**About the Special Issue Editors**

**Beverly J. Tepper** Ph.D., is Professor of Sensory Science at the Department of Food Science, Rutgers, The State University of New Jersey, USA, where she directs the Sensory Evaluation Laboratory. Her research program combines food sensory science with nutritional science and psychology to better understand the links between taste, diet, and health. Specific research areas include the influence of genetic variation in taste perception on the pathways linking oral sensations to food preferences, diet selection, and body weight; the role of salivary proteins in sensory perception and oral health; the influence of personal traits on consumer behavior; and sensory evaluation and consumer testing of natural products and novel food ingredients and technologies. She is also the co-founder and director of the Center for Sensory Sciences & Innovation (CSSI) at Rutgers, where she conducts basic and applied research in partnership with the food industry. Dr. Tepper is a Fellow of the Institute of Food Technologists.

**Iole Tomassini Barbarossa** is a Full Professor of Physiology at the Department of Biomedical Sciences, University of Cagliari, Italy. During the last 10 years, she has built a strong and internationally recognized research profile, mainly due to her role as the principal investigator in multidisciplinary studies aimed at analyzing the physiology of the sense of taste and its role in food preferences, nutritional status, and human health. By integrating psychophysics, molecular biology, neurobiology, genetics, nutrition, and electrophysiology methods, these studies have focused on the identification of the physiological basis of individual taste variability; the relationships between taste sensitivity, food behavior, and nutritional status; and on modifications of taste perception. Recently, she designed and patented a new technique based on electrophysiological recordings of the bioelectric potentials generated in the taste cells of the human tongue by taste stimulation, thus providing a direct, objective, and quantitative measure of the peripheral taste function.

### *Editorial* **Taste, Nutrition, and Health**

#### **Beverly J Tepper 1,\* and Iole Tomassini Barbarossa <sup>2</sup>**


Received: 16 December 2019; Accepted: 16 December 2019; Published: 6 January 2020

**Abstract:** The sensation of flavour reflects the complex integration of aroma, taste, texture, and chemesthetic (oral and nasal irritation cues) from a food or food component. Flavour is a major determinant of food palatability—the extent to which a food is accepted or rejected—and can profoundly influence diet selection, nutrition, and health. Despite recent progress, there are still gaps in knowledge on how taste and flavour cues are detected at the periphery, conveyed by the brainstem to higher cortical levels and then interpreted as a conscious sensation. Taste signals are also projected to central feeding centers where they can regulate hunger and fullness. Individual differences in sensory perceptions are also well known and can arise from genetic variation, environmental causes, or a variety of metabolic diseases, such as obesity, metabolic syndrome, and cancer. Genetic taste/smell variation could predispose individuals to these same diseases. Recent findings have also opened new avenues of inquiry, suggesting that fatty acids and carbohydrates may provide nutrient-specific signals informing the gut and brain of the nature of the ingested nutrients. This special issue on "Taste, Nutrition, and Health" presents original research communications and comprehensive reviews on topics of broad interest to researchers and educators in sensory science, nutrition, physiology, public health, and health care.

#### **1. Sweet Taste**

Understanding the role of sweet taste in health and nutrition has been a major focus of chemosensory research for more than 50 years. Although significant strides have been made in this area, a complete understanding of the complex links between sweet taste perception, liking, and intake remains elusive. Tan and Tucker [1] reviewed the current state of knowledge in this area, concluding that current measures of sweet taste perception and liking may have limited capacity to predict dietary behaviours. The characterization of individuals as "sweet likers" or "sweet dislikers" has been a useful concept for understanding person-to-person differences in hedonic reactions to sweetness across a range of intensities. Building on their previous work, Iatridi, Hayes, and Yeomans [2] presented a new methodological approach for fine-tuning sweet-liker/-disliker classifications. These advances are taking place against a backdrop of escalating public health concerns about excess sugar in the diet and are reflected in current dietary guidelines in the United States [3] and many other countries across the globe [4], which now limit daily sugar consumption. To achieve the goal of sugar reduction at the population level, consumers would need to change their behaviours by making different diet choices, selecting sugar-reduced products, or a combination of these activities. Sugar reduction has been an ongoing focus of the food industry. Wee, Tan, and Forde's [5] study of 16 sweeteners provides an up-to-date and comprehensive guide for comparing the potencies of several classes of sweeteners to sucrose, the goal standard. Sweetener classes include, e.g., saccharides and polyols, non-nutritive synthetics (e.g., aspartame, sucralose), and non-nutritive naturals such as stevia.

#### **2. Food Preferences**/**Individual Di**ff**erences**

Understanding individual differences in food preferences and eating behaviours has important implications for both food research and nutrition monitoring. Many of the contributions in this issue examine individual differences, from a variety of perspectives such as age, gender, culture/ethnicity, and genetic variation. For example, to gain insight into food preferences in a cross-cultural context, Wanich et al. [6] compared liking ratings for foods tasted in the laboratory to general liking responses obtained by questionnaire. Jilani et al. [7] studied a large European family cohort (>12,000 respondents) to establish the validity of a single instrument collecting food preference data from children, adolescents, and adults. The review by Keller et al. [8] presents a new conceptual model and fresh look at sex differences in eating behaviours in children. Two papers address the role of genetic variation in food preferences and choice. De Toffoli et al. [9] examined the interaction between PROP taste sensitivity (a marker for bitter taste) and psychological traits on the selection of astringent, polyphenol-rich foods, while the short review by Robino et al. [10] proposes that other genes and phenotypes (in addition to traditional taste-modifying genes) may play a role in food preferences.

#### **3. Umami and Fat Taste**

The role of other taste sensations in nutrition and health remains a vibrant and active area of research interest. Two contributions in this issue focus on fatty acid taste sensations. Sollai et al. [11] utilized a novel technique to measure electrophysiological responses from the gustatory cells of the human tongue following the direct application of oleic acid. They report strong associations between physiological signals and self-reports of fat taste sensations, demonstrating the reliability of this technique. Furthermore, Peterschmitt et al. [12] showed that direct lingual application of long-chain fatty acid to the circumvallate papillae of the mouse activated brain circuits involved in taste signaling, reward, and memory. Together, these studies reveal important features of the gustatory, peripheral, and central mechanisms involved in fat taste that are relevant to both animals and humans.

Finally, Hartley, Liem, and Keast [13] re-examine the notion that umami qualifies as a basic taste. They argue that umami meets most of the criteria for a basic taste—it is elicited by a distinct class of stimuli (e.g., L-glutamate), it activates specific receptor(s), (e.g., T1R1/T1R3), etc., but it does not generate a unique taste quality. They propose a new subclassification called "alimentary taste" for umami, and other taste qualities (such as fat) that may be more important signals for regulating postingestive metabolism than as sensory cues for the presence of specific nutrients in foods.

#### **4. Disease States and Role of the Gut**

Alterations in taste or smell are well-known features of a variety of metabolic diseases and pathological states. However, for many of these conditions, data from well-described clinical populations are scarce. In this issue, Singh et al. [14] present comprehensive findings on taste disruptions and oral complaints in patients with Sjögren's syndrome, an autoimmune disease affecting exocrine glands, such as the salivary glands, which results in dry mouth, burning mouth, and poor oral health. Importantly, this study included patients with Sjögren's syndrome, individuals with so-called "sicca" complaints who do not meet the diagnostic criteria for the disease (and are rarely studied), and healthy controls. There is also a critical need to develop food products that help patients with nutritional diseases to adhere to prescribed diets. Proserpio et al. [15] assessed the acceptability of different formulations of low-phenylalanine foods using a check-all-that-apply (CATA) methodology in individuals with phenylketonuria.

Obesity is increasingly characterized as an inflammatory disease arising from gut dysbiosis associated with an obesogenic diet. In the study by Bernard et al. [16], mice chronically fed a high-fat diet exhibited a blunted preference for sucrose that was partially corrected by supplementing the diet with a prebiotic (10% inulin-type fructan). Examination of caecal contents showed a greater abundance of beneficial bacteria in the diet-induced obese mice fed the prebiotic supplement. These interesting findings suggest that prebiotic supplementation warrants more attention as an aid to the dietary management of obesity.

Lastly, taste receptors are expressed throughout the gastrointestinal tract and are known to release satiety hormones such as GLP-1, CCK, and PYY. In a single-blind, crossover trial, Klaassen et al. [17] delivered a tastant mixture via a naso-duodenal-ileal catheter to healthy participants and measured food intake and satiety from a subsequent meal. However, no differences in outcome measures were observed as a function of duodenal (proximal) or ileal (distal) infusions.

#### **5. Lifestyle Factors**

Two papers examine the extent to which lifestyle factors influence taste perception and food preferences in healthy individuals. Using fMRI, Gramling, Kapoulea, and Murphy [18] demonstrate that chronic caffeine consumers and nonconsumers experience differential activation in neuronal areas involved in reward, memory, and information processing when they are exposed to bitter and sweet tastants. Likewise, Feeney et al. [19] showed that in men, habitual physical activity selectively alters taste perceptions. Specifically, active men gave higher intensity ratings to sweet and umami solutions in comparison to nonactive men.

The study by Larsen et al. [20] examined the complex interrelationships between taste and diet in a cohort of chronic smokers who were also overweight or obese. Because obese smokers reportedly use smoking as a means of controlling their appetite and weight [21], gaining greater insights into taste changes and smoking-related dietary behaviors in this population may have important implications for treatment and prevention. Notably, participants also rated a liking for sweet e-juice, which is used to flavor e-cigarettes, a popular alternative to tobacco cigarettes. Using structural modeling, Larsen et al. [20] showed that taste (including e-juice liking) was associated with body mass index (BMI) in chronic smokers through liking of fats/carbohydrates and that smoking-related dietary behaviors (assessed by questionnaire) could influence BMI by a separate pathway. These novel findings could help to inform the development of new smoking intervention strategies.

#### **6. New Product Formulations**

This volume would not be complete without addressing consumer acceptance of new products and formulations designed to enhance health and wellbeing. Grapefruit is rich in vitamins, antioxidants, and anti-inflammatory compounds, but is rejected by many consumers due to its bitter taste. Gous et al. [22] developed 36 model grapefruit beverages varying in taste, aroma, flavor, and color to characterize their sensory profiles and to identify the formulations best-liked by consumers. Franks et al. [23] present unique findings showing that the type of water (tap, bottled, or deionized) used to brew tea influences sensory characteristics and nutrient extraction. Color, flavor, and epigallocatechin gallate (EGCG) extraction were higher for teas (especially green tea) made with purified water, but consumer liking was higher for less intensely flavored green tea made with tap water. These findings suggest that the consumer's choice of water source can maximize the flavor or health benefits of tea according to their personal preferences.

#### **7. Olfaction**

The determination of the odor detection threshold is a classic technique for assessing smell function, but such methodology is time-consuming and not well suited to diagnostic evaluation in the clinical setting or in the field with a large number of subjects. Using Sniffin' Sticks (odour-impregnated pens) and a Bayesian adaptive algorithm (QUEST protocol), Höchenberger and Ohla [24] established a rapid method with reduced testing duration and less variability between measurements.

**Author Contributions:** B.J.T. and I.T.B. wrote the Editorial. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Review* **Sweet Taste as a Predictor of Dietary Intake: A Systematic Review**

#### **Sze-Yen Tan <sup>1</sup> and Robin M. Tucker 2,\***


Received: 7 December 2018; Accepted: 4 January 2019; Published: 5 January 2019

**Abstract:** Taste is frequently cited as an important factor in food choice, and while a number of studies have attempted to identify relationships between taste function and dietary intake, a systematic review of these studies has been lacking. This review identified studies that examined associations between taste function or taste perception and dietary intake. The purpose was to determine which taste measure was most closely associated with dietary intake in healthy adults. Studies that measured some component of dietary intake, either acutely or longer-term, were eligible for inclusion. Studies were grouped into three categories: those that measured sensitivity (thresholds), intensity, or hedonic responses to sweet stimuli. Sensitivity and intensity studies demonstrated little association with dietary intake measures. Hedonic measurements were more likely to be associated with dietary intake, especially if sweet likers were analyzed separately from sweet dislikers, but the degree of heterogeneity among stimulus concentrations and dietary measures as well as small sample sizes likely obscured more consistent relationships between hedonic evaluation and dietary intake. Due to the potential for within-day and between-day variability in both taste function and dietary intake, future work should explore obtaining more than one taste measurement before comparing results to longer-term dietary assessments and attempts to standardize methods.

**Keywords:** sweet taste; psychophysics; nutrition; diet; threshold; intensity; liking

#### **1. Introduction**

The sense of taste is commonly referred to as the "gatekeeper" of food intake [1]. This concept is supported by consumer surveys that report food choices are made primarily based on the flavor of the selected foods, with considerations about healthfulness or cost typically rated as less important [2]. Taste is an important component of the chemosensory attributes (taste, smell, chemesthesis or chemical irritation) that comprise flavor [3], and thus, guide food selection and intake. Dietary intake, in turn, influences nutritional status and body composition. Thus, individual differences in taste function and perception may lead to differences in dietary behaviors and risk of chronic disease [4].

Each taste quality has been associated with specific nutrients that are important to health and well-being. For example, sweet taste is commonly thought to help identify sources of carbohydrate, sour taste with the presence of vitamins, salty taste with essential electrolytes, and umami with protein [5]. Bitter taste likely serves as a warning against potentially dangerous compounds [5]. If these purported functions are accurate, then positive associations between taste function and/or preference for these taste qualities and related nutrient intake should exist.

Research regarding taste is typically concerned with one of two questions. First, how well does the system function? Sensitivity testing, which involves determining the absolute minimum concentration of a stimulus that can be reliably detected (detection threshold) or recognized (recognition threshold), is frequently performed, but perceived intensity measurements of suprathreshold concentrations are also used. Threshold measurements can take several forms, but these tests usually involve presenting the participant with several samples – only one of which contains the stimulus of interest. The participant is required to identify the sample that contains the stimulus. A variety of approaches in terms of the number of samples to present and number of correct answers needed to stop the experiment exist [6]. Intensity measurements typically involve presenting a stimulus to the participant and asking for a rating of the intensity. Scales commonly used include a visual analog scale [7], a category scale [8], or a general Labeled Magnitude Scale [9]. The second question typically assesses a hedonic aspect, such as, how much is the stimulus liked, the preferred stimulus when a participant is asked to compare two or more stimuli of different concentrations, or the optimal stimulus concentration—often determined using an adjustment method where the participant increases or decreases the concentration of the taste quality. All of the taste measures just described are considered to be independent of each other, providing separate but complementary information about how the stimulus is detected and perceived [10].

When research is conducted on a specific taste quality, model stimuli, often consisting of a prototypical stimulus dissolved in deionized water, are typically used. For example, commonly used prototypical stimuli for sweet taste include sucrose or glucose solutions; whereas, sodium chloride solutions comprise the typical salty stimulus. Participants usually swish and then expectorate the liquid samples, but other approaches, including filter paper impregnated with stimuli [11], cotton swabs [12], edible wafers [13], or edible films [14] have been used. The simplicity of model systems allows for attention to be focused on the taste quality of interest with minimal distraction, but the obvious drawback of the model system is that it does not reflect the complex sensory experiences provided by foods and beverages. Thus, the question that arises is: how closely do taste test results using model systems correlate with dietary intake?

Given their simplicity but seemingly limited ecological validity [15], the ability of taste tests using model solutions to adequately predict dietary intake was previously considered limited [16,17]. However, few studies had adequately assessed intake when this question was first considered [16]. The question remains relevant, as recent work has examined how results from taste testing are associated with dietary intake. For example, the proposal of "fat" as another taste quality has led to renewed interest in connecting taste measurements to dietary intake and weight status (for a recent meta-analysis, see [18]). This suggests that relationships between taste measures and intake remain of interest to taste researchers.

