L., Riddell, L., Liem, G., Whitelock, S., & Keast, R. (2011). The influence of sodium on liking and consumption of salty foods. Journal of Food science, 76(1), 572-576. Mela, D. J. (1990). Sensory preferences for fats: what, who, why? Food Quality and Preference, 2(2), 95-101. Mela, D. J., & Marshall, R. J. (1991). Sensory properties and perceptions of fats. In D. J. Mela (Ed.), Dietary fats: Determinants of preference, selection and consumption (pp. 27-41). New York: Elsevier Applied Science. Monneuse, M. O., Bellisle, F., & Louis-Sylvestre, J. (1991). Impact of sex and age on sensory evaluation of sugar and fat in dairy products. Physiology & Behavior, 50(6), 1111-1117. Nunnally, J. C. (1978). Psychometric theory (2nd ed.). New York: McGraw-Hill. Pangborn, R. M. (1970). Individual variation in affective response to taste stimuli. Psychonomic Science, 21, 125-126. Pangborn, R. M. (1988). Sensory attributes and acceptance of fat, sugar, and salt in dairy products. In D. M. H. Thomson (Ed.), Food acceptability (pp. 413-429). London: Elsevier Applied Science. Rolls, E. T. (2008). Functions of the orbitofrontal and pregenual cingulate cortex in taste, olfaction, appetite and emotion. Acta Physiologica Hungarica, 95(2), 131-164. Running, C. A., Mattes, R. D., Tucker, R. M. (2013). Fat taste in humans: Sources of within- and between-subject variability. Progress in Lipid Research 52(4), 438-445.Stone, L. J., & Pangborn, R. M. (1990). Preferences and intake measures of salt and sugar, and their relation to personality traits. Appetite, 15(1), 63-79. Verhagen, J. V., Kadohisa, M., & Rolls, E. T. (2004). Primate insular/opercular taste cortex: neuronal representations of the viscosity, fat texture, grittiness, temperature, and taste of foods. Journal of Neurophysiology, 92(3), 1685-1699.

Figure 1. Perceived intensity1 and hedonic ratings2 (mean ± overall SE) of each level of fat (A), salt (B) and sweet (C) as averaged over the compounding products (9 for fat, 10 for salt and 12 for sweet) and subjects3.

1

averaged over 12 trained panelists. The possible minimum and maximum intensity that any subject could give was 0 and 10, respectively 2 averaged over 567 subjects. The possible minimum and maximum hedonic ratings that any subject could give was 1 and 9, respectively 3 Different letters indicate that means are significantly different (P<0.01) within the perceived intensity ratings (letters under the plain line) or within the hedonic ratings (letters above the dashed line), respectively.

8

liking

7

intensity

Mean rating

6 5 4

3 2 1 0

-2

-1

0

1

2

Level of fat 8

liking

7

intensity

Mean rating

6 5 4

3 2 1 0

-2

-1

0

1

2

Level of salt 8

liking

7

intensity

Mean rating

6 5 4

3 2 1 0

-2

-1

0

Level of sweet

1

2

d

Figure 2. Distribution of the overall liking scores for fat (A) , salt (B) and sweet (C) as determined on 543 to 563 consumers.

Fat

A

Salt

B

Overall liking scores (percent of subjects)

Sweet

C

Table 1. Food products tested during the development of the tests.

Fat

Salt

Sweet

(22 products)

(20 products)

(27 products)

Fat-and-salt Crisps

Broccoli terrine

Apple sauce

Leeks

Butter

Cake

Lentil puree

Carrot puree

Chocolate cake

Mashed potatoes

Cream cheese

Chocolate custard

Mixed vegetables

Cucumber soup

Chocolate milk

Pasta + butter

Green beans

Chocolate mousse

Polenta

Ham

Coconut macaroon

Rillettes

Mashed potatoes

Corn flakes + milk

Soft white cheese

Mayonnaise

Crème brûlée

Vegetable soup

Mustard and herbs stick

Desert roses

Tapenade

Pasta + Bolognese

Elephant ears

Tuna + mayo

Polenta

Far breton

Tuna + oil

Salmon terrine

Ketchup

Zucchini + white sauce

Salty stick

Lemonade

Fat-and-sweet

Semolina

Milk shake

Almond cake

Soft white cheese

Orange juice

Cake

Spinach cake

Soft white cheese

Chocolate mousse

Tomato juice

Sparkling water

Chocolate paste

Vegetable soup

Stick

Chocolate milk

Zucchini + white sauce

Strawberry syrup

Petits suisses

Lemon Tea

Shortbread biscuits

Tea

Yogurt

Tea jelly Vanilla ice cream Verbena infusion Whipped cream Yogurt

Table 2. Percentage (g per 100 g of food) of added lipids, salt and sugar (+ sweetener) within the selected food products per tastant level.

