Cross-modal interactions as a strategy for sugar reduction in products targeted 7 at children : Case study with vanilla milk desserts 8 9

The high availability of products with high sugar content, particularly among those targeted as children, has been identified as one of the factors that contribute to the childhood obesity epidemic. For this reason, product reformulation has been recommended as one of the strategies that can be implemented to achieve short-term reductions in children's sugar intake. In this context, the objective of this study was to evaluate the feasibility of using cross-modal (taste-odor-texture) interactions as a strategy for reducing the sugar content of products targeted at children, using milk desserts as case study. A series of 5 vanilla milk desserts were formulated: a control sample with 12% added sugar and 4 sugar-reduced samples (7% added sugar) prepared following a 2 × 2 experimental design by varying vanilla (0.4% and 0.6% w/w) and starch (4.3% and 4.7% w/w) concentrations. A total of 112 children (8-12 years old) tasted the desserts and performed a dynamic sensory characterization task using either temporal check-all-that-apply or temporal dominance of sensations. In addition, they assessed the overall liking of all samples. Results showed that sugar-reduced samples did not significantly differ from the control sample in terms of their average overall liking scores. However, individual differences in children's hedonic reaction were found; three clusters of children with distinctive liking patterns were identified. The increase in vanilla and starch concentration led to an increase in overall liking for over 80% of the children. Sensory dynamic profiles revealed significant but subtle differences among samples. Results from the present work suggest that cross-modal interactions could contribute to minimizing the sensory changes caused by sugar reduction, which could enable to achieve larger reductions if implemented in the context of gradual sugar reduction programs.

recommended as one of the strategies that can be implemented to achieve short-term 23 reductions in children's sugar intake. In this context, the objective of this study was to 24 evaluate the feasibility of using cross-modal (taste-odor-texture) interactions as a 25 strategy for reducing the sugar content of products targeted at children, using milk 26 desserts as case study. A series of 5 vanilla milk desserts were formulated: a control 27 sample with 12% added sugar and 4 sugar-reduced samples (7% added sugar) prepared 28 following a 2x2 experimental design by varying vanilla (0.4% and 0.6% w/w) and starch 29 (4.3% and 4.7% w/w) concentrations. A total of 112 children (8 to 12 years old) tasted 30 the desserts and performed a dynamic sensory characterization task using either 31 temporal check-all-that-apply or temporal dominance of sensations. In addition, they 32 assessed the overall liking of all samples. Results showed that sugar-reduced samples 33 did not significantly differ from the control sample in terms of their average overall liking 34