In recent years, sugar intake has been proposed as a potential cause of the increasing prevalence of obesity globally [19,20]. The relationship is especially strong between intake of sugar-sweetened beverages and obesity [21]. As a result, recommendations that added sugar in habitual diets should not exceed 10% of total daily energy intake have been made by a number of governmental and non-governmental organizations including the United States Dietary Guidelines for Americans [22], the Australian Dietary Guidelines [23], and the World Health Organization [24]. Mechanistically, scientists posit that sugar consumption is driven by hedonics, i.e., its pleasant sweet taste, and evidence also suggests that sweet taste enhances the liking and wanting of sweet-tasting foods [25]. Some studies further demonstrated that sugar activates the opioid (e.g., nucleus accumbens) and dopaminergic (e.g., ventral tegmental area and right amygdala) reward centers in the brain [26,27], leading to the notion that sugar is 'addictive' and leads to excessive food intake and subsequent weight gain. Together, these mechanistic studies appear to suggest that sweet taste triggers food seeking behaviors and dietary intake. Although a number of individual studies have performed sweet taste testing using model systems and assessed associations with intake, to our knowledge, a systematic review summarizing these findings has not been undertaken. Therefore, the purpose of this review was to determine if psychophysical tests for sweet taste were associated with dietary intake and, if possible, to determine which test is the most closely associated with dietary intake.

#### **2. Materials and Methods**

A systematic literature search of the electronic databases PubMed, PsycInfo, Web of Science, and CINAHL was conducted. The search string used in PubMed was ("Taste" (Mesh)) AND ("Diet, Food, and Nutrition" (Mesh)); filters included Adult 19+, English, and Human. These filters were used in the other databases when available. Review articles that were identified were searched to identify articles that the searchers missed. Studies that recruited generally healthy individuals and collected at least one psychophysical measure of sweet taste and reported some sort of dietary intake measure, either acute or long-term were included. There was no restriction on adiposity, that is, all categories of body mass index were accepted. Studies were excluded if the populations were currently or had previously been ill, for example diabetes, alcoholism, or eating disorders; had known changes or deficits in chemosensory function, for example gastric bypass surgery patients; were pregnant; or were smokers. The review protocol was registered with PROSPERO, review #CRD42018111833.

After the initial searches were completed and duplicate entries removed, all potential studies were entered into a master database. Initial screenings by title and abstract were completed by the authors. In the case that a determination to include or exclude could not be made based on the abstract, the full paper was reviewed. The authors discussed questions about inclusion or exclusion until consensus was reached. The authors searched the reference lists of relevant articles to identify potential articles (*n* = 2) that were missed by the systematic search.

#### **3. Results**

In total, 3206 publications were identified and 17 were included in this review (Figure 1). Studies were placed into three categories based on psychophysical method utilized: (1) sensitivity measurements consisting of detection and recognition thresholds (*n* = 6), (2) intensity measures (*n* = 8), and (3) hedonic evaluations, namely liking and preference (*n* = 13). Some studies used more than one method; those that did were examined multiple times. Given the heterogeneity of psychophysical measures [10] and stimuli concentrations [28] as well as differences in stimuli tested (glucose vs. sucrose vs. non-nutritive sweeteners) [29] and dietary intake assessment methods [30], a meta-analysis could not be attempted.

**Figure 1.** A total of 17 articles meeting the inclusion criteria were identified.

#### *3.1. Sensitivity Testing*

A total of six studies examined relationships between taste sensitivity and dietary intake [9,16,29,31–33] (Table 1). Studies varied in terms of the stimuli used, e.g., glucose vs. sucrose vs. non-nutritive sweeteners, the ranges of concentration tested, and the dietary assessment methods employed. Sensitivity was measured based on detection threshold [9,29,31,32], recognition threshold [9,16,29], and/or ability to correctly identify a 9 mM sucrose solution three times in a row using a triangle test [33]; individuals who could perform this task correctly were classified as "highly sensitive". Of the six studies identified, only two observed significant associations between sweet taste thresholds and dietary intake [32,33]. One of the studies (*n* = 30) was an acute experimental study that reported that individuals who were highly sensitive to a 9 mM sucrose solution consumed significantly less carbohydrate and more non-sweet foods, dietary protein, and protein as a percent of energy at an *ad libitum* feeding opportunity 60 min after exposure to either a sweet, non-sweet (umami), or "no-taste" soup [33]. The use of a 9 mM sucrose solution to establish sweet taste sensitivity is not an approach that was used by any other study in this review, and the validity of this approach has not been established. The second study (*n* = 56) reported that aspartame threshold was negatively associated with energy intake as assessed by a 7-day food diary [32]. However, the association was very weak, albeit statistically significant, and may have limited implications (beta coefficient = −0.003, *p* < 0.0009); no further association between sucrose threshold and any diet measures were observed. Another study examining non-nutritive sweetener thresholds did not identify diet-taste relationships [29]. Differences in diet assessment methods (FFQ [29] vs. 7-day food diaries [32]) could contribute to these disparate results.

To summarize, most available studies failed to observe a significant relationship between sweet sensitivity and dietary intake, suggesting that testing for sweet taste threshold is not likely to be predictive of dietary intake. The only studies that reported an association found that sweet-sensitive individuals consumed less carbohydrate and more non-sweet foods [33]. The methodological limitations and small samples sizes of these studies also limit the generalizability of the findings.


Abbreviations: [ ] concentration, CHO = carbohydrate, DT = detection threshold, E = energy, FFQ = food frequency questionnaire, F = female, M = male, PRO = protein, RT = recognition threshold, w/v = weight for volume.

#### *3.2. Intensity Testing*

Eight studies examined relationships between measures of sweet taste intensity and dietary intake [7,9,16,29,34–37] (Table 2). As with the sensitivity studies, stimuli and concentrations tested also varied widely. Only two of the ten studies observed significant relationships [9,29]. The first study (*n* = 42) reported negative associations between diet and intensity ratings for a 250 mM glucose stimulus [9]. Intensity was negatively correlated with total energy, carbohydrate (starch as well as total sugar, glucose, and fructose), but not sucrose intake. Sweet food intake was also negatively associated with intensity ratings of the 500 mM and 1000 mM samples. In this study, dietary intake was measured both by 4-day weighed food records as well as by an unvalidated sweet food FFQ and a sweet beverage liking questionnaire. The second study (*n* = 60) reported that intensity ratings for Rebaudioside A and sucralose, both non-nutritive sweeteners, were positively associated with mean total energy intake (*p* < 0.01 for both) [29]. No associations between intensity ratings and other dietary measures, including carbohydrate, sugar, or starch were observed, and no associations with the other sweet stimuli tested (glucose monohydrate, fructose, sucrose, or sucralose) were noted [29]. This study relied on the validated Cancer Council of Victoria Food Frequency Questionnaire [38] to assess dietary intake.

In conclusion, only two studies demonstrated the utility of sweet taste intensity ratings in reflecting dietary intake, and neither study used sucrose—a prototypical sweet taste stimulus. The negative association between sweet taste intensity rating of glucose and energy as well as carbohydrate intake was consistent with the findings from the sensitivity studies that also reported significant negative associations [9,29]. On the other hand, associations with non-nutritive sweeteners (Rebaudioside A and sucralose) were present but positively associated with dietary intake. Further study is needed to understand the underlying mechanisms that contribute to these distinct relationships.



#### *3.3. Hedonic Testing*

A total of 13 papers examined relationships between hedonic evaluation and dietary intake [7–9, 16,28,31,34,36,37,39–42]. As before, the concentrations of sweet solution used in these studies varied considerably as did dietary assessment methods (Table 3). In contrast to the sensitivity and intensity studies, all but one [9] used sucrose. Hedonic measurements included determining the preferred concentration out of a range of stimuli [31] or through an adjustment task [16,42] or a rating of how much the stimulus was liked, typically using either a visual analog [7,28,37,40,41], labelled magnitude scale [9,34,36], or likert-style hedonic scales [8,39]. Five of the studies that measured hedonics also classified participants as sweet "likers" or "dislikers" [28,34,37,40,41]. A sweet liking phenotype has been associated with different hedonic responses to sweetness (for a recent review, see [37]), so failure to identify sweet liker phenotype could influence findings. That is, if the study population was comprised predominantly of sweet likers or dislikers, results could be skewed. Therefore, these studies are presented separately from the others. One study analyzed the data with and without sweet liker classification [37], so it is reported twice – both with those studies that did and did not identify sweet likers.

#### 3.3.1. Studies that Determined Sweet Liking Phenotypes

Among the five studies that distinguished between sweet likers and dislikers, the classification methods used to determine sweet liker status varied greatly [28,34,37,40,41]. Classification was performed by hierarchical cluster analysis [28,41]; by preferred concentration cut-off, i.e., favorable ratings above a specific concentration [34,40]; a mean favorable rating over all concentrations tested [41]; and a pattern of increasing hedonic scores [37]. Among these six papers, three observed relationships between hedonics and dietary intake measures [28,37,40]. Among the studies demonstrating associations with sweet liker status and intake, one (*n* = 418) reported that energy intake from sugar-sweetened beverages was higher among likers compared to dislikers (*p* = 0.008) based on a beverage food frequency questionnaire [28]. A second study (*n* = 196) that examined sweet liker and PROP taster status combinations observed that individuals who were both sweet likers and PROP tasters reported consuming more energy from beverages and fiber as measured by two 24-h recalls [40]. The last study (*n* = 132) reported positive associations between the preferred level of sucrose and frequency of sweet food consumption, intake of refined sugars, and total sugars [37]. Two studies did not observe taste-diet relationships, but the reported sample sizes raise questions about the power of these studies to detect relationships (*n* = 12 (6 sweet likers) [34] and *n* = 36 (12 sweet likers)) [41]. Overall, sweet likers appear to consume more energy from sugar-sweetened beverages and more energy from refined and total sugars. It appears that identifying an individual's sweet liking phenotype may increase the likelihood that relationships between hedonic scores and dietary intake will be observed, especially if sample sizes are sufficiently large enough.

#### 3.3.2. Studies that Did Not Determine Sweet Liking Phenotypes

Among the nine studies that did not classify sweet likers, associations between hedonic responses and intake were observed in five [9,16,31,37,42] but not in the other four [7,8,36,39] (Table 3). Preferred sweetness concentration was associated with greater total energy intake [31], carbohydrate intake [31,42], percent of sweet calories consumed [37,42], refined and total sugars [37], and frequency of carbohydrate-rich food selections [42], while one study observed positive associations with liking ratings of glucose at 500 mM and 1000 mM and total energy and carbohydrate (total sugar, fructose, glucose) but not starch and sucrose intake [9]. One study observed a negative association between preferred sweetness concentration and carbohydrate intake [16]. The studies finding associations between hedonic evaluations and dietary intake used one 24-h recall [31], 4-day weighed food records [9], and 7-day diet records [16,42]. Sample sizes for these studies ranged from *n* = 25 [42] to *n* = 51 [31]. Studies not observing associations reported sample sizes ranging from *n* = 17 [8] to *n* = 100 [7]. In summary, hedonic measures appear to be better correlated with dietary intake, and these relationships are strengthened when sweet likers are analyzed separately.


**3.**Hedonic Studies Examining Taste-Diet Relationships.

**Table** 


**Table 3.** *Cont.* Abbreviations: [ ] = concentration, CHO = carbohydrate, EPIC = European Prospective Investigation into Cancer and Nutrition study, E = energy, FFQ = food frequency questionnaire,female, M = male, w/v = weight for volume.

 F

=

#### *Nutrients* **2019** , *11*, 94

#### **4. Discussion**

The sensory properties of food, including taste, play an important role in food selection and intake [2]. Psychophysical studies exploring taste function and perception have sought to determine if responses obtained in these studies can be associated with dietary intake. Given the challenges of assessing dietary intake [43], a proxy measure that is a simple, quick, and reliable predictor of intake would be welcomed.

Of the taste testing methods used—sensitivity testing, intensity measures, or hedonic evaluation—hedonic ratings proved to be superior in their ability to correlate with dietary intake, although these studies also did not report consistent findings. The fact that sensitivity was not a reliable indicator of dietary intake was not unexpected, as others have noted that an individual's sensitivity to a taste quality often fails to predict intake since these exposures can be quite dissimilar to the suprathreshold exposures experienced while eating [16,44]. Intensity measures lacked predictive power as well. One study observed positive associations between dietary intake and hedonic evaluation but not with intensity [37]. Another study reported that intensity evaluations between sweet likers and dislikers did not differ [28]. These results further support the argument that measuring sensitivity, intensity, and hedonic responses provides distinct but complementary information about the taste sensations experienced by an individual [10], but that, based on the available data, hedonic evaluation may provide a more reliable indication of dietary intake.

Further, among the studies that classified sweet likers and dislikers, three of the five studies reported that sweet likers were more likely to demonstrate associations between dietary intake measures and hedonic evaluations. Sweet likers are typically classified by increasingly favorable hedonic responses to increasingly sweeter stimuli [45]. Thus, the positive associations between hedonic responses and intake of sugar sweetened beverages and sugar intake make intuitive sense. The two studies [34,41] that failed to see associations between hedonic responses and intake in sweet likers had small sample sizes of sweet likers (*n* ≤ 12). Intriguingly, while the methods used to assess sweet liking phenotype differed, results were consistent across studies. This agrees with others who reported that among these methods, no single classification approach demonstrated superiority [45].

The differences in both taste and diet measurements likely contribute to the discrepancies reported. First, a discussion of the taste measurement differences. The stimuli and concentrations used will have a direct impact on results. While different nutritive sweeteners were noted to have detection and recognition thresholds as well as intensity scores that were correlated with each other, actual values differed [46]. This is unsurprising, as different sugars have different potencies; sucrose, for example, is sweeter than glucose at the same concentration [47]. Further, the human sweet receptor responds to many compounds besides mono- and disaccharides, including amino acids, proteins, and non-nutritive sweeteners [48]. Sucrose and glucose are presumed to be the best stimuli to correlate with dietary intake, but this has not been tested, and one study reported that the threshold for the non-nutritive sweetener aspartame was negatively associated with energy intake, unlike sucrose [32]. The concentrations of the sweet stimulus presented to a participant can also influence taste results. Smaller differences between successive concentrations will allow for more precise determination of the taste threshold, but additional trials add to participant burden and increase the risk of fatigue. There is no standardized procedure for determining the difference in concentration between one stimuli and the next. The range of concentrations presented to participants in order to determine sweet liker/disliker phenotypes also varied by study [28]. It is conceivable that some individuals could be classified as sweet likers with one set of concentrations and sweet dislikers if the concentrations presented were higher. This is especially true if sweet liker phenotype is determined by the response to one concentration. Thus, if individuals were misclassified, results could change.

In terms of dietary assessment, it is well known that self-reported dietary information is subject to over- and under-reporting [49]. Over- or under-reporting could obscure taste-diet relationships. In addition, due to the high degree of variability in intake from one day to the next, depending on the nutrient of interest, many days of intake in the form of diet diaries or records must be recorded [50]. For example, at minimum, two weeks of intake records are needed to estimate average energy intake in an individual, which is impractical for many studies, and accuracy declines over time [51]. This number falls to three days when estimating energy intake for groups of people [50]. Even with this reduction, dietary record keeping can be burdensome for participants [43] and items consumed can be poorly estimated or forgotten entirely.