Fat (% added lipids) Leeks (cream) Tuna + mayonnaise Mixed vegetables (mayonnaise) Polenta (pork lard) Almond cake (butter) Cake (oil) Lentil puree (pork lard) Mashed potatoes (pork lard) Soft white cheese (cream) Zucchini + white sauce (cream) Salt ( % of added salt) Green beans Carrot puree Broccoli terrine Mashed potatoes Zucchini + white sauce Pasta + Bolognese Spinach cake Polenta Vegetable soup Salmon terrine

L-2

L-1

Level L0

1 5 3 2 3 9 2 1 0 0

3 10 6 5 6 16 5 2 3 1

5 17 12 9 17 21 9 9 7 2

8 27 21 13 24 27 17 17 10 4

13 35 27 17 29 37 29 29 17 7

0 0 0 0 0 0 0 0.1 0 0

0.5 0.2 0.1 0.2 0.3 0.2 0.2 0.5 0.1 0.1

1 0.7 0.4 0.5 0.7 0.6 1.2 1 0.3 0.4

2 1.1 0.8 1 1.1 1.7 1.6 1.5 0.6 0.8

3.9 1.5 1.6 1.5 1.6 2.5 2.6 3.6 1.2 1.5

5 6.5 8 6 5 10 12 + 0 5 2.5 9 18 + 0.9 2

11 15 14 11 11 26 17 + 0.6 9 8 16 18 + 4 5

20 22 25 20 17 42 21 + 6 17 14 29 17 + 8 9

Sweet (% of added sugar + % of added sweetener) Chocolate milk 0.5 2 Soft white cheese 0.4 2 Corn flakes + milk 0 2 Milk shake 1 3 Tea 1 3 Whipped cream 0 3 Chocolate custard 3+0 7+0 Verbena infusion 1 2 Strawberry syrup 0 1 Apple sauce 0.5 2 Cake 9+0 17 + 0 Orange juice 0 0.5

L+1

L+2

Table 3. Mean perceived intensity1 of fat, salt or sweet in food products according to the grading levels (L-2 to L+2) of fat, salt or sweet, respectively, as determined by a trained panel. Level

Sensation and

Number

P-value for

P-value for

food product

of judges

L-2

L-1

L0

L+1

L+2

LSD

effect level2

linear trend

Leeks

10

1.1 (d)

1.3 (dc)

1.8 (c)

2.4 (b)

3.0 (a)

0.5688

<0.0001

<0.0001

Tuna + mayo

12

1.2 (c)

1.8 (c)

2.8 (b)

3.2 (b)

4.1 (a)

0.6068

<0.0001

<0.0001

Mixed vegetables

11

1.9 (b)

2.1 (b)

2.4 (b)

3.9 (a)

4.1 (a)

0.8917

<0.0001

<0.0001

Polenta

11

2.2 (b)

2.4 (b)

3.1 (ab)

2.6 (b)

3.8 (a)

1.1092

0.0414

0.008

Almond cake

10

2.8 (b)

2.9 (b)

3.9 (a)

4.4 (a)

4.4 (a)

0.9005

0.0007

<0.0001

Cake

11

1.5 (c)

2.0 (cb)

2.2 (b)

2.9 (a)

3.2 (a)

0.6403

<0.0001

<0.0001

Lentil puree

12

2.2 (cd)

2.1 (d)

3.1 (cb)

4.1 (b)

5.5 (a)

0.9425

<0.0001

<0.0001

Mashed potatoes

11

2.3 (c)

2.1 (c)

2.4 (c)

3.9 (b)

5.0 (a)

0.9731

<0.0001

<0.0001

Soft white cheese

12

2.5 (c)

2.7 (cb)

4.0 (a)

3.7 (ab)

3.6 (ab)

0.9764

0.0124

0.005

Zucchini + white sauce

12

3.4 (a)

3.8 (a)

3.8 (a)

4.1 (a)

3.9 (a)