Introduction 46
Childhood overweight and obesity are one of the most serious health problems 47 9 make it more enticing to children. The instructions were given by a cartoon character (a 212 detective monkey). 213 The study consisted of two tasks: a familiarization step and a sample testing, 214 involving dynamic sensory characterization and hedonic evaluation of the samples. 215 Children were divided into two groups, each of which used a different method for 216 evaluating the temporal sensory characteristics of the desserts: TCATA (n=53) or TDS 217 (n=59). Chi-square tests showed no significant differences in age (p=0.596) and gender 218 (p=1.000) distribution of the two groups. Instructions were given for the sample tasting using a similar procedure (monkey 234 character) to the familiarization task. Children received six milk dessert samples and they 235 were asked to describe them using a TCATA or TDS task. Desserts (20 g) were served 236 in black plastic cups coded with 3-digit random numbers at 8°C. They were presented 237 following a Williams' Latin Square design to avoid order and carry over effects. Still 238 mineral water was used for rinsing between samples. A warm-up sample was included 239 to familiarize the children with the tasting protocol. The warm-up sample was equal to 240 the 7% w/w added sugar dessert identified as "Sugar Reduced" in Table 1, but with a 241 different vanilla. 242 Children were asked to carefully read the list of words before starting the test and 243 to indicate if they had any doubt about their meaning. Attribute definitions were verbally 244 provided if children expressed that their meaning was not clear. Then, they had to place 245 a spoonful of sample in their mouths and immediately touch the "start" button in the 246 screen to describe the sensory characteristics of samples using either TCATA or TDS. 247 Children were instructed to eat the whole spoonful of sample at once and they were not 248 allowed to taste it again. TCATA was performed as described by Castura et al. (2016). 249 Children were instructed to check all the words that applied to describe what they 250 perceived at each time while consuming the sample. They were free to select several 251 attributes concurrently. If a word was no longer perceived, children had to uncheck it. 252 For the TDS task children were instructed to select the word that described the sensation 253 that catch their attention the most at a given time (Pineau et al., 2009). After the dynamic sensory characterization task, children were asked to rate their 262 overall liking using a 9-point hedonic scale (1=dislike very much and 9=like very much) 263 with emoji anchors ( =dislike very much and =like very much was applied on standardized overall liking data to explore segmentation. A linear mixed 280 model was used to evaluate the existence of significant differences among samples 281 within each cluster. In addition, the effect of the factors considered in the 2x2 282 experimental design on overall liking was of interest. In order to evaluate this, both for 283 the whole sample of children and for each cluster, a mixed linear model was used on the 284 overall liking data of the four samples formulated using the experimental design 285 considering vanilla, starch and their interaction as fixed effects. 286 The identified groups were compared in terms of their gender distribution and the 287 temporal method used to evaluate samples using chi-square test. In addition one-way 288 ANOVA was used to compare the groups in terms of their age.

Overall liking 318
When data was analyzed considering the whole sample of children, no significant 319 differences (p=0.14) among milk dessert samples were found in terms of their overall 320 liking. As shown in Table 2, the average liking scores for all samples were close to 7 in 321 the 9-point hedonic scale. This suggests that, on average, children showed a highly 322 positive hedonic reaction to samples, regardless of their sugar content and concentration 323 of vanilla and starch. 324 However, when only the data of the four sugar-reduced samples was analyzed, 325 significant main effect of vanilla was found ( Table 3) were clustered into three groups, with clearly different liking patterns (Table 2). No 337 significant differences between the clusters were found in their age (p=0.643) or the 338 temporal method used for evaluating samples (p=0.368). However, a significant 339 difference in the gender distribution of the samples was found (p=0.035). Cluster 1 and 340 3 were composed by a higher percentage of girls compared to Cluster 2 (63 % and 78% 341 vs 43%). 342  All the other samples showed a significantly lower overall liking score (Table 2). 353 According to the design of experiment, only the main effect of vanilla B showed a 354 significant effect on overall liking of the sugar reduced samples (Table 3). As shown in 355  (Table 2). In this case, linear 361 mixed model focused on the experimental design was not able to identify any significant 362 effect (Table 3). However, vanilla B concentration had a marginal effect (p=0.062). 363 Children in Cluster 3 tended to give higher liking scores to the samples with more vanilla. 364 365

3.2.
Temporal evaluation using TCATA 366  (Table 4). The SR.Vanilla sample also differed from the Control in the citation proportion 379 of the term vanilla flavor, whereas the SR.Starch sample showed a higher citation 380 proportion of the term off-flavor during a small period of time and a lower citation 381 proportion of the term soft for a considerable part of the evaluation (Table 4). Finally, the 382 sample with increase of starch and vanilla did not significantly differ from the Control 383 sample in any sensory attribute (Table 4). Temporal evaluation using TDS 400 The TDS task was not able to capture the temporal evolution of the attributes for 401 most of the samples. As shown in Figure 4, the curves were mostly flat, suggesting that 402 most children selected only one attribute during the whole evaluation. In addition, the 403 citation proportions of all the attributes were lower than 0.35 for all samples. For this 404 reason, few attributes were found to be significantly dominant. 405 The control sample was characterized by the dominance of the term sweet during 406 the majority of the evaluation period and by the dominance of creamy at the beginning 407 of the evaluation. In addition, off-flavor was on the limit of dominance in the first half of 408 the evaluation time (Figure 3). 409 The TDS curve of the reduced sample showed that off-flavor and creamy were Regarding differences among the sugar-reduced samples (Table 5)