There are two main approaches to reduce participant burden when assessing dietary intake. These include the 24-hour diet recall, where participants are asked to remember what they ate during the previous day rather than recording it as each food and beverage is consumed, or a food frequency questionnaire (FFQ) [43]. The 24-h recall allows dietary information to be recorded at one time point, but accurate information collection relies on trained staff and suffers from recall bias [43]. FFQs employ a checklist approach, where participants can indicate how much and/or how often they consume certain foods. The main drawback of this approach is that the ability to accurately remember and quantify intake is severely compromised [43]. While both approaches are valuable, diet diaries are considered to be more accurate measures [43].

Based on the studies examined, there was no clearly superior method of dietary assessment that was more likely to identify taste-diet relationships. For the sensitivity studies, among the studies observing relationships, one utilized an acute intake measurement, i.e., consumption following a pre-load [33], while the other used 7-day food diaries [32]. Studies not observing relationships between taste sensitivity and dietary intake relied on 4-day weighed food records [9], food frequency questionnaires [9,29], 24-hour recall [31], and 7-day food diaries with predominant taste recorded [16]. For intensity, studies that observed relationships between taste and diet used 4-day weighed food records as well as an unvalidated sweet food FFQ and a sweet beverage liking questionnaire [9] and a validated FFQ not used by any other of the studies included in this review [29]. Studies failing to find associations between intensity measures and diet used two 24-h food recalls [7], multiple (3–14) day diet records [16,34,35,39], *ad libitum* intake of specific test foods [8], and various food frequency questionnaires [35,36,39]. Studies measuring hedonic responses that observed associations used multiple day (3–7) food records [9,16,42], 24-hour recalls [31,40], and food frequency questionnaires [9,28]. Studies that did not find associations used multiple day (3–14) food records [34,39], food frequency questionnaires [36,41], 24-h recalls [7], and food preference surveys [39]. At this time, it is not possible to make a recommendation for one dietary assessment method over the other.

The majority of the studies relied on a one-time measure of taste response and attempted to map this response to dietary intake that spanned over days or months—a further limitation of the literature. Taste responses can vary throughout the day [52] or across days [31], posing problems in terms of test-retest reliability [53]. Day-to-day variability in both taste responses and dietary intake could obscure more immediate or acute relationships. One study noted that taste-diet relationships were observed after a night of sleep that lasted less than 7 h but saw no relationships after a night of longer sleep [31]. Sleep or other confounding variables may obscure taste-diet relationships. One of the two studies that did assess acute intake observed that sweet taste sensitivity correlated with a greater amount of non-sweet foods, protein, and protein as a percent of energy consumed by highly sensitive participants, and those participants also consumed less carbohydrate as a percent of energy [33]. The other study that assessed acute intake observed no relationships between intensity and hedonics [8]. The selection of the foods available for *ad libitum* intake could influence intake; thus, in addition to the different taste measures, it is difficult to compare these studies. Further exploration of whether taste measures are superior predictors of acute intake compared to longer-term intake needs to be undertaken.

There are several limitations to this review. As with all systematic reviews and meta-analyses, the inclusion criteria dictate the findings. While all studies were considered, taste testing studies are at high risk of bias due to the reliance on non-random selection of subjects and failure or inability to blind researchers and participants to the test stimuli or purpose of the study. The decision to focus solely on sweet taste limits generalizability to other taste qualities. The heterogeneity of taste testing and dietary assessment methods makes definitive conclusions difficult. Further work examining taste-diet relationships in children and populations with chronic conditions should be undertaken.

#### **5. Conclusions**

In conclusion, only a small proportion of available studies reported significant associations between taste sensitivity, intensity, and hedonics with dietary intake. However, of those that reported significant associations, sensitivity and intensity measurements (sensory function) were negatively associated with intake, while liking and preferred concentration measurements (hedonics) were positively associated with intake in all but one study. Measures of taste liking and preference appear to provide relatively superior insight into dietary behaviors compared to sensitivity and intensity measures. Future considerations regarding standardizing methods are imperative.

**Author Contributions:** Conceptualization, S-Y.T. and R.M.T.; methodology, S-Y.T. and R.M.T.; validation, S-Y.T. and R.M.T.; formal analysis, S-Y.T. and R.M.T.; data curation, S-Y.T. and R.M.T.; writing—original draft preparation, R.M.T.; writing—review and editing, S-Y.T. and R.M.T.; supervision, S-Y.T. and R.M.T.; project administration, S-Y.T. and R.M.T.; funding acquisition, none.

**Funding:** This work received no external funding.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Quantifying Sweet Taste Liker Phenotypes: Time for Some Consistency in the Classification Criteria**

**Vasiliki Iatridi 1,\*, John E. Hayes 2,3 and Martin R. Yeomans <sup>1</sup>**


Received: 17 December 2018; Accepted: 4 January 2019; Published: 10 January 2019

**Abstract:** Taste hedonics is a well-documented driver of food consumption. The role of sweetness in directing ingestive behavior is largely rooted in biology. One can then intuit that individual differences in sweet-liking may constitute an indicator of variations in the susceptibility to diet-related health outcomes. Despite half a century of research on sweet-liking, the best method to identify the distinct responses to sweet taste is still debated. To help resolve this issue, liking and intensity ratings for eight sucrose solutions ranging from 0 to 1 M were collected from 148 young adults (29% men). Hierarchical cluster analysis (HCA) revealed three response patterns: a sweet-liker (SL) phenotype characterized by a rise in liking as concentration increased, an inverted U-shaped phenotype with maximum liking at 0.25 M, and a sweet-disliker (SD) phenotype characterized by a decline in liking as a function of concentration. Based on sensitivity and specificity analyses, present data suggest the clearest discrimination between phenotypes is seen with 1.0 M sucrose, where a liking rating between −15 and +15 on a −50/+50 scale reliably distinguished individuals with an inverted U-shaped response from the SLs and the SDs. If the efficacy of this approach is confirmed in other populations, the discrimination criteria identified here can serve as the basis for a standard method for classifying sweet taste liker phenotypes in adults.

**Keywords:** sweet taste; hedonics; sweetness; taste test; individual differences; classification method

#### **1. Introduction**

Hedonic responses to taste stimuli are dissociable construct from motivation or a desire to eat (i.e., "liking" vs. "wanting") as proposed by Berridge [1], and these responses influence dietary intake [2–4]. Elsewhere, a conceptual model linking sensation to intake via affective/hedonic responses has also been proposed [5]. Under these models, it is highly plausible that interpersonal variations in hedonic responses to sweet taste—in conjunction with genetic and epigenetic inputs, environmental forces, and other acquired individual characteristic—may contribute to variations in the susceptibility for obesity and obesity-related diseases. For almost half a century, observations of distinct individual liking patterns to sweet taste stimuli have repeatedly been made, thereby challenging the widespread belief that sweetness is universally highly liked. Witherly and colleagues, for example, speculated that humans exhibit up to four distinguishable responses to various sweetened beverages [6], which, as was also illustrated later by Drewnowski [7], could be described as a rise in liking with increasing sweetener concentration followed by a decline (Type I), a rise and then a plateau (Type II), a monotonic decline (Type III), and a non-systematic change in liking (Type IV).

Since the pioneering work of Pangborn [8], sensory scientists using simple sucrose solutions and multiple different scaling methods in laboratory settings have similarly identified at least four different sweet taste liker phenotypes. As summarized in Figure 1, the associated response patterns are characterized by either a positive slope, a horizontal ("flat") slope, an inverted U-shape, or a negative slope. Simpler schemes also exist, where participants are dichotomized into sweet likers (SLs) and sweet dislikers (SDs). The SL phenotype (sometimes reported as the Type II phenotype) generally refers to liking for ever-higher sweetness (e.g., in References [9,10]) and accounts for 48.5% of the published literature [11]. In contrast, the SD phenotype, which shares a very similar distribution (48.2%) with the SL phenotype [11], has been defined differently across various studies: it can describe either as a monotonically decreasing liking as sucrose concentration increases (e.g., in References [12,13]), or a liking for moderate levels of sweetness, which is graphically presented as an inverted U (e.g., in Reference [14]) and sometimes also called Type I phenotype (e.g., in References [15,16]). To note, a few studies identifying both subtypes of the SD response pattern classified them into a single group reported as SD phenotype, as well (e.g., in References [17,18]).

**Figure 1.** Graphical representation of the most commonly reported sweet taste liker phenotypes. The green line corresponds to a phenotype characterized by a rise in liking with increasing sucrose concentration (e.g., sweet liker phenotype), yellow line illustrates an inverted U-shaped hedonic response as a function of sucrose concentration (e.g., inverted-U phenotype), grey line represents an insensitive response to changes in sucrose concentration, and red line corresponds to a phenotype characterized by a decline in liking as sucrose concentration increases (e.g., sweet disliker phenotype). Adapted with permission from Reference [11].

Accordingly, an important question to be addressed is how these distinct hedonic responses to sweet taste can be defined and identified. Among 71 studies we recently reviewed [11], four main phenotyping methods (each relying on different classification criteria) were identified: the visual or algorithmic interpretation of hedonic responses from multiple sucrose concentrations (Method 1a and Method 1b, respectively), the "highest preference using ratings" method that dichotomizes participants based on whether they like the highest sucrose concentration tested the most (Method 2), the "average liking above mid-point" or "positive/negative liking" method where liking ratings are compared to one or two predefined cut-off scores (Method 3), and the "highest preference via paired comparisons" method that categorizes participants based on which sucrose concentration they prefer the most (Method 4). As detailed in our recent review [11], Method 2 and Method 3 suffer from arbitrariness

associated with the strength of the taste stimuli and/or the classification rating thresholds, and both methods are prone to misclassification. The dependence on visual inspection in Method 1a raises the potential for subjective interpretation, and Method 4 involves a choice paradigm based on preference rather than liking per se.

Considering these methodological challenges, along with the ongoing debate over the role of sugar intake as a factor in obesity [19–22], there is strong need for a more precise and consistent method to identify sweet taste phenotypes. The numerous prior studies that have investigated the presence of different sweet taste liker phenotypes and their potential relationship to dietary intake (e.g., in References [14,18,23]) or to body mass index (BMI: e.g., in References [13,16,24–26]) have used widely different methods to define phenotypes; presumably, this has contributed to the inconsistencies reported across studies. Accordingly, in our recent review [11], we suggested that a rapid and reliable phenotyping method is needed to facilitate comparisons across future studies. In our review, we proposed that an optimal sucrose concentration be identified that best separates distinct sweet taste liker phenotypes, in terms of sensitivity and specificity. In 2015, Asao et al. [27] piloted this idea in order to discriminate SLs from SDs. However, as commonly happens with small pilot studies, their sample size likely affected the phenotyping process, potentially leading to an underestimation of the true number of distinct response patterns, a limitation the authors noted in their report. Further, the total number of stimuli they used was rather large [27], raising additional issues of fatigue, adaptation, and inattentiveness. Finally, their participants were tested after they had fasted for an average of 12.1 h [27], which may influence the appetitiveness of the stimuli.

The present study aimed to extend the approach used by Asao et al. [27] while also eliminating some of the methodological issues mentioned above toward a goal of defining a new standardized phenotyping method. We had three aims. First, we identified different sweet taste liker phenotypes statistically. Second, we analyzed these phenotyping data to identify a single sucrose concentration where an application of one or two specific cut-off liking scores ensures the most reliable and replicable definition of each of the identified phenotypes. Last, potential relationships between the motivational state and baseline characteristics of our participants with these sweet taste liker phenotypes were explored.

#### **2. Materials and Methods**

#### *2.1. Participants*

A total of 148 non-diabetic participants aged 18–34 were recruited from students and staff at the University of Sussex between September and December 2017 (Table 1). Cohort size was determined by the suggested minimum of 100 participants in our recent methodological review for the successful identification of the main sweet taste liker phenotypes [11], which was further increased to adjust for the expected underrepresentation of the SD phenotype in our young adult population. Inclusion criteria comprised being medication free (other than oral contraception), smoking less than five cigarettes a week, and having no history of diagnosed eating disorders. Individuals with a current respiratory illness or having recently (less than two weeks) undergone a dental procedure, those being on a weight loss or a medically induced special diet, and women with an irregular menstrual cycle were also excluded. At enrollment, participants gave their written informed consent for inclusion in the study, but they were naive to the study's hypothesis until they had completed all tasks (debriefing provided). The University of Sussex Science and Technology Cross-Schools Research Ethics Committee approved the protocol on the 22 September 2017 (ER/VI40/1), and the study was conducted in accordance with the 1964 Declaration of Helsinki.



BMI, body mass index; Q1, 25th percentile; Q3, 75th percentile. All frequencies reported refer to valid percentages. <sup>1</sup> Participants demonstrating erratic responses to sweet stimuli (*n* = 2) were excluded from this analysis. <sup>2</sup> *p* > 0.05 for all between group comparisons performed with chi-square or Kruskal Wallis tests.

#### *2.2. Taste Test*

#### 2.2.1. Taste Stimuli

To ensure sufficient individual ratings for the development of hedonic curves while trying to minimize confounding effects of adaptation [28] and sensory specific satiety [29], the taste test consisted of seven different aqueous sucrose solutions (0.03125, 0.0625, 0.125, 0.25, 0.5, 0.67, and 1 M) and one water blank, replicated in two separate blocks, for a total of 16 tastings.

The particular concentration range tested was equivalent to sucrose solutions between 1.07% and 34.23% (*w/v*) based on density at 20 ◦C [30], and were chosen to reflect four different considerations: (1) previously reported effects of age on sucrose recognition thresholds [31–33]; (2) the most commonly used sucrose concentrations in prior relevant studies (reviewed in Reference [11]); (3) the sweetness typically encountered in sugar-sweetened beverages [34]; and (4) a compromise between equal log spacing and serial dilution for sample preparation.

All sweet stimuli were prepared at least 24 hours in advance by dissolving food-grade sucrose in mineral water at room temperature. Solutions were stored at 4 ◦C until used. On the experimental day, solutions were allowed to warm up to room temperature prior to presentation, and were presented as 10 mL samples in transparent 60 mL glass cups labelled with random three digit codes. For the solute and rinsing, we used a commercial mineral water with the lowest dry residue concentration available at the time (Volvic, Danone Waters London and Ireland Ltd., London, U.K.).

#### 2.2.2. Rating Scales

Participants evaluated liking and intensity for each stimulus using a horizontal visual analogue scale (VAS) end-anchored with "dislike extremely" (scored −50) and "like extremely" (scored +50) and a vertical generalized labeled magnitude scale (gLMS) with properly positioned descriptors ranging from "no sensation" (scored 0) to "strongest imaginable sensation of any kind" (scored +100), respectively. To ensure within and between-subjects compliance, training for both scales was provided. The practice session for VAS involved rating liking for a series of non-food items, while training in the use of gLMS was applied by evaluating responses to noise and light [35].

On the basis of Cabanac's theory regarding possible enhancement of stimulus value by internal state ("alliesthesia" [36]), two series of VAS appetite ratings [37] were completed before the first and after the second taste test block. All ratings were collected using the Sussex Ingestion Pattern Monitor (SIPM version 2.0.13, University of Sussex, Falmer, U.K.), a computer-based system developed to record and score rating data.