0.9316

0.6377

0.221

Green beans

11

0.5 (d)

1.0 (d)

2.9 (c)

5.2 (b)

7.2 (a)

0.6703

<0.0001

<0.0001

Carrot puree

12

0.8 (e)

1.7 (d)

3.2 (c)

4.6 (b)

5.6 (a)

0.7231

<0.0001

<0.0001

Fat

Salt

Broccoli terrine

11

0.6 (c)

1.5 (dc)

2.0 (c)

3.9 (b)

5.9 (a)

0.8385

<0.0001

<0.0001

Mashed potatoes

10

0.7 (d)

1.7 (cd)

2.5 (c)

4.6 (b)

5.8 (a)

0.8675

<0.0001

<0.0001

Zucchini + white sauce

12

0.8 (e)

1.8 (d)

3.3 (c)

5.3 (b)

6.8 (a)

0.8557

<0.0001

<0.0001

Pasta + Bolognese

12

1.5 (e)

2.3 (d)

3.4 (c)

5.1 (b)

5.8 (a)

0.6492

<0.0001

<0.0001

Spinach cake

12

1.2 (d)

2.1 (d)

4.4 (c)

5.4 (b)

6.5 (a)

1.0184

<0.0001

<0.0001

Polenta

12

0.5 (d)

1.0 (d)

3.9 (c)

5.2 (b)

7.6 (a)

0.7652

<0.0001

<0.0001

Vegetable soup

11

1.2 (d)

1.5 (cd)

2.2 (c)

4.5 (b)

5.9 (a)

0.9111

<0.0001

<0.0001

Salmon terrine

12

2.4 (d)

2.3 (d)

3.4 (c)

4.7 (b)

5.8 (a)

0.6231

<0.0001

<0.0001

Sweet

1

Chocolate milk

11

0.8 (c)

2.7 (b)

3.7 (b)

6.0 (a)

6.1 (a)

1.3611

<0.0001

<0.0001

Soft white cheese

12

0.3 (e)

1.3 (d)

3.6 (c)

6.1 (b)

7.3 (a)

0.6176

<0.0001

<0.0001

Corn flakes + milk

11

1.6 (c)

2.2 (c)

4.4 (b)

5.9 (a)

6.3 (a)

0.9044

<0.0001

<0.0001

Milk shake

10

0.9 (e)

2.5 (d)

4.5 (c)

6.4 (b)

7.3 (a)

0.6381

<0.0001

<0.0001

Tea

12

0.3 (e)

1.8 (d)

3.3 (c)

6.7 (b)

7.7 (a)

0.7104

<0.0001

<0.0001

Whipped cream

10

0.6 (d)

2.2 (c)

4.8 (b)

6.7 (a)

7.3 (a)

0.7830

<0.0001

<0.0001

Chocolate custard

12

1.2 (e)

2.4 (d)

4.4 (c)

6.5 (b)

7.6 (a)

0.8452

<0.0001

<0.0001

Verbena infusion

11

1.3 (d)

2.0 (d)

4.1 (c)

6.2 (b)

7.3 (a)

1.0049

<0.0001

<0.0001

Strawberry syrup

12

1.2 (e)

2.0 (d)

2.9 (c)

6.2 (b)

7.3 (a)

0.4962

<0.0001

<0.0001

Apple sauce

11

3.1 (e)

3.9 (d)

5.5 (c)

6.5 (b)

7.1 (a)

0.6039

<0.0001

<0.0001

Cake

12

2.0 (c)

2.8 (cb)

3.5 (b)

5.3 (a)

5.4 (a)

0.9153

<0.0001

<0.0001

Orange juice

12

2.2 (d)

2.3 (d)

3.0 (c)

4.2 (b)

5.9 (a)

0.6877

<0.0001

<0.0001

Perceived intensity per product and level averaged over the subjects. The possible minimum and maximum intensity that any subject could give was 0

and 10, respectively 2

Based on an anova model with level and subject as factors

3

LSD : Least Significant Difference.