Heterogeneity in children's reaction to cross-modal interactions 501
Careful interpretation of the impact of sugar reduction should be paid since it is 502 known that food preferences in children are influenced by multiple genetic and 503 environmental factors (Wardle & Cooke, 2008). This leads to individual differences in 504 food preference and choice, which are likely to influence success of sugar reduction 505 strategies. Despite the majority of children liked all the samples, three groups were 506 identified with distinctive liking patterns. 507 One small group tended to strongly dislike the sample with the highest 508 concentration of sugar which was highly liked by the rest of the children. Differences in 509 sweet preferences among children have been identified due to early experiences, 510 genetic variances and cultural components (Liem & Mennella, 2002;Mennella, Pepino, 511 Yanina, and Reed, 2006;Pepino & Mennella, 2005). For instance, the existence of sweet 512 dislikers among children has been reported by Garneau, Nuessle, Mendelsberg, 20 preference for high sweetness levels, their liking decreased as the concentration of 515 sucrose increased. 516 Considering that the aim of product reformulation is to at least maintain liking of 517 the control sample, it is interesting to note that added sugar reduction of around 40% led 518 to maintained or increased liking for 37.5% of the children (Clusters 1 and 3), while for 519 the remaining 62.5% (Cluster 2) liking decreased but could be restored by the addition 520 of high starch and vanilla levels. Another relevant point is that, even though around 80% 521 of the children gave the highest overall liking to the dessert formulated with the highest 522 levels of vanilla and starch, one group of children showed a strong dislike for this sample. 523 Although the findings regarding individual differences were interesting, it is 524 important to take into account that the number of children in each cluster was small. 525 Future studies should be conducted with a larger consumer sample to confirm the trends 526 found here. In addition, whether the individual differences found in hedonic perception 527 are due to differences in sensory perception, or if they are just the result of differences 528 in children's preference patterns, deserves further investigation. 529 Individual differences could also be related to the nutritional status of children. In 530 this sense, Proserpio et al. (2016) showed that certain aromas had a higher impact on 531 the sensory perception of obese adult woman than normal weight ones. Although in the 532 present study data on children's body mass index was not collected, this information 533 could be valuable for future research. 534 535

Methodological considerations 536
The present study is the first to report the use of dynamic sensory methods with 537 children. Although children reported to understand both methods and were able to 538 complete the tasks, results showed that children mostly used the methods as static. As 539 shown in Figure 4, TDS curves were mostly flat, suggesting that children tended to select 540 only one attribute during the whole evaluation period. In the case of TCATA, although 541 Although the authors reported that this variant did not outperform the classic methods 556 with adults, it might improve children's performance since it could keep their attention for 557 longer, as participants are more actively involved during the evaluation. Moreover, it 558 might help to eliminate false dominance periods at the end of the mastication period or 559 due to hesitation. 560 In addition, it could be interesting to evaluate the application of dynamic sensory 561 methods with solid products that undergo larger changes in their sensory characteristics 562 throughout consumption. The fact that most variation in TCATA curves occurred in the 563 first fifth or quarter of the evaluation period also suggest that children tended to use this 564 method as static: once attributes were selected no further changes were registered. Another methodological consideration of this study is the sugar reduction level that 583 was used. Although ~ 40% reduction in added sugar led to a decrease in overall liking 584 for the majority of the children, the sugar reduced sample was not disliked. Future studies 585 should consider higher reduction levels in order to achieve children's' rejection of the 586 reformulated product, in which compensation strategies such as cross-modal interaction 587 would be more relevant to achieve reformulation goals. 588 589

Conclusions 590
Results from the present work suggest that it is feasible to reduce the added sugar