#### 2.2.3. Procedure

The taste test was conducted approximately 2 h after breakfast (between 09.30 am and 12.30 pm depending on each participant's personal routine). Participants were also asked to abstain from smoking, chewing gum, and tooth brushing for the 2 h prior to testing; no restrictions applied to water consumption. During both taste test blocks, a "sip and spit" protocol was followed: participants were instructed to place the entire 10-mL solution in their mouth, swirl it around for 10 s, and expectorate it. They then rated their liking and sweetness intensity before rinsing their mouth with water and proceeding to the next sample. Stimuli were presented in randomized order with participants blinded to the concentration of sucrose tasted each time. After the taste test was complete, demographic (date of birth, sex, and ethnicity) and lifestyle characteristics ("Have you ever been on a diet in order to lose weight?" with possible answers "Yes, one or more times in the past" or "Never," and "Did you usually add more sugar in your coffee, tea or cereals when you were younger?" with possible answers "Yes, I used to add more sugar in my coffee, tea or cereals when I was younger," or "No, I add the same sugar as I did in the past," or "Never added sugar in my coffee, tea or cereals") were collected.

#### *2.3. Anthropometry*

To minimize any possible interactions between the sensory ratings and anthropometric measures, participants revisited the laboratory for a separate early morning session (08:30–10:30) for anthropometry; this visit was scheduled between two days and two weeks after the taste test. Height was measured to the nearest 0.1 cm using a stadiometer and weight to the nearest 0.1 kg using a calibrated body composition analyzer (MC-780MA P, TANITA, Tokyo, Japan). Standardized procedures were followed, including wearing light clothing and no shoes [38].

#### *2.4. Statistical Analysis*

Our primary goals were to (a) algorithmically identify the different sweet taste liker phenotypes in our study cohort and (b) to determine the specific sucrose concentration and associated cut-off score(s) for liking ratings that most reliably allowed for the identification of those distinct phenotypes. Assumptions of normality were tested prior to the main statistical analyses using visual inspection (histograms, Q-Q plots, and bloxplots), and summary statistics (skewness and kurtosis *z*-scores computed by dividing skewness or kurtosis values with the associated standard errors). *Z*-scores (absolute values) larger than 1.96 were indicative of a normal distribution. All ratings are reported as means and standard errors (normally distributed), while medians and ranges are used for age and BMI (not normally distributed); categorical characteristics are expressed as percentages.

Interclass correlation coefficients (ICCs) were calculated to assess test–retest reliability of liking ratings over the two taste test blocks. Given our experimental design, an average measures absolute agreement two-way mixed-effects model was selected [39]. Per the guidelines, an ICC value less than 0.5 indicates poor reliability, values between 0.5 and 0.75 reflect moderate reliability, and values between 0.75 and 0.9 indicate good reliability [40].

As the first step to achieve the principle aim of the current study, an agglomerative hierarchical cluster analysis (HCA) was performed and meaningful groups (clusters) of participants who shared similar liking patterns within each group but were heterogeneous in the between-group contrasts were identified. The mean liking ratings from the eight replicated concentrations in the two taste test blocks were treated as the dimensions for the HCA. The squared Euclidean distance between pairs of cases or clusters and the between-groups (average) linkage method were selected to assist with the merging process [41]. The final decision on the true number of clusters in our dataset was dictated graphically by interpreting the scree plot of coefficients of the agglomeration schedule we designed (Office Excel 2013 for Windows, Microsoft, Washington, DC, USA) and then applying this information ("stopping rule") to the dendrogram provided by the statistical software on the HCA output [41].

We then implemented a two-by-two cross tabulation function to estimate the dyads of sucrose concentration and liking score with the highest sensitivity and specificity in predicting the three distinct sweet taste liker phenotypes. In each two-by-two cross tabulation table, the phenotyping results emerged when a specific dyad of sucrose concentration and liking score was used as the classification criteria for the identification of the sweet taste liker phenotype under investigation were contrasted with the associated phenotyping results suggested by the HCA. The number of true positives (e.g., classified as SL by both the dyad tested and the HCA) and the number of true negatives (e.g., not classified as SL by both the dyad tested and the HCA) indicated the sensitivity and specificity attached to that particular dyad of sucrose concentration and liking score, respectively. Reported liking ratings for stimuli from 0.03125 M to 1.0 M sucrose and potential cut-off values ranging between −20 and +20 in 5-point increments were tested for their prediction value. A K-1 series of sensitivity-specificity tests were conducted, where k represents the number of main clusters previously identified in the HCA.

To test the hypothesis that the sucrose concentration (within subject factor) and the initial clusters or subsequent sweet taste liker phenotypes (between subject factor), as well as their interaction, affect liking and intensity ratings of the presented sweet taste stimuli, two-way mixed ANOVAs with Greenhouse-Geisser correction were carried out. We also employed separate one-way ANOVAs to contrast liking and intensity (both mean ratings and ratings across each of the eight concentrations) by sweet taste liker phenotype. In cases of violation of the equal variances assumption, Brown–Forsythe tests were applied, instead [42]. Post hoc Fisher's least significant difference (LSD) and Games-Howell tests were used as appropriate to further understand the nature of specific paired comparisons.

Nonparametric (Mann–Whitney) tests for the previously reported not normally distributed continues variables (age and BMI) and Pearson's chi-square tests for the categorical variables (gender, ethnicity, dieting history, and habitual use of table sugar) were used to investigate for differences in participant characteristics across the distinct sweet taste liker phenotypes. To explore whether there were also gender differences in measures of interest, additional chi-square tests were performed. Phi symmetric measures instead of Pearson's results are reported in cases of cells with an expected count less than 5.

To ensure participants' compliance with the taste test protocol, changes in hunger and thirst before and after delivering the taste test were explored using paired *t*-tests. We also calculated multiple linear regressions to investigate the degree to which pre- and post-test hunger and thirst predicted liking and intensity ratings across the study sample. The influence of pre- and post-test levels of hunger and thirst was further explored using either one-way ANOVAs or Brown–Forsythe tests [42] to detect differences across the distinct sweet taste liker phenotypes.

The extent to which our method for the identification of the distinct sweet taste liker phenotypes agrees with those in previous literature (see Introduction for details) was assessed by Cohen's Kappas and 95% confidence intervals (CIs) based on the "Estimate ± 1.96 × Standard Error" formula [43]; participants exhibiting an inverted U-shaped response were excluded from this analysis due to the bimodal nature of the phenotyping results elicited by Method 2 and 3. The relevant frequency distributions were also estimated. For the comparison with Method 2 participants who rated the highest sucrose concentration, namely the 1 M solution, as the most pleasant were considered as SLs, whilst all remainder liking patterns were classified into the SD phenotype [44,45]. The agreement with Method 3 was tested using the 0.5 M sucrose solution and the corresponding neutral cut-off hedonic score of 0 (zero) as the classification criteria to discriminate SLs from SDs [23].

Unless otherwise stated, data were analyzed using SPSS Statistics for Windows, version 24.0 (IBM Corp., Armonk, NY, USA). An alpha level of 0.05 was set as the threshold for statistical significance and all performed statistical tests were two-tailed.

#### **3. Results**

#### *3.1. Participant Characteristics*

Participant characteristics are summarized in Table 1; three (two women and one man) failed to report to the laboratory for both sessions. As a whole the cohort tested here was relatively young and lean (*Mdn* = 20.2 years and *Mdn* = 22.1 kg/m2, respectively) and was mainly comprised of women (70.9%); most self-identified as Caucasian (75.7%). Nearly half of the participants reported that they currently add less sugar in their drinks and cereals than when they were younger, and one in three had been on a diet for weight loss at least once in the past. Overall, the women were slightly younger than the men (*Mdn* = 21.1 years for men and *Mdn* = 20.1 years for women; *U* = 1454.5, *Z* = −3.263, *p* = 0.001), and had a lower average BMI (*Mdn* = 23.4 kg/m2 for men and *Mdn* = 21.6 kg/m2 for women; *U* = 1475.5, *Z* = −2.861, *p* = 0.004). This was expected, as it reflects the typical differences in BMI between men and women and the differences in BMI across different age groups in the U.K. [46].

#### *3.2. Taste Test*

Test-retest reliability analysis comparing liking ratings across the two taste test blocks indicated moderate to good reproducibility based on the ICC cut-offs suggested by Portney and Watkins [40] for all but the 0.125 M solution (Figure 2). The two highest sucrose concentrations (0.67 and 1.0 M), and water were associated with the strongest agreement between the two repetitions. As expected, there was a main effect of concentration on liking across all participants with significantly different mean hedonic scores reported for different solutions (*F*(2.12, 312.15) = 10.65, *p* < 0.001, *ηp*<sup>2</sup> = 0.068).

**Figure 2.** Interclass correlation coefficient (ICC) scores (95% confidence interval) for liking ratings from the two taste test blocks across the different taste stimuli.

#### 3.2.1. Identifying Distinct Responses to Sweet Taste: HCA

HCA resulted in ten subgroups of distinct responses to sweet taste with a significant effect of cluster on liking (*p* < 0.001 for all eight sucrose concentrations and effect sizes ranged from 0.22 for the 0.125 M solution to 0.80 for the 1.0 M solution). Three main clusters that accounted for 92% of the study sample were observed. Cluster 1 (*n* = 44) and cluster 3 (*n* = 22) described hedonic response patterns consistent with the sweet liker (SL) and sweet disliker (SD) phenotypes. Both trends were particularly evident for solutions with added sucrose above 0.125 M. Notably however, almost half of the study sample fell into cluster 2 (*n* = 70), where liking increased modestly with concentration up to an intermediate level of sucrose (0.25 M) and then decreased as the concentration continued to increase. Remarkably, participants who were classified into cluster 2 rated both the lowest (*M* = 1.0, *SEM* = 0.76

for 0.03125 M) and the highest (*M* = −1.5, *SEM* = 1.44 for 1.0 M) sucrose concentration as neutral; that is, they neither liked them nor disliked them (*t*(69) = 1.46, *p* = 0.148 for the paired comparison between the lowest versus the highest concentration).

Regarding the 12 participants classified into one of the remaining clusters (clusters 4 to 10), plotting liking as a function of concentration revealed that participants in cluster 9 (*n* = 2) and those in cluster 10 (*n* = 3) followed a classical SL and a SD liking pattern, respectively. Their ratings from the eight different sucrose concentrations resulted, however, in steeper liking curves ("extreme" responses) than those in our main SL and SD clusters, which explains why they emerged as separate groups during the clustering procedure. Indeed, before we applied the "stopping rule" as appropriate (see Section 2.4 for details), participants grouped into clusters 9 and 10 and those grouped into clusters 1 and 3, respectively, had been considered as homogenous only subsequent to the inverted U-shaped phenotype merged with the SL phenotype. Likewise, an inverted U-shaped response corresponding to corresponding to that of cluster 2 was observed for participants classified into cluster 4 (*n* = 2), cluster 7 (*n* = 2), and cluster 8 (*n* = 1): among the heterogeneous mean liking ratings to those of cluster 2, a different optimal sweetness (0.5 M for cluster 4 and 0.67 M for cluster 8) and a higher rating for the breakpoint concentration of 0.25 M sucrose (*M* = 8.9, *SEM* = 1.15 for cluster 2 and *M* = 28.5, *SEM* = 4.50 for cluster 7, *t*(70) = −2.84, *p* = 0.006) stand out. Two single cases of erratic responses were also identified and eliminated from further analysis (cluster 5 and cluster 6).

#### 3.2.2. Identifying Distinct Sweet Taste Like Phenotypes: New Classification Criteria

With regard to the specific sucrose concentration and liking thresholds that best discriminated between the three main clusters, the 1 M solution and liking scores of −15 or lower for the identification of SDs and +15 or higher for the identification of SLs were associated with the lowest number of false negative classifications (90.9 and 97.7 percentage sensitivity for SDs and SLs, respectively) and the lowest number of false positive classifications (93.9 and 93.5 percentage specificity for SDs and SLs, respectively). The results are shown in Tables 2 and 3.


**Table 2.** Sensitivity and specificity checks to discriminate sweet dislikers (cluster 3) from the rest of sweet taste liker phenotypes.

Percentages (%) with an asterisk (\*) indicate the dyad of sucrose concentration and liking cut-off score with the highest combined sensitivity and specificity for the prediction of the sweet disliker phenotype across all dyads tested.

We then applied these classification criteria individually to participants who were assigned to the remaining clusters. The revised grouping (SL phenotype: *n* = 46; 31.5%, inverted U-shaped phenotype: *n* = 73; 50%, SD phenotype: *n* = 27; 18.5%) was in agreement with the classification suggested by the visual interpretation of the shape of the relevant liking curves in all participants except those initially classified into cluster 4. Those participants met the new SD phenotype criteria rather the criteria associated with the inverted U-shaped response pattern. A closer inspection of their hedonic responses revealed that they actually had rated all sucrose solutions as neutral or unpleasant. In addition, integrating the very small clusters into the main groups of responses reduced overfitting and allowed for the subsequent statistical analyses required.


**Table 3.** Sensitivity and specificity checks to discriminate sweet likers (cluster 1) from the rest of sweet taste liker phenotypes.

Percentages (%) with an asterisk (\*) indicate the dyad of sucrose concentration and liking cut-off score with the highest combined sensitivity and specificity for the prediction of the sweet liker phenotype across all dyads tested.

Confirming the diverse nature of the sensory responses to sweet taste among participants classified into the three main sweet taste liker phenotypes, overall liking and intensity significantly varied across these newly defined distinct groups, *F*(2, 56.21) = 89.44, *p* < 0.001 for liking and *F*(2, 77.95) = 5.74, *p* = 0.005 for intensity. A main effect of sucrose concentration (*F*(4.44, 635.19) = 8.53, *p* < 0.001, *ηp*<sup>2</sup> = 0.056), as well as a strong interaction effect between sucrose concentration and phenotype (*F*(8.88, 635.19) = 78.65, *p* < 0.001, *ηp*<sup>2</sup> = 0.524) on liking were also found. As shown in Figure 3, follow-up analysis indicated that participants with an inverted U-shaped response liked the three lower sucrose concentrations at a similar level when compared with both SLs and SDs. When liking ratings of those stimuli were separately contrasted between the two extreme phenotypes, we found that SLs rated them as less pleasant than SDs did. Liking for the 0.125 M sucrose solution did not differ between groups, whereas liking ratings for the rest of the sweet taste stimuli significantly differed by cluster (*p* < 0.001 for most paired comparisons).

**Figure 3.** Liking ratings (mean ± standard error of the mean) as a function of sucrose solutions by the three sweet taste liker phenotypes. Ratings were averaged across the two taste test blocks. The response pattern for the sweet liker phenotype is displayed with a dotted line, the response pattern of inverted U-shaped phenotype with a solid line, and the response pattern of sweet disliker phenotype with a dashed line. Different colors denote the different ranges of liking ratings for 1 M sucrose which, according to the relevant sensitivity and specificity checks (see Tables 2 and 3 for details), could be used for the reliable discrimination between the three distinct sweet taste liker phenotypes: green color corresponds to the range of liking ratings for 1 M sucrose representing sweet likers, yellow color indicates the hedonic response spectrum to 1 M sucrose characteristic of the inverted U-shaped phenotype, and red color corresponds to the range of liking ratings for 1 M sucrose for sweet dislikers.

We next sought to examine the perceived variations in sweetness for the different stimuli between the three sweet liker phenotypes. Paired comparisons between the intensity ratings for each successive concentration and the intensity ratings for the previous indicated that participants were clearly able to distinguish between the different sucrose concentrations (*p* = 0.002 for water and 0.03125 M, and *p's* < 0.001 for all remainder pairs). Rated intensity also increased as sucrose concentration increased across all three sweet taste like phenotypes, *F*(2.32, 336.30) = 535.25, *p* < 0.001, *ηp*<sup>2</sup> = 0.787 (Figure 4). SDs overall perceived the taste stimuli as sweeter (*M* = 23.3, *SEM* = 1.62) than both SLs (*M* = 17.2, *SEM* = 0.73; *p* = 0.001) and participants classified in the inverted U-shaped phenotype (*M* = 19.2, *SEM* = 0.96; *p* = 0.015). No interaction effect between concentration and sweet taste like phenotype on intensity was, however, observed, *F*(4.67, 333.68) = 521.10, *p* = 0.082, *ηp*<sup>2</sup> = 0.027.