Table 4. Mean hedonic ratings1 of food products according to the grading levels (L-2 to L+2) of fat, salt or sweet, respectively, as determined by the subjects. Level

Sensation and

Number of

food product

subjects

L-2

L-1

L0

L+1

L+2

effect level

Leeks

564

4.5 (c)

4.8 (b)

5.1 (a)

5.3 (a)

4.8 (b)

<0.0001

Tuna + mayo

563

4.6 (c)

5.2 (b)

6.2 (a)

5.4 (b)

4.1 (d)

<0.0001

Mixed vegetables

561

4.6 (e)

5.2 (c)

5.9 (a)

5.5 (b)

4.8 (d)

<0.0001

Polenta

561

4.2 (c)

4.7 (ba)

4.9 (a)

4.7 (b)

4.2 (c)

<0.0001

Almond cake

563

5.9 (b)

6.1 (ba)

6.3 (a)

5.9 (b)

5.0 (c)

<0.0001

Cake

565

4.6 (d)

5.2 (b)

5.5 (a)

5.7 (a)

4.9 (c)

<0.0001

Lentil puree

559

5.7 (a)

5.8 (a)

5.7 (a)

4.3 (b)

3.0 (c)

<0.0001

Mashed potatoes

561

6.4 (ba)

6.6 (a)

6.3 (b)

4.9 (c)

3.2 (d)

<0.0001

Soft white cheese

562

5.7 (b)

6.1 (a)

6.2 (a)

6.1 (a)

5.4 (c)

<0.0001

Carrot puree

563

4.2 (c)

5.4 (b)

6.1 (a)

5.3 (b)

3.8 (d)

<0.0001

Broccoli terrine

566

3.2 (c)

4.0 (b)

5.3 (a)

5.1 (a)

3.2 (c)

<0.0001

Bolognese sauce

562

3.5 (c)

5.0 (b)

6.0 (a)

4.8 (b)

3.1 (d)

<0.0001

Green beans

564

3.7 (c)

5.0 (b)

5.6 (a)

5.2 (b)

2.9 (d)

<0.0001

Pvalue for

Fat

Salt

Zucchini + white sauce

565

3.3 (c)

4.5 (b)

5.2 (a)

4.3 (b)

2.7 (d)

<0.0001

Mashed potatoes

545

4.0 (d)

5.4 (b)

6.1 (a)

4.9 (c)

3.0 (e)

<0.0001

Spinach cake

564

3.8 (c)

4.9 (a)

4.4 (b)

3.6 (c)

2.6 (d)

<0.0001

Polenta

541

3.1 (c)

4.2 (b)

4.7 (a)

4.1 (b)

1.8 (d)

<0.0001

Vegetable soup

566

3.8 (c)

4.7 (b)

5.4 (a)

5.6 (a)

3.5 (d)

<0.0001

Salmon terrine

565

5.4 (b)

5.8 (a)

5.8 (a)

4.7 (c)

3.0 (d)

<0.0001

Chocolate milk

558

3.0 (d)

4.2 (c)

5.3 (a)

4.6 (b)

3.2 (d)

<0.0001

Soft white cheese

563

3.7 (c)

4.3 (b)

5.4 (a)

4.5 (b)

3.5 (c)

<0.0001

Corn flakes + milk

553

4.2 (d)

5.2 (b)

5.8 (a)

4.6 (c)

3.5 (e)

<0.0001

Milk shake

561

3.4 (d)

4.3 (c)

5.3 (a)

4.8 (b)

3.4 (d)

<0.0001

Tea

562

4.4 (b)

4.8 (a)

4.9 (a)

4.0 (c)

3.2 (d)

<0.0001

Whipped cream

562

3.2 (d)

5.2 (b)

6.3 (a)

4.3 (c)

3.2 (d)

<0.0001

Chocolate custard

564

3.7 (c)

4.8 (b)

6.0 (a)

4.9 (b)

3.1 (d)

<0.0001

Verbena infusion

565

4.7 (b)

5.1 (a)

5.2 (a)

4.2 (c)

3.1 (d)

<0.0001

Strawberry syrup

565

3.4 (c)

4.2 (b)

4.8 (a)

4.1 (b)

3.2 (c)

<0.0001

Apple sauce

566

4.9 (d)

5.7 (b)

6.2 (a)

5.3 (c)

3.9 (e)

<0.0001

Cake

564

3.7 (e)

5.3 (c)

6.0 (a)

5.7 (b)

4.8 (d)

<0.0001

Orange juice

563

4.8 (c)

5.0 (cb)

5.4 (a)

5.1 (b)

4.1 (d)

<0.0001

Sweet

1

Hedonic ratings for each product and at each level averaged over the subjects. The possible minimum and maximum ratings that any subject could

give was 1 and 9, respectively 2

Based on an anova model with level and subject as factors.

Table 5. Distribution of the shape of the predicted liking-level curves for fatty, salty and sweet.

%

Shape of the curve Fatty

Salty

Sweet

67.1

84.2

78.1

outside of the curvea

10.0

6.1

9.3

U-shapeb

7.6

4.3

5.1

Flatb

15.3

5.4

7.5

Inverted U-shapea U-shape with Lpref

a

The predicted preferred level could be determined.

b

The predicted preferred level could not be determined.

33

Table 6 Exploratory factor analyses1 performed on the predicted preferred level of fat, salt or sweet in the food products compounding the fat, salt or sweet sensation: parameter estimates and internal consistency (Cronbach’s alpha). Sensation and

Standardized

Cronbach’s

food products

factor loading2

alpha

Fat Leeks

0.533

Tuna + mayo

0.510

Mixed vegetables

0.504

Polenta

0.428

Almond cake

0.398

Cake

0.396

Lentil puree

0.393

Mashed potatoes

0.326

Soft white cheese

0.278

0.596

Salt

34

Green beans

0.649

Carrot puree

0.648

Broccoli terrine

0.608

Mashed potatoes

0.597

Zucchini + white sauce

0.596

Pasta + Bolognese

0.584

Spinach cake

0.536

Polenta

0.529

Vegetable soup

0.519

0.826

Salmon terrine

0.484

Sweet Chocolate milk

0.751

Soft white cheese

0.689

Corn flakes + milk

0.669

Milk shake

0.669

Tea

0.655

Whipped cream

0.642

Chocolate custard

0.636

Verbena infusion

0.629

Strawberry syrup

0.610

Apple sauce

0.579

Cake

0.578

Orange juice

0.531

1

Maximum likelihood estimation.

2

Values statistically significant at p < 0.001.

35

0.883

Table 7. Skewness and kurtosis coefficients of the distributions of the predicted preferred levels of fat, salt or sweet1.

Number of Skewness

Kurtosis

subjects

1

Fat

543

0.31

0.06

Salt

559

0.22

-0.67

Sweet

563

0.50

0.41

Strong violation of normality would be indicated by coefficients higher than 2 for Skewness and 7

for Kurtosis in absolute value.

36

Table 2. Percentage (g per 100 g of food) of added lipids, salt and sugar (+ sweetener) within the selected food products per tastant level. L-2

L-1

Level L0

Leeks (cream)

1

3

5

8

13

Tuna + mayonnaise

5

10

17

27

35

Mixed vegetables (mayonnaise)

3

6

12

21

27

Polenta (pork lard)

2

5

9

13

17

Almond cake (butter)

3

6

17

24

29

Cake (oil)

9

16

21

27

37

Lentil puree (pork lard)

2

5

9

17

29

Mashed potatoes (pork lard)

1

2

9

17

29

Soft white cheese (cream)

0

3

7

10

17

Zucchini + white sauce (cream)

0

1

2

4

7

0

0.5

1

2

3.9

Carrot puree

0

0.2

0.7

1.1

1.5

Broccoli terrine

0

0.1

0.4

0.8

1.6

Mashed potatoes

0

0.2

0.5

1

1.5

Zucchini + white sauce

0

0.3

0.7

1.1

1.6

Pasta + Bolognese

0

0.2

0.6

1.7

2.5

Spinach cake

0

0.2

1.2

1.6

2.6

0.1

0.5

1

1.5

3.6

L+1

L+2

Fat (% added lipids)

Salt (% of added salt) Green beans

Polenta

37

Vegetable soup

0

0.1

0.3

0.6

1.2

Salmon terrine

0

0.1

0.4

0.8

1.5

2

5

11

20

Sweet (% of added sugar + % of added sweetener) Chocolate milk 0.5 Soft white cheese

0.4

2

6.5

15

22

Corn flakes + milk

0

2

8

14

25

Milk shake

1

3

6

11

20

Tea

1

3

5

11

17

Whipped cream

0

3

10

26

42

Chocolate custard

3+0

7+0

12+0

17+0.6

21+6

Verbena infusion

1

2

5

9

17

Strawberry syrup

0

1

2.5

8

14

Apple sauce

0.5

2

9

16

29

Cake

9+0

17+0

18+0.9

18+4

17+8

0

0.5

2

5

9

Orange juice

38