**Figure 4.** Intensity ratings (mean ± standard error of the mean) as a function of sucrose solutions by the three sweet taste liker phenotypes. Ratings are averaged across the two taste test blocks. The intensity curve of the sweet liker phenotype is displayed with a dotted line, the intensity curve of the inverted U-shaped phenotype with a solid line, and the intensity curve of the sweet disliker phenotype with a dashed line.

To explore whether the identified sweet taste liker phenotypes were merely indirect consequences of differences in perceived intensity rather than true differences in hedonics per se, liking ratings were also plotted as a function of intensity separately for the three main clusters. As shown in Figure 5a–c, no such indication was found.

#### 3.2.3. Pre- and Post-Test Levels of Hunger and Thirst

Pre-test levels of hunger (*M* = −7.5, *SEM* = 2.11) and thirst (*M* = 0.3, *SEM* = 1.68) confirmed participants' compliance with the taste test preparation instructions, whereas the increase in hunger (*t*(147) = −3.25, *p* = 0.001) and decrease in thirst (*t*(147) = 2.32, *p* = 0.022) over time was also in line with the effects of the "sip and spit" and "mouth rinsing with water" parts of the taste protocol. Neither hunger nor thirst ratings before taste test block 1 or after taste test block 2 predicted liking (*F*(2, 145) = 2.065, *p* = 0.130 for pre-test levels of hunger and thirst; *F*(2, 145) = 0.607, *p* = 0.546 for post-test levels of hunger and thirst) or intensity (*F*(2, 145) = 1.041, *p* = 0.356 for pre-test levels of hunger and thirst; *F*(2, 145) = 0.403, *p* = 0.669 for post-test levels of hunger and thirst) across the study sample. When ratings of hunger and thirst were examined against the three distinct sweet taste liker phenotypes, non-significant differences were found (*F*(2, 143) = 2.410, *p* = 0.093, and *F*(2, 143) = 0.094, *p* = 0.910 for pre-test levels of hunger and thirst, respectively; *F*(2, 76.22) = 0.986, *p* = 0.378, and *F*(2, 143) = 0.107,

*p* = 0.899 for post-test levels of hunger and thirst, respectively). These data clearly show that the group differences in sweet liking cannot be attributed to the observed changes in hunger or thirst.

**Figure 5.** Individual ratings of liking as a function of perceived intensity for the sweet taste stimuli in (**a**) sweet likers, (**b**) individuals exhibiting an inverted U-shaped hedonic response, and (**c**) sweet dislikers. Lines represent the average ratings across individuals classified within each phenotype.

#### *3.3. Participant Characteristics by Sweet Taste Liker Phenotype*

Possible variations in participant characteristics relative to sweet taste liker phenotype were also examined. Gender (*χ*2(2, *N* = 146) = 2.39, *p* = 0.302), ethnicity (*φ* = 0.152, *p* = 0.496), dieting history (*χ*2(2, *N* = 144) = 1.84, *p* = 0.400), habitual use of table sugar (*φ* = 0.194, *p* = 0.240), age (*H*(2) = 2.60, *p* = 0.273) and BMI (*H*(2) = 0.67, *p* = 0.717) did not differ between groups. All associated values by phenotype are summarized in Table 1.

#### *3.4. Comparison to Existing Classification Methods*

When Method 2 (rating the 1 M sucrose solution or not as the most pleasant) and Method 3 (rating the 0.5 M sucrose solution higher than 0 or not) were used to distinguish the different sweet taste liker phenotypes, the proportions of SD and the SL were respectively overestimated: 113 participants were classified as SDs according to Method 2 and 108 as SLs according to Method 3. Compared with our phenotyping method, in both cases, the majority of those participants (56.6% of SDs in Method 2 and 53.7% of SLs in Method 3) exhibited an inverted U-shaped response. Focusing on Method's 2 phenotypic classification, all 27 participants classified as SDs using our method were also identified as SDs using Method 2. Regarding the SL phenotype, 22 out of 46 participants initially fell into the SL phenotype were classified as SDs using Method 2. Those 22 participants liked the 1 M sucrose solution significantly lower than the previous concentration (*M* = 25.3 for 1 M versus *M* = 30.6 for 0.67 M, *p* = 0.014), while no significant difference was observed when compared with the third higher sucrose concentration (*M* = 25.3 for 1 M versus *M* = 28.4 for 0.5 M, *p* = 0.222). The kappa coefficient was accordingly low at 0.447 (95% CI: 0.286 to 0.608). In contrast, the agreement with Method 3 was good with a Kappa coefficient at 0.879 (95% CI: 0.764 to 0.993). All SLs identified using our method were also classified as SLs by Method 3. The two phenotyping approaches were also in line regarding the SD phenotype: only four SDs using our method were discordantly classified as SLs using Method 3. Those participants had a mean liking for the 0.5 M barely over the neutral point (*M* = 1.1) and their liking rating for the 1 M, which was our concentration of choice for distinguishing sweet taste liker phenotypes, was as low as −28.7. A graphical representation of the level of consistency/disagreement among the methods compared here is provided in Figure 6.

**Figure 6.** Comparison of the distribution of sweet taste liker phenotypes in our study sample when different classification methods were used. Method 2 (rating the 1 M sucrose solution or not as the most pleasant) and Method 3 (rating the 0.5 M sucrose solution higher than 0 or not) were, by definition, limited to a two-response group phenotyping outcome (binomial distribution), while HCA method (rating the 1 M sucrose solutions higher than +15, lower than −15, or between −15 and +15) allowed for the identification of three distinct sweet taste liker phenotypes. 133 participants (77.4%) versus 27 (18.5%) were classified as SDs and 108 participants (74.0%) versus 46 (31.5%) were classified as SLs when Method 2 and Method 3 were contrasted with the method we proposed here (HCA method), respectively. Different colors of the stacked columns and the associated data labels (numbers) correspond to the number of participants classified into the phenotype of the same color when the HCA method was used. Data labels (numbers) within each column add up to the total number of participants classified into the phenotype illustrated at the upper end of the relevant column. Asterisks (\*/\*\*) denote alternatives to our definition for SLs and SDs. SDs, sweet dislikers; SLs, sweet likers.

#### **4. Discussion**

#### *4.1. General Findings*

The present report describes how hedonic responses to taste stimuli of varied sweetness can be algorithmically interpreted using HCA, and clustered into groups that represent similar sweet-liking patterns. For the current dataset, consistent differences in liking ratings across the eight sucrose solutions were found, which then allowed a clear characterization of participants as SLs, those with an inverted U-shaped response, or as SDs. Another key feature of the study was the subsequent identification of the 1 M aqueous sucrose solution and the VAS-based cut-off liking scores of −15 and +15 as the statistically reliable criteria to efficiently categorize individuals into these three different sweet taste liker phenotypes.

#### *4.2. HCA Selection Advantages*

Regarding our decision to use HCA for the identification of different sweet taste liker phenotypes, this was principally driven by the need for a statistically robust and unbiased merging of individuals into groups. Indeed, using an advanced statistical clustering technique allowed the three sweet taste liker phenotypes to emerge, whereas this would have been difficult to discern using more traditional visual inspection methods, particularly if the inspector was assuming a simple dichotomous mode. HCA is also based on hedonic responses across multiple stimuli rather than based on an arbitrarily selected single liking rating or the average value of hedonic scores of different stimuli. Accordingly, most elements of subjectivity and arbitrariness noted in the other phenotyping methods discussed earlier were controlled for. When we re-analyzed our current data using other widely used methods (defined as Methods 2 and 3 in the introduction, and in our recent review [11]), many participants were misclassified relative to the cluster analysis performed here, as the bimodal phenotyping approach in those methods assumes a priori that there are only two distinct response patterns. Critically, the HCA analysis shown here, as well as other recent studies [9,13], all suggest that response patterns for sweet stimuli are better described by three distinct phenotypes. Regarding the observed overestimation of SDs by Method 2 and of SLs by Method 3, this was a consistent feature of those methods in our recent evaluation of the impact of different sweet taste liker classification approaches [11]. In contrast, discriminating participants between the different sweet taste liker phenotypes based on a single sucrose concentration and predetermined cut-off liking scores as used in Method 3, led to the least misclassifications, further supporting the utility of such a phenotyping approach.

#### *4.3. Phenotyping Results*

Our findings confirm some [8,9,13,47,48] but not all, studies using phenotyping methods that allowed for a non-dichotomous identification of sweet-liking patterns. Indeed, in some published reports, participants with an inverted U-shaped response were considered as outliers [12,15,17], whilst elsewhere they were treated as homogeneous with the SDs [49–51]. Here, the generated icicle plot of our statistical output (not shown) revealed that during the final stages of the clustering process, SLs merged with those from the inverted U-shaped phenotype before SDs joined them both, uncovering a greater resemblance of the SL rather than of the SD phenotype to the inverted U-shaped response group. It is then plausible to assume that eliminating or misclassifying this intermediate phenotype is problematic and possibly obfuscates potential relationships between sweet taste liker phenotypes and health outcomes of interest. We also noticed that the sucrose concentration associated with the highest liking in the inverted U-shaped response group (i.e., the 0.25 M), was in line with the concentration observed in most previous work [15–18,27,52,53], although lower values have also been reported [8,14,48,54]. Practically speaking, this commonly identified 0.21–0.3 M range of sucrose concentration threshold in individuals who like intermediate levels of sweetness is lower than the sugars composition of the commercially available sweetened beverages [34]. This may potentiate the argument for reexamining the utility of sugar-tax policies [55]. The multisensory aspects of tasting real-life products should not, however, be disregarded [56], as well as the possible attenuating or enhancing effects of other flavor components on perceived sweetness in complex products [57–60]. As sagely noted by Pangborn, "a change in one ingredient can cause multiple physical-chemical interactions which alter several sensory attributes simultaneously: appearance, aroma, texture, taste etc." [61] (p. 65).

Turning now to the frequency distribution of the identified sweet taste liker phenotypes, one third of our participants were classified as SLs, a proportion consistent with observations by others who also used HCA as their phenotyping method of choice [9,13,14]. Conversely, results in Asao et al. [27] and Kim et al. [62] indicate that this sweet-liking pattern accounted for roughly 50% of their study samples. Two possible explanations can be considered. First, the absence of a monotonically negative slope implies that individuals in both cohorts generally exhibited stronger liking for sweetness. Notably, in Kim et al. [62], two thirds of those classified in the inverted U-shaped phenotype rated 0.7 M as the most liked, a sucrose concentration breakpoint twice as high as the concentration we identified. Second, in those studies, sweet-liking was assessed under extreme motivational states with participants' hunger [27,62] and/or satiety [62] being manipulated. Critically, when the same Korean researchers replicated their study using a more typical pretest protocol (i.e., refraining from eating for one to two hours prior to the taste test), their measures generally correspond with the data shown here. Focusing on the frequency distribution of the monotonically negative slope regardless of the SD label, our findings disagree with previous observations. For example, of the 650 age diverse adults tested by Garneau et al. [13], only 55 exhibited decreasing liking as concentration increased. Presumably, this is due to the relatively low sucrose concentrations they used; indeed, the highest concentration they used (0.40 M) fell near the concentration breakpoint we identified for our inverted U-shaped phenotype. In contrast, SDs in Kim et al. [9] were approximately as frequent as SLs and as participants in the inverted U-shaped phenotype (31.7, 32.5, and 35.8%, respectively). Nonetheless, they reported that, for the purposes of the study, two distinct clusters were treated as a single sweet-liking pattern representing the SD phenotype, with no further information provided; each of those clusters accounted for 10 and 21.7% of the total sample, respectively [9].

Here, despite the similar liking ratings of the lowest and the highest sucrose concentration by participants classified into the inverted U-shaped phenotype, perceived sweetness varied considerably when intensity ratings of those stimuli were contrasted. Therefore, this type of response cannot be attributed to reduced sensitivity to taste stimuli or from differences in recognition thresholds; rather, it appears to reflect a distinct liking pattern. Figure 5a,c indicated that this is also true for the SL and the SD phenotype, since inclusion of intensity ratings in the liking plots generated the expected liking patterns. In previous research, any differences in sweetness intensity between participants, when reported, were interpreted independent of the associated phenotyping results (e.g., in References [45,63,64]). The few studies that have contrasted sweetness intensity between the defined sweet taste liker phenotypes have had mixed outcomes: some studies report greater overall sweetness intensity in SDs than in SLs and/or than in other phenotypes in line with what we observed here [12,15,49,65], but the majority found no differences in sweet taste perception [10,13,16,66–71]. These inconsistencies could arise from several factors including the phenotyping methods and the stimuli concentrations used in these studies. Many of the most relevant studies did not, however, specifically report differences in sweetness intensity between their defined sweet taste liker phenotypes, limiting meaningful contrasts between our findings and prior work.

#### *4.4. Recommended Criteria for the Identification of Distinct Sweet Taste Liker Phenotypes*

Except for a pilot experiment [27], this is the first study suggesting specific criteria for the identification of the distinct sweet taste liker phenotypes that could be considered as both statistically robust and easy-to-apply. One core element of the proposed simpler approach is the administration of a single sucrose concentration that allows for both a less time-consuming and resource-demanding assessment process and for elimination of potential issues from the contrast effects which are

"hard-wired" to longer protocols [72]. Within the taste literature, this in a not a novel concept. In 1980, Lawless addressed the need to identify an efficient classification method that could be used to rapidly screen large cohorts in terms of bitter taste phenotypes for phenylthiocarbamide (PTC) and 6-n-propylthiouracil (PROP), i.e., thiourea tasters and nontasters [73]. After using multiple approaches within the same study cohort, he concluded intensity ratings (on a 7-point scale) for a single antimodal concentration of PTC or PROP presented in a two-series taste test allowed for a rapid and reliable separation of the tasters from the nontasters [73].

Despite using a similar analysis to that of Asao et al. [27], we concluded that approximately twice the concentration of sucrose, compared to the concentration they proposed, is required to deliver the highest sensitivity and specificity in the discrimination between distinct sweet taste liker phenotypes. A small sample size, dichotomous grouping, and participants' pre-test fasting state in the earlier pilot experiment [27] raise questions about the broader applicability of the concentration (0.598 M sucrose) recommended in their study. Indeed, other studies using multiple sweet taste stimuli identified concentrations ranging from 0.83 M (e.g., in References [66,74–78]) to 0.99 M (e.g., in References [79–81]). Moreover, the 0.6 M sucrose solution referred in Tuorila et al. [23] was actually shortlisted from their previous work where two additional lower concentrations were tested but not any higher [82]. Finally, the replication in our sample of the proposed by Asao and colleagues' U-shaped association between sucrose concentration and reproducibility of the liking ratings across the repeated blocks of the taste test [27] may also bear critically upon sweet-liking protocols based on intermediate concentrations. Indeed, taste measures for about 40% of the adult sample in Garneau et al. [13] indicated indifferent responses to a range of stimuli between 0 M and 0.4 M sucrose.

Considering the comparatively less sophisticated and less restrictive concepts of the VAS compared to the labelled magnitude or Likert-type scales, the decision to record liking on an analogue scale further strengthens our classification criteria proposal. In particular, VAS-based ratings are independent of the range of prior sensory experiences and of the assumption that the same descriptors (labels) reflect equivalent meaning across different responders [83,84]. That said, in our lab, we have repeatedly observed that participants find VAS to be more straightforward than gLMS, although when we directly contrasted the two scales in a sample of young educated adults, VAS and gLMS yielded similar results [17]. Additionally, VAS is appropriate for recording the multi-dimensional continuum of human responses that a fixed pre-coded format does not by principle permit [85]. Clearly, no scaling approach is perfect: the "anchor effect" phenomenon (centering bias) characterized by less use of the extreme response has been associated with most rating scales, the VAS included [72]. Overall, we propose that utilizing VAS for sweet-liking assessment when phenotyping protocols are applied to groups of diverse characteristics is likely to come with the least challenges.

#### *4.5. Controlling for Protocol Conditions*

Although previous research presents an inconclusive picture [16,62,86], some studies report an effect of hunger [10,87,88] and thirst [89] on liking for sweet taste stimuli. It was thereby critical to ensure that recorded sensory responses were not driven by participants' motivational state and that the motivational state did not differ between the contrasted sweet taste liker phenotypes. Analysis of the pre- and post-test levels of hunger and thirst across our study sample and between-groups confirmed this was not so.

The nature of changes in levels of hunger and thirst over the test period (increased and decreased by 15.2% and 10.1%, respectively) also indicated little or no likely influence of post-ingestive effects of sucrose on the sensory-related measures [90], suggesting the "sip and spit" protocol worked as expected. Notably, Running and Hayes [91] observed no significant differences in the rated intensity of a 0.5 M sucrose solution when "sip and spit" and "sip and swallow" protocols were contrasted. Nonetheless, the differences in the density of taste buds [92] and in the associated saliva [93] across the different regions of the oral cavity and the known role of gastrointestinal tract's sweet taste receptors in metabolic regulation [94,95], suggest a need for both explicit instructions and subsequent compliance

checks in sensory evaluations, particularly when a wide range of concentrations or a relatively strong solution are being tested.

#### *4.6. No Effect of Sweet Taste Liker Phenotype on Participant Characteristics*

Analysis of this young healthy sample found no effect of sweet taste liker phenotype on the few demographic, lifestyle, and anthropometric characteristics we examined. First, the frequency distribution of the SL phenotype did not differ between women and men. With the exception of the multi-ethnic cohort of Thai et al. [53], lack of sex differences in sweet-liking is consistent with previous published work focusing on sweet taste liker phenotypes generated from simple sucrose solution-based taste tests and where women and men were represented equally [27,49,52,64,66,77]. In his recent review, Spence [96] argues that individual differences rather than sex differences might be the most important influence on shaping our taste worlds, particularly when the hedonic aspects of taste are studied. Animal models provide equivocal results on sucrose sensory properties by sex [97]. These findings fail to support Katz's theory of "gendered eating patterns" generated by either evolution or, according to others, by cultural norms [98], as well as baseline reports from the NutriNet-Santé study where, remarkably, men and not women liked sweet tastes more [99]. It is worth stressing though that sensory data in the French cohort were collected indirectly using "Pref-Quest," a proxy of laboratory-based taste tests that measures recalled liking for different taste modalities via asking questions on selective food items and eating habits [100]. In the present work, we also failed to observe an effect of age on hedonic responses to sweet taste. This stands in direct contrast to the fairly consistent effect of age on sweet-liking whenever children or adolescents were compared with adult populations [101–104], and may be due to the relatively restricted age range tested here. To note, in some [13,16,74,76,78,105–108] but not all [13,81,109–111] studies testing middle-aged or older adults, SDs and those with an inverted U-shaped response outnumbered SLs. Critically, methodological limitations that may lead to possible overestimation of the SD phenotype in prior studies cannot also be overlooked [11].

Other factors worth exploring with regard to humans' responses to sweet taste are dieting and BMI. Regarding attempts to investigate how being on a weight loss diet affects classification into the distinct sweet taste liker phenotypes, evidence has been loose and is drawn on research on sweet-liking either as a continuous measure (e.g., in References [112–114]) or assessed via questionnaires instead of laboratory-based taste tests [99]. As discussed in a recent review, bariatric surgery is also likely to augment gustatory sensitivity to sweet taste and to attenuate relevant hedonic responses post-operatively [115]. In our study, being a former dieter was more apparent in SDs. This may seem counterintuitive to the sensory specific satiety theory (decline in pleasantness for a food stimulus subsequent to consumption compared with the uneaten [29]), but could be backed up within the hedonic hunger context (motivation to consume palatable foods in the absence of food deprivation [116]). Nonetheless, no explicit information on the timing, duration, or mode of the dietary regime or the extent of weight loss and weight regain was collected. Additionally, considering the small size of this particular subgroup and the subsequent lack of significance, caution is advised in interpreting this observation until replicated. BMI, on the other hand, did not differ across the three sweet taste liker phenotypes. Although one can argue that this was due to the limited range of BMI in our sample, our finding was consistent with a sizable body of published evidence [13–15,17,24,49,53,66,69,76,106,117–119]. It is also of note that some early reports testing individuals of greater BMIs showed that obese were more often classified into the SD phenotype than normal-weight participants [16,26,54,120,121]; only one study of 12 participants has provided suggestive evidence for the opposite association [25].

#### *4.7. Potential Mechanisms*

Different mechanisms may account for the observed variations in affective responses to sweet taste, and fundamental biology likely plays a part. Sweet tasting substances activate various neural circuits

including some associated with dopamine-linked reward centers in the prefrontal cortex [122–124]. This activation accommodates the urge to meet physiological needs such as the central nervous system's energy supply (e.g., in Reference [125]). Internal state-specific factors ("homeostasis") have also been implicated in explaining the variation of hedonic responses to sweet taste as a function of deprivation state. In this context enhanced sweet-liking in fetuses [126,127] and infants [128–130] may relate to the increased needs for energy during the stages of rapid growth [131]. Likewise, Coldwell and colleagues reported that SL adolescents had higher levels of a bone growth factor compared with their SD peers [49]. Similarly, negative gustatory alliesthesia, which refers to diminishing liking as a response to internal energy abundance (as in satiety or obesity) [36], has been proposed to contribute to the apparent inverse relationship between BMI and sweet-liking.

Later advances have implicated taste genetics with sweetness, both directly and indirectly. TAS1R2 and TAS1R3 taste receptor genes have directly been linked to sweet taste perception [132–134]. The heterodimeric protein encoded by these genes is expressed in taste receptor cells in the oral cavity, providing the mechanism by which sweet taste occurs [135]; subsequently, these receptors have also been found in extra oral tissues [123]. Salivary glucose levels and salivary protein profile have recently identified as additional potential determinants of sweet taste perception [136]. Finally, some reports suggest that differences in the density of structures that house taste cells (i.e., fungiform papillae) may explain differences in suprathreshold taste intensity, including sweetness [92,137], although others account conflict with this explanation [138–141].

#### *4.8. Limitations*

The present study has some limitations that require further confirmatory analyses in different populations to allow the proposed method to be applied universally. First, we had a gender-imbalanced sample of young adults primarily from European Caucasian ancestry. Past literature has partly identified more SLs than SDs when direct contrasts between younger and older adults were performed [16,26,47,77]. Whether sweet taste liker phenotypes vary by ethnic group is, however, not yet well understood [18,23,49,53,76,107]. Nevertheless, due to the higher risk of many non-Caucasian ethnic groups and of older versus much younger individuals in developed countries for non-communicable diseases [142], this research area is worthy of further investigation. Our findings may also not translate to populations with a different habitual intake of sugar. Studies in the U.S., for example, suggest a slightly higher daily intake of free sugars [143] compared with U.K.-based cohorts [144], whereas the recommended daily allowance [145] is also double the U.K. recommendations [146]. On the basis of the conflicting evidence surrounding the influence of exposure in sweet-tasting foods on hedonic responses to sweetness [147,148], this limitation may leave particular populations vulnerable to any possible interplay between sweet-liking patterns and eating patterns and therefore much still need to be learned. Moreover, women and men in our sample were not of a representative BMI for their age-matched group [46]. Whilst this is presumably a caveat for the generalizability of our results, the reader is advised to consider that, as noted earlier, both in our study and elsewhere, BMI did not differ by sweet taste liker phenotype. Still, the fact that the observed proportion of SDs was relatively low, although it was expected from phenotyping results from prior studies using HCA (see Section 4.3 for details), it also means that group contrasts need to be treated with some caution. Finally, no phenotyping method is beyond limitations. The one inherent in using HCA is the lack of a formal "stopping rule" in the clustering process; the researcher is called to indicate the number of stages displayed in the agglomeration schedule that need to be eliminated from further merging and then manually incorporate this decision on the generated dendogram [41].

#### **5. Conclusions**

The present study confirms that the expression of sweet-liking is not universal but responses to sweet taste stimuli vary considerably across people. What is new is the statistical determination of some robust but concurrently usable classification criteria for the identification of the different sweet taste liker phenotypes in a large-scale study. Despite limitations arising mainly from participant characteristics, there is good reason to believe that our approach might still be widely applicable as HCA-based liking patterns between our U.K. based study and those by American [13] and Korean [9] researchers largely align. Conceivably, the potential of a broader use of the psychophysical comparisons we delivered herein in epidemiological studies and clinical trials could have a fruitful impact on research associated with health and wellbeing. Accordingly, we may now have appropriate tools to finally address a longstanding issue first Mattes noted over 30 years ago, that is: "The question remains whether individual responsiveness to sweet taste can tell us anything about the individual, his or her physiological or nutritional status, or the likely patterns of food selection." [149].

**Author Contributions:** Conceptualization, V.I., M.R.Y., and J.E.H.; methodology, V.I., and M.R.Y.; software, V.I.; validation, V.I., and M.R.Y.; formal analysis, V.I.; investigation, V.I.; resources, M.R.Y.; data curation, V.I.; writing—original draft preparation, V.I.; writing—review and editing, M.R.Y., and J.E.H.; visualization, V.I., and J.E.H.; supervision, M.R.Y.; project administration, V.I., and M.R.Y.; funding acquisition, M.R.Y.

**Funding:** This research was funded by the World Sugar Research Organization (WSRO) and the Doctoral School of the University of Sussex. J.E.H. also receives salary support from United States Department of Agriculture Hatch Act funds [PEN04565] and the Pennsylvania State University. M.R.Y. is employed as a Professor at University of Sussex.

**Acknowledgments:** The authors would like to thank Rosalie Considine-Moore (University of Sussex) for her contribution to data collection.

**Conflicts of Interest:** V.I. declares no conflict of interest. J.E.H. has received speaker fees, travel reimbursements, and/or consulting fees from federal agencies, nonprofit organizations, trade/commodity groups, and corporate clients in the food industry. M.R.Y has received direct research funding from numerous sources including national and international companies, as well as speaker fees, travel reimbursements, and consultancy fees from various companies, none of which impact on the work reported here. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **A Comparison of Psychophysical Dose-Response Behaviour across 16 Sweeteners**

#### **May Wee 1, Vicki Tan <sup>1</sup> and Ciarán Forde 1,2,\***


Received: 4 September 2018; Accepted: 26 October 2018; Published: 2 November 2018

**Abstract:** Reduction or replacement of sucrose while maintaining sweetness in foods is challenging, but today there are many sweeteners with diverse physical and caloric compositions to choose from. The choice of sweetener can be adapted to match reformulation goals whether these are to reduce calories, lower the glycaemic response, provide bulk or meet criteria as a natural ingredient. The current study sought to describe and compare the sweetness intensity dose-response, sweetness growth rate, sweetness potency, and potential for calorie reduction across 16 different sweeteners including sucrose. Sweetness growth rate was defined as the rate of change in sweetness intensity per unit of sweetener concentration. Sweetness potency was defined as the ratio of the concentration of a sweetener to that of sucrose at equivalent sweetness intensity, whereas the potential for calorie reduction is the caloric value of a sweetener compared to sucrose at matched sweetness intensities. Sweeteners were drawn from a range of nutritive saccharide (sucrose, dextrose, fructose, allulose (D-psicose), palatinose (isomaltulose), and a sucrose–allulose mixture), nutritive polyol (maltitol, erythritol, mannitol, xylitol, sorbitol), non-nutritive synthetic (aspartame, acesulfame-K, sucralose) and non-nutritive natural sweeteners stevia (rebaudioside A), *luo han guo* (mogroside V). Sweetness intensities of the 16 sweeteners were compared with a sensory panel of 40 participants (*n* = 40; 28 females). Participants were asked to rate perceived sweetness intensity for each sweetener series across a range of concentrations using psychophysical ratings taken on a general labelled magnitude scale (gLMS). All sweeteners exhibited sigmoidal dose-response behaviours and matched the 'moderate' sweetness intensity of sucrose (10% *w*/*v*). Fructose, xylitol and sucralose had peak sweetness intensities greater than sucrose at the upper concentrations tested, while acesulfame-K and stevia (rebA) were markedly lower. Independent of sweetener concentration, the nutritive sweeteners had similar sweetness growth rates to sucrose and were greater than the non-nutritive sweeteners. Non-nutritive sweeteners on the other hand had higher potencies relative to sucrose, which decreases when matching at higher sweetness intensities. With the exception of dextrose and palatinose, all sweeteners matched the sweetness intensity of sucrose across the measured range (3.8–25% *w*/*v* sucrose) with fewer calories. Overall, the sucrose–allulose mixture, maltitol and xylitol sweeteners were most similar to sucrose in terms of dose-response behaviour, growth rate and potency, and showed the most potential for sugar replacement within the range of sweetness intensities tested.

**Keywords:** sweeteners; sugar reduction; psychophysical dose-response; sweetness growth rate; sweetness potency

#### **1. Introduction**

Sweetness is a key driver of liking in food products and a heightened liking for sweet tastes has been associated with increased intakes of foods with added sucrose [1]. The rising incidence of obesity and type-2 diabetes has been linked with excessive sucrose intake, and fuelled the need for reducing added sucrose in food products [2,3]. Countries such as the United Kingdom and Singapore have pledged to cut sucrose to either 5% free sugars in foods [4] or a 25% sucrose reduction from the current levels [5], namely through reducing added sucrose, using non-nutritive sweeteners and public health education. Sweetness intensity is associated with liking and reducing sucrose can negatively impact the hedonic appeal of a product and consumer acceptance of reformulated products, thereby limiting the widespread reduction of sucrose to achieve these public health goals. Non-nutritive sweeteners can be used to maintain product sweetness, while reducing the negative health impact of excessive sucrose intake, including increased body weight and risk of type-2 diabetes and cardiovascular diseases [6–8]. Sweet taste intensity has been shown to be associated with sucrose content of a product, but not with its energy content [9,10] thus creating an opportunity to reduce energy whilst matching sweet taste intensity and liking through the use of lower calorie sweeteners. As such, there has been a rising trend in the use of non-nutritive sweeteners such as sucralose and aspartame, in line with an increasing consumer demand for reduced-calorie foods. In the United States, 1 in 4 consumers include non-nutritive sweeteners in their diets based on a 24-hour diet recall [11]. This may be an effective strategy to improve public health, and a recent meta-analysis has shown that transition to lower-energy sweeteners in place of sucrose leads to reduced energy intake and body weight in both children and adults, as energy reductions associated with the intake of these sweeteners is often not fully compensated for during subsequent eating episodes [7].

Synthetic non-nutritive sweeteners like aspartame and sucralose are still the most widely consumed due to their low cost, quality of their sweet taste and calorie-free advantage, although their long-term metabolic impacts are still being investigated [8]. In addition to reduced sucrose and calories, in recent years there has been a rise in consumer demand for 'natural' and clean-label ingredients [11]. As a result, many food manufacturers have shifted towards the use of natural sweeteners such as plant-based glycoside extracts from stevia (rebaudioside, stevioside) and monk fruit (*luo han guo*; mogroside V). Alternative sugars such as the rare sugar D-psicose (allulose) and the isomerized sucrose isomaltulose (palatinose) have also gained interest due to their natural source, clean sweet taste and post-ingestive anti-glycaemic effects [12–14]. Polyol sweeteners are a group of sugar alcohols that have been reported to have excellent sweetness quality [15], fewer calories than sucrose and can also act as bulking agents in low sucrose foods, giving them an advantage over several non-nutritive sweeteners [16,17]. To date, the sweetness intensity and dose-response behaviour of many of these more recent sweeteners such as allulose, palatinose, a sucrose–allulose mixture and *luo han guo*, have not been compared alongside sucrose and other sweeteners.

Dose-response relationships have previously been reported for a range of different commercial sweeteners using, for example, the magnitude estimation or spectrum scaling method standardised with reference sucrose solutions [18–21]. This method obtains relative perceived sweetness intensity values but the comparison to other studies as is highly dependent on the reference solution, scaling method and extent of participant training [21,22]. More recently, psychophysical approaches have compared perceived sweetness intensity using ratings made on the general labelled magnitude scale (gLMS) [23–25]. This technique allows for relative comparisons of perceived sweetness intensity between sweeteners across concentrations, and can be useful for determining sweetening capabilities of a novel sweetener in relation to sucrose and other sweeteners [24].

The current study sought to characterise the perceived sweetness intensity of a wide range of different sweeteners to sucrose using a contemporary psychophysical approach. Based on the change in sweetness intensity across a range of concentrations, the dose-response behaviour of each sweetener was compared for their sweetness growth rate, sweetness potency, and potential to support calorie reduction at an equivalent sweetness intensity. These sweetness characteristics can be used as indices to gauge the sweetness and concentration-dependency of a sweetener in relation to sucrose. The sweetness growth rate is the Stevens' power law exponent in psychophysical terms, and is defined as the slope of the psychophysical relationship describing the rate of change in sweetness intensity with the rate of change in concentration [19]. Sweetness potency was defined as the ratio of the concentration of a sweetener to that of sucrose at an equivalent sweetness intensity [24], whereas the potential for calorie reduction is the caloric value of a sweetener compared to sucrose at matched sweetness intensities. These characteristics were examined across the selected sweeteners to compare sweetener suitability when attempting to reduce or replace sucrose.

#### **2. Materials and Methods**

#### *2.1. Materials*

A wide range of different sweeteners was selected to represent a diverse sample of common commercially available sweeteners. The sweeteners used in this study were sucrose (SIS, NTUC Fairprice Supermarket, Singapore), dextrose monohydrate (Suntop Enterprise, Singapore), fructose (Suntop Enterprise, Singapore), allulose (D-psicose; Matsutani Co., Osaka, Japan), palatinose (isomaltulose; Beneo, Singapore), xylitol (Roquette, Lille, France), sorbitol (Suntop Enterprise, Singapore), mannitol (Roquette, Lille, France), erythritol (iHerb, Perris, CA, USA), acesulfame-K (Celanese, Irving, Texas, USA), sucralose (Tate & Lyle, McIntosh, AL, USA), aspartame (Suntop Enterprise, Singapore), *luo han guo* extract (50.6% mogroside V; Hunan Huacheng Biotech Inc., Hunan, China) and stevia (68.0% rebaudioside A; Suntop Enterprise, Singapore). The sucrose–allulose mixture was prepared as a 1:1 blend of sucrose and allulose by weight. Table 1 summarises these sweetener properties including energy density, glycaemic index and bulk properties.


**Table 1.** Characteristics of the 16 sweeteners used.

‡ 1:1 sucrose−allulose mixture (weight basis). A indicates that the sweetener belongs to the respective category.

#### *2.2. Participants*

Forty healthy adult participants (12 males and 28 females; mean age: 26 ± 6 years) were recruited from the campus of the National University of Singapore (NUS) and surrounding areas. Participants were pre-screened for eligibility, basic taste sensitivity and recruitment criteria including being aged between 21 and 50 years old, normal weight (body mass index (BMI) 20–25 kg/m2), non-smoker, non-denture wearer, no self-reported sinus, taste or smell dysfunction, not currently following a special diet, no specific food dislikes, allergies or intolerances and not phenylketonuric, diabetic or pregnant. Eligible participants provided informed consent and were compensated for their time. This study (reference: 2017/00787) was approved by the Domain Specific Review Board of the National

Healthcare Group, Singapore and complies with the Declaration of Helsinki for research involving human subjects.

#### *2.3. Training and Test Procedure*

All participants underwent a total of 5 one-hour sessions on separate days, including 1 training session and 4 test sessions. During training, participants familiarised themselves with rating perceived sweetness intensity using the general Labelled Magnitude Scale (gLMS), based on a previously published approach [25]. Participants were asked to rate the overall taste intensity for seven imagined and/or recalled sensations described verbally including the warmth of lukewarm water and the pain from biting their tongue. Thereafter, participants were presented with four basic taste samples and asked to rate sweet (sucrose), salty (NaCl), sour (citric acid) and bitter (caffeine) to ensure they understood how to use the line scale and practice making ratings using the gLMS.

During each test session, participants rated the sweetness intensities of four sweetener sets, with eight samples for each sweetener set. The order of sweetener presented was randomised and balanced across participants and test sessions using a William's (Latin Square) design. The order of sample presentation within each sweetener set was also randomised. Participants rated a total of 16 sweetener sets over 4 test sessions. For each sweetener set, there is a water sample, six different concentrations of the sweetener (Table 2), and a warm-up sample with a duplicate concentration to one of the samples (sample 4/5; Table 1). The warm-up sample was presented at the beginning to reduce first order effects (data not included in analysis). The concentrations used in this study are by weight basis (% *w*/*v*) presented in Table 2, and the same concentrations expressed in molarity (mmol/L) are provided in the supplementary material (Table S1). For ease of interpretation of the calorie reduction potential and application to product re-formulation, the dose-response behaviour of the sweeteners was expressed on a weight basis in the current study, in line with previous reports [18,21,24]. Therefore, discussions made in this study are based on weight of sweeteners and not by molarity.


**Table 2.** Concentrations tested for each of the sweetener set by weight basis (% *w*/*v*).

‡ 1:1 sucrose–allulose mixture (weight basis).

All data were collected using Compusense Cloud software as part of the Compusense Academic Consortium (Compusense Inc., Guleph, ON, Canada), in sensory booths under red lights which conform to international standards for the design of test rooms [26]. Participants were instructed to take the sample (15 mL) in their mouth, hold it for 5 s, and rate the sweetness intensity before expectorating. Between samples, participants were instructed to rinse their mouth with filtered water during the mandatory 45-second inter-stimulus interval, to reduce carryover between samples. Solutions were prepared at least 24 h prior to sensory testing using filtered water and stored at refrigeration temperature. The concentration ranges chosen were based on previously published

results for each sweetener [23,27,28], and to reflect the sweetness intensities encountered in commercial food and beverage products. Preliminary testing was done to confirm that the sweetness intensities rated for each sweetener were comparable to one another.

#### *2.4. Psychophysical Scaling*

Perceived sweetness intensity was rated using a general labelled magnitude scale (gLMS) [22,29,30]. The scale is partitioned by descriptors: no sensation (0), barely detectable (1.5), weak (6), moderate (17), strong (35), very strong (52) and strongest imaginable sensation (100). Individual scaling behaviours for gLMS ratings were standardized within participants using a previously published weight comparison modality matching task [25,31]. All participants were asked to make intensity ratings using the gLMS across a series of different weight stimuli while holding the container on the palm of their non-dominant hand. There was a significant correlation between the overall sweetness ratings and overall mean heaviness ratings (*r* = 0.472, *p* < 0.05). Assuming the intensity ratings of sweetener samples and the heaviness of the bottles were unrelated, the significant correlation indicated that gLMS ratings were due to individual scale-use rather than differences in sweeteners, and thus required standardization across participants. For each participant, a personal standardization factor was obtained using the grand mean for heaviness ratings across weights and participants divided by the individual's average heaviness ratings. The sweetness intensity rankings for each participant were then multiplied by their individual personal standardization factor to correct for idiosyncratic differences in scale use.

#### *2.5. Mathematical Modelling and Data Analysis*

#### 2.5.1. Dose-Response Curves

Dose-response curves were fitted using the software Origin Pro 8.1 (OriginLab, Northampton, MA, USA) with the Hill equation for sigmoidal curves:

$$R = R\_{\rm min} + \frac{R\_{\rm max} - R\_{\rm min}}{1 + 10^{(\log \text{EC}\_{50} - \mathbb{C}) \times \text{HillSlope}}} \tag{1}$$

where R is the predicted sweetness intensity, and C is the sweetener concentration expressed in % *w*/*v*. Rmin is the minimum sweetness which was constrained to zero, and Rmax is the predicted maximum sweetness achievable. The midway point between Rmin and Rmax is EC50, and the slope of the linear portion of the model is the Hill slope [23]. The fitted parameters are summarised in the supplementary material (Table S2).

#### 2.5.2. Sweetness Growth Rate

Dose-response curves were also converted to log sweetness intensity vs. log concentration plots, which were originally derived from the power law function *R* = kCn in the linear form:

$$
\log \mathbf{R} = n \log \mathbf{C} + \log \mathbf{k} \tag{2}
$$

where R is the predicted sweetness intensity, C is the sweetener concentration expressed in % *w*/*v*, *n* is the sweetness growth rate (slope of the line or Stevens' power law exponent), and k is the constant (intercept). The sweetness growth rate provides an overall average index of the rate of change for sweetness intensity with change in sweetener concentration. A sweetener with a steep slope (>1) increases in their perceived intensity at a rate that is faster than changes in concentration, whereas for flatter slopes (<1), greater increases in sweetener concentration are required to produce an equivalent change in sweetness intensity. The log k (intercept) values also refer to the log sweetness intensity of the sweetener at a concentration of 1% *w*/*v*.

#### 2.5.3. Sweetness Potency

Sweetness potency is the ratio of the concentration of sucrose to that of a sweetener at equivalent sweetness intensities (Equation (3)). A sweetness potency of >1 indicates that a smaller concentration of a sweetener is required to achieve the same sweetness intensity at a particular sucrose concentration and, therefore, the sweetener could be considered as 'more potent' than sucrose. Sweetness potency is often quoted as a single value e.g., 'acesulfame-K is 120 times sweeter than sucrose', however this value should always be reported with the concentration of sucrose at which it was calculated, since sweeteners often have different sweetness growth rates to sucrose.


#### 2.5.4. Statistical Analysis

A two-way analysis of variance (ANOVA) was run to confirm absence of first-order and carryover effects. A one-way repeated measures ANOVA analysis was used to test the effect of sweetener type and effect of concentration and statistical significance was set at 5% (α = 0.05). Post hoc pairwise comparisons, using Bonferroni corrections, were used to compare differences in sweetness intensity scores across sweeteners (16 levels) and concentration of sweeteners (6 levels) using the statistical analysis software SPSS (IBM SPSS Statistics for Windows, Version 22.0, IBM Corporation, Armonk, NY, USA).

#### **3. Results**

The dose-response for all sweeteners are illustrated on semi-log curves (Figure 1A–C) and fitted with the Hill equation (Equation (1)) with *<sup>r</sup>*<sup>2</sup> ≥ 0.95 for all sweeteners. The fitting parameters are listed in Table S2. Repeated-measures ANOVA confirmed that all sweeteners exhibited a concentration dose-dependency for sweetness intensity (*F*5,39 = 142.12, *p* < 0.001). Sweetener type had a significant effect on sweetness intensity as concentrations increased (*F*15,39 = 18.05, *p* < 0.001) and this was confirmed as a significant interaction between concentration and sweetener type (*F*75,39 = 4.20, *p* < 0.001).

#### *3.1. Concentration Dose-Response of Sweeteners*

The sweetness intensity of sucrose ranged from 'barely detectable' (3) to 'strong' (33) on the gLMS for the concentration range of 3.8 to 25% *w*/*v*. Nutritive saccharide and polyol sweeteners sucrose, dextrose, allulose, palatinose, maltitol, sorbitol, mannitol, xylitol and erythritol exhibited sigmoidal dose-response functions. By contrast, fructose displayed a more linear response and had a higher sweetness intensity than sucrose and other nutritive sweeteners, across all sucrose concentrations (Figure 1A,B). The sucrose–allulose mixture, maltitol and xylitol had dose-response curves closely matched to sucrose within the range of 3.8 to 25% *w*/*v* sucrose. The dose-response curve for xylitol was similar to sucrose at lower concentrations but had higher sweetness intensity above 11.7% *w*/*v*. Allulose was perceived as less sweet than sucrose at equivalent concentrations, but when allulose and sucrose were blended in a 1:1 mixture, this blend achieved a near identical dose-response curve to sucrose. Palatinose required the highest concentration to match the sweetness intensity of sucrose, and only produced a noticeable rise in sweetness intensity as the concentration went above 10% *w*/*v*. Dextrose, erythritol, sorbitol and mannitol all had lower sweetness intensities than sucrose across the concentration range tested.

Non-nutritive sweeteners exhibited sigmoidal dose-response functions (Figure 1C) and stevia (rebA) and acesulfame-K had flatter responses at low and high concentrations, where increased concentration produced smaller increments in perceived sweetness intensity. In addition, maximum sweetness for these sweeteners peaked below sucrose at 'moderate' (25). Sucralose had a higher peak sweetness intensity (35) compared to sucrose (33) at the highest concentration, and was higher than the other non-nutritive sweeteners across equivalent concentrations. Aspartame and *luo han guo* both had similar peak sweetness to sucrose, and their dose-response curves were similar to each other. Their sweetness intensities were weaker than stevia (rebA) at low concentrations (0.01–0.1% *w*/*v*) but stronger at higher concentrations (>0.1% *w*/*v*) when the sweetness intensity of stevia (rebA) plateaued.

**Figure 1.** Sweetness intensity with concentration for (**A**) saccharide, (**B**) polyol and (**C**) non-nutritive sweeteners (sucrose is plotted using the secondary x-axis below (0.1–100% *w*/*v*)).

#### *3.2. Comparison of Sweetness Growth Rates across Sweeteners*

The sweetness growth rate is represented by the slope of the log-log sweetness intensity concentration curves (% *w*/*v* basis) (Figure 2 and Table 3). Sucrose had a sweetness growth rate of 1.31 whereas saccharide sweeteners (dextrose, palatinose, fructose, allulose, sucrose–allulose mixture,) had sweetness growth rates >1, ranging from 1.08 (fructose) to 1.46 (sorbitol). The bulk polyol sweeteners (sorbitol, xylitol, mannitol and erythritol) had sweetness growth rates with similar slopes to sucrose (~1.3–1.4), indicating a similar growth rate to sucrose such that changes in concentration produce similar changes in sweetness intensity. Palatinose and fructose yielded much flatter sweetness growth rates (slopes ≈ 1) amongst the nutritive sweeteners, with 1.10 and 1.08 respectively. By contrast, non-nutritive sweeteners had compressed sweetness growth rates < 1, ranging from 0.65 (sucralose) to 0.84 (aspartame).

**Figure 2.** Log sweetness intensity vs. log concentration for 16 sweeteners.

**Table 3.** Slope and y-intercept values of linear fit between log sweetness intensity and log concentration (% *w*/*v*).


‡ 1:1 sucrose–allulose mixture (weight basis).

#### *3.3. Sweetness Potency of Sweeteners Relative to Sucrose*

Sweetness potency as well as the concentration of sweetener required to achieve equivalent sweetness intensity to sucrose concentrations at 5%, 10% and 15% *w*/*v* are summarised in Table 4. Saccharide and polyol sweeteners had sweetness potencies <1, with the exception of xylitol (at 15% *w*/*v* sucrose) and fructose. Sweetness potency values for allulose increased from 5% to 15% *w*/*v* sucrose respectively, whereas the sweetness potency for maltitol, xylitol and sucrose–allulose mixture were closer to sucrose across sucrose concentrations, emphasising the similarity of their dose-response functions (Figure 1). Non-nutritive sweeteners had decreasing sweetness potencies at increasing sucrose concentrations. Sucralose has the highest sweetness potency across all sweeteners, but also the largest decline, from sweetness potency of 521 at 5% *w*/*v*, to 201 at 15% *w*/*v* sucrose. Aspartame had the smallest decline in sweetness potency among the non-nutritive sweeteners at higher sucrose concentrations.


**Table 4.** Concentrations matching for equi-sweetness and sweetness potency of 15 sweeteners to 5%, 10% and 15% *w*/*v* sucrose.

‡ 1:1 sucrose–allulose mixture (weight basis).

#### *3.4. Potential for Calorie Reduction across Sweeteners*

Figure 3 shows the caloric value across the different nutritive sweeteners at sweetness intensities ranging from weak (6) to strong (35). The equivalent sweetness intensity to sucrose per unit calorie provides a summary of the calorie saving potential across the different sweeteners. With the exception of dextrose and palatinose, all of the other nutritive sweeteners enable calorie saving at an equivalent perceived sweetness intensity to 10% sucrose (indicated by red line on Figure 3). Allulose and erythritol have the lowest energy densities (0.2 kcal/g) and can achieve an equivalent sweetness intensity to 10% sucrose with very few calories (~95% reduction). For example, a product with 10% *w*/*v* sucrose could potentially be reduced from 40 kcal to <5 kcal/100ml by substituting with allulose or erythritol. Mannitol, sucrose–allulose mixture, xylitol, fructose, maltitol and sorbitol provide about 5–20 kcal/100 mL savings in terms of energy required to achieve equivalent sweetness to 10% sucrose.

**Figure 3.** Energy content (kcal/100 mL) of nutritive saccharide and polyol sweeteners to achieve sweetness intensities ranging from weak (6) to strong (35).

#### **4. Discussion**

In order to support sugar reduction, sweeteners must first match the sweetness intensity of sucrose across the range of perceived intensities commonly encountered in foods and beverages. From a public health perspective, the reduction in sucrose should also support calorie reduction while maintaining consumer appeal beyond sensory-matching perceived sweetness. In addition to their sweetening capacity, sweeteners that can confer additional functionality such as acting as bulking agents, supporting clean labelling or providing an additional anti-glycaemic effect are also highly desirable. A wide variety of sweeteners are currently available and the present study sought to evaluate the sweetening capabilities of these sweeteners in comparison to sucrose based on their dose-response behaviour, sweetness growth rate, sweetness potency and potential calorie savings at equal sweetness intensities.

All sweeteners exhibited sigmoidal dose-response behaviours although fructose displayed a more linear response across the concentration range tested. This sigmoidal relationship between concentration and perceived intensity is the result of the binding kinetics of the sweetener molecules to taste receptors, which plateaus at higher concentration when receptors become saturated [32,33]. From the dose-response curves, all sweeteners were found to match the perceived sweetness intensity of a 10% *w*/*v* sucrose solution, which represents a 'moderate' sweetness associated with 10% sugar that is frequently found in many commercially available sweetened and carbonated beverages (e.g., Arizona Ice Tea 10.1 g/100 mL) [34]. This aligns with similar findings from other studies, where the sweetness intensity of ~10% *w*/*v* sucrose was also found to be of 'moderate' intensity with rating scores approximately 15 to 20 on the gLMS [23–25]. Interestingly, the perceived sweetness intensity of the sucrose–allulose mixture (1:1) was nearly identical to that of sucrose by weight basis, although allulose on its own had consistently lower sweetness intensity than sucrose across the concentrations tested. Previous research has demonstrated that the sweetness intensity of binary

mixtures of sweeteners is often an intermediate of the two compounds when tasted alone and at the same total molarity as the mixtures [35,36]. Since the weight of allulose (monosaccharide) is half that of sucrose (disaccharide), the dose-response behaviour of the sucrose–allulose mixture was expected to be between that of sucrose and allulose when expressed in terms of total molarity. Other nutritive sweeteners with smaller molecular weights than sucrose, such as fructose or xylitol, would be relatively even less sweet than sucrose on a molarity basis as compared to weight basis [37]. Nonetheless, for purposes of sweetener application to product re-formulation and interpretation of the calorie reduction potential, the dose-response behaviour of the sweeteners were expressed on a weight basis in the current study, in line with previous reports [18,21,24]. Fructose, xylitol and sucralose were the only sweeteners which had greater peak sweetness intensities than sucrose at the highest concentrations tested, and this has previously been demonstrated across a range of previous studies [23,38,39]. This suggests that 'high-intensity' sweeteners such as aspartame and sucralose may be more accurately described as 'high-potency' sweeteners, as proposed previously by Antenucci and Hayes [23]. The peak sweetness intensities for the non-nutritive sweeteners acesulfame-K and stevia (rebA) were markedly lower than that of sucrose, reaffirming that these high-intensity sweeteners are not necessarily higher in perceived sweetness intensity than sucrose. Further concentration increments of acesulfame-K, stevia (rebA) and sucralose have been shown to produce a further decrease in sweetness intensity [23,24], which was likely due to bitter taste antagonism at higher concentration among these sweeteners [40]. This decrease in sweetness was not observed for any sweeteners at the concentrations used in the current study. The low peak sweetness of acesulfame-K and stevia (rebA) could limit their use in foods where higher sweetness intensities are required. Nevertheless, it is difficult to determine the true peak sweetness achievable unless a plateau in sweetness can be clearly observed [21], on the condition that the intensity scaling method is not limited by a ceiling effect [29,41]. The concentration ranges for the sweeteners were selected prior to the study based on literature and preliminary experiments, although we acknowledge that further concentration increments would likely result in greater perceived sweetness for some sweeteners. In this case, comparing the sweetness growth rate would be a better indicator of the dose-response trajectory rather than the peak sweetness of each sweetener, to understand whether they are likely to match or surpass the sweetness of sucrose.

The sweetness growth rate is the Stevens' power law exponent or slope of the log relationship between changing concentration and the perceived sweetness. It should be noted that the sweetness growth rate obtained in this study is a product of the concentration ranges from which they are derived. These range effects mean that sweetness growth rates can change to be higher or lower depending on the range of concentrations tested, and a higher sweetness growth rate is obtained with a smaller concentration range [42]. Sucrose had a sweetness growth rate of 1.3 which is consistent with previous findings which reported sweetness growth rates of 1.15 to 1.3 [18–20,24]. The sweetness growth rates of sucralose, stevia (rebA), dextrose and mannitol were also found to be comparable to those previously reported [18,24] and collectively these findings highlight that sucrose, and other nutritive sweeteners exhibit sweetness growth rates greater than non-nutritive sweeteners. This appears counterintuitive since only a small quantity of non-nutritive sweetener is required to impart an intense sweetness, which should be perceived as a higher growth rate. However, sweetness growth rates are based on sweetness intensity changes *per* unit log-concentration, and in relative terms greater quantities of non-nutritive sweeteners are required to achieve a proportional increment in perceived sweetness. This may also be due to the emergence of bitter side-tastes and taste–taste antagonism at higher concentrations [40], or different binding mechanisms across sweeteners, which often remain poorly understood [43]. Sweeteners with lower sweetness growth rates to sucrose are capable of matching the sweetness intensity, albeit over a limited concentration range. The implication is that sweetness growth rates should be considered when estimating the predicted sweetness intensity of a sweetener at concentrations beyond those reported in the dose-response curves. For example, sucralose matched the sweetness intensity of sucrose to an upper concentration of 25% *w*/*v*, but displayed a smaller growth rate, suggesting that the peak sweetness intensity for sucralose would be lower than that of sucrose. This is further supported by the flatter dose-response curve for sucralose at higher concentrations (0.172–0.350% *w*/*v*).

Sweeteners with the same growth rate as sucrose will increase in perceived sweetness intensity with equal increases per unit concentration. Sweetness potency or relative concentration of sweetener required to produce an equi-intense sweetness to sucrose would, therefore, be similar across a range of different concentrations. By contrast, sweeteners with growth rates that differ significantly from sucrose would have sweetness potencies that vary with sucrose concentration, as demonstrated previously [19,20,24,44]. While sweetness growth rate and sweetness potency are closely related indices, sweetness potency is more often used as a quick indication of the quantity required to achieve an equi-intense sweetness to sucrose at a given concentration. Non-nutritive sweeteners have growth rates significantly lower than sucrose, and therefore their sweetness potency is also highly concentration-dependent. The sweetness potency values reported for non-nutritive sweeteners in the current trial were not fully consistent those reported previously. For example, aspartame was found to be 128 and 185 times more potent than sucrose at 5 and 10% sucrose equivalence by Tunaley, Thomson and McEwan [45], and Cardello et. al. [20] respectively, as compared to the 173 (5% *w*/*v*) and 121 (10% *w*/*v*) found in our study. Differences were also found for sweetness potencies of stevia (rebA) and sucralose [24]. Sucrose–allulose mixture, xylitol and maltitol have sweetness potencies closest to 1, indicating that the quantities required to achieve an equivalent sweetness intensity on a weight basis are similar to sucrose. This is an important consideration when the replacement sweetener is also required to substitute some of the bulking properties of the removed sucrose. When calorie reduction without the addition of bulk is the main goal of sucrose reduction, low calorie and/or high potency sweeteners may be more effective, particularly among certain products (i.e., beverages) as lower concentrations are required to match sweetness intensity.

With the inclusion of several low-calorie nutritive sweeteners in the study, it is still possible to achieve calorie reduction while maintaining sweetness, even when sweetness potency was not equivalent or higher than sucrose. With the exception of dextrose and palatinose, the nutritive sweeteners profiled supported reductions in total calories to varying extents while meeting the equivalent sweetness intensity of a 10% sucrose solution. Allulose and erythritol in particular have the lowest calories at a sweetness intensity equivalent to 10% sucrose, and could be used to support substantial calorie savings. When allulose was mixed 1:1 to partially replace sucrose, the sucrose–allulose mixture showed very similar sweetening properties to sucrose, while supporting significant reduction in overall calories. Considering the sweetness intensity, sweetness growth rate, sweetness potency and potential calorie reduction together, the sucrose–allulose mixture, maltitol and xylitol were most similar to sucrose, across the concentrations studied. All three sweeteners can provide bulk, support a clean label, reduce total calories for equivalent sweetness intensity and in the case of sucrose–allulose mixture, also impart an additional anti-glycaemic effect post-ingestively [12,14]. When selecting the appropriate sweetener for use in sugar reduction, the sensory, physical, nutrient and metabolic impact of the selected sweetener should be considered, and in some cases sweeteners will have desirable characteristics for some properties but not others. For example, palatinose has an anti-glycaemic benefit which is desirable for products that support the management of glucose homeostasis, but it is required at a greater concentration and energy content to achieve an equi-intense sweetness intensity to sucrose [13]. Non-nutritive sweeteners are calorie-free but may have certain undesirable side-taste attributes, especially at higher concentrations [23,24,40,46] which may limit their usage. With these factors considered, it may be judicious to blend sweeteners with sucrose to optimise the sensory profile and sweetening capacity, and compromise on some elements of the nutrient or metabolic profile. Results from the current study demonstrate that blends like the sucrose–allulose mixture provide encouraging results with excellent sweetness characteristics in line with sucrose, at a fraction of the calories and a potential post-consumption anti-glycaemic benefit.

Findings from the current study are aligned with previously reported differences in sweetness dose–response, growth rate and potency, although some subtle differences were observed in the absolute values reported. These are likely to be due to differences in methodological approach, individual variability, sweetener source, matrix effects, concentration range used, pH and temperature [20,21,23,25,47]. Our findings are most closely aligned with those previously reported by Antenucci and Hayes which were collected using the same standardised gLMS approach to rate sweetness intensity. This approach minimises ceiling effects and produced comparable intensity ratings for many of the same sweeteners [23]. There is currently no official standardised approach to quantify the perceived sweetness intensity of a sweetener, although the comparability of results would be greatly enhanced if future efforts adopt a consistent objective approach, such as used in the current and previous studies [23–25].

In choosing to focus on sweetness intensity alone, we have not accounted for other temporal and qualitative taste differences between the sweeteners reported elsewhere [46]. In addition, perceived sweetness intensity rated in water does not account for matrix effects or taste–taste interaction that would occur when these same sweeteners and concentrations are used in foods and beverages [48,49]. The current findings present an overview of the psychophysical dose-response behaviour of a wide range of different sweeteners, and provides guidance on the similarity of various sweeteners to sucrose and the likely calorie savings that could be achieved if they are used to reduce or replace sugar. Future research should aim to extend this further by profiling the temporal and qualitative differences between sweeteners and characterising their performance in food and beverages.

#### **5. Conclusions**

The current paper characterized the psychophysical dose-response behaviour of 16 sweeteners and identified differences in the peak sweetness, sweetness potency and sweetness growth rate. Sucrose—allulose mixture, maltitol and xylitol exhibited similar psychophysical behaviours to sucrose in terms of peak sweetness intensity, sweetness growth rate and sweetness potency, and showed the greatest potential to match the sweetness of sugar, for significantly fewer calories. Non-nutritive sweeteners offer significant calorie savings, but had lower peak sweetness intensities and lower sweetness growth rates, which may not limit their ability to match sweetness intensity over a wider range of sucrose concentrations. Differences in the psychophysical relationships identified in the current paper should be considered when selecting sweeteners to support sucrose reduction or replacement, and offer significant opportunities to match the perceived sweetness of sugar, while supporting energy density reductions.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2072-6643/10/11/1632/ s1, Table S1: Concentrations tested for each of the sweetener tastant series in molarity (mmol/L), Table S2: Fitting parameters for dose-response of sweeteners using the Hill equation.

**Author Contributions:** C.F. conceived and planned the experiments. M.W. and V.T. collected the data. M.W. and C.F. analysed the data and prepared the manuscript. All authors approved the final version of the manuscript.

**Funding:** Research supported by Biomedical Science Institute Strategic Positioning Fund Grant (G00067; BMSI/13-80048C-SICS: (Sensory Nutritional Science; PI: C. Forde).

**Acknowledgments:** We would like to thank the following companies for providing some of the sweetener samples for this study: Roquette, Matsutani and Beneo. The Compusense Cloud software was also used as part of the Compusense Academic Consortium (Compusense Inc., Guleph, ON, Canada).

**Conflicts of Interest:** C.F. has received reimbursements for speaking at conferences sponsored by companies selling food ingredients and nutritional products, and currently serves on the scientific advisory council of a commercial ingredient manufacturer. None of these organizations had any influence on the design or interpretation of the findings in the current study. All other authors declare no conflicts of interest.

#### **References**

1. Drewnowski, A.; Mennella, J.A.; Johnson, S.L.; Bellisle, F. Sweetness and Food Preference–3. *J. Nutr.* **2012**, *142*, 1142S–1148S. [CrossRef] [PubMed]


© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
