Skip to main content
NCBI home page
The Journal of Nutrition logoLink to The Journal of Nutrition
. 2021 Jun 10;151(9):2843–2851. doi: 10.1093/jn/nxab160

Data-Driven Clustering Approach to Derive Taste Perception Profiles from Sweet, Salt, Sour, Bitter, and Umami Perception Scores: An Illustration among Older Adults with Metabolic Syndrome

Julie E Gervis 1, Kenneth K H Chui 2, Jiantao Ma 3, Oscar Coltell 4,5, Rebeca Fernández-Carrión 6,7, José V Sorlí 8,9, Rocío Barragán 10,11, Montserrat Fitó 12,13,14, José I González 15,16, Dolores Corella 17,18,✉, Alice H Lichtenstein 19
PMCID: PMC8861513  PMID: 34114008

ABSTRACT

Background

Current approaches to studying relations between taste perception and diet quality typically consider each taste—sweet, salt, sour, bitter, umami—separately or aggregately, as total taste scores. Consistent with studying dietary patterns rather than single foods or total energy, an additional approach may be to study all 5 tastes collectively as “taste perception profiles.”

Objective

We developed a data-driven clustering approach to derive taste perception profiles from taste perception scores and examined whether profiles outperformed total taste scores for capturing individual variability in taste perception.

Methods

The cohort included 367 community-dwelling adults [55–75 y; 55% female; BMI (kg/m2): 32.2 ± 3.6] with metabolic syndrome from PREDIMED-Plus, Valencia. Cluster analysis identified subgroups of individuals with similar patterns in taste perception (taste perception profiles); quantitative criteria were used to select the cluster algorithm, determine the optimal number of clusters, and assess the profiles’ validity and stability. Goodness-of-fit parameters from adjusted linear regression evaluated the individual variability captured by each approach.

Results

A k-means algorithm with 6 clusters best fit the data and identified the following taste perception profiles: Low All, High Bitter, High Umami, Low Bitter & Umami, High All But Bitter and High All But Umami. All profiles were valid and stable. Compared with total taste scores, taste perception profiles explained more variability in bitter and umami perception (adjusted R2: 0.19 vs. 0.63, respectively; 0.40 vs. 0.65, respectively) and were comparable for sweet, salt, and sour. In addition, taste perception profiles captured differential perceptions of each taste within individuals, whereas these patterns were lost with total taste scores.

Conclusions

Among older adults with metabolic syndrome, taste perception profiles derived via data-driven clustering may provide a valuable approach to capture individual variability in perception of all 5 tastes and their collective influence on diet quality. This trial was registered at https://www.isrctn.com/ as ISRCTN89898870.

Keywords: taste, individual differences, cluster analysis, sweet, salt, sour, bitter, umami, data-driven

See corresponding editorial on page 2503.

Introduction

Accumulating evidence suggests that perception of the 5 basic tastes—sweet, salt, sour, bitter, and umami—plays an influential role in diet quality and diet-related health outcomes. Taste perception differs substantially among individuals, partially attributable to genetic polymorphisms in taste-related genes (1) as well as physiological differences in the gustatory system (2), sex and age (3), and lifestyle habits such as smoking and alcohol use (4, 5). Evolutionarily, perception of all 5 tastes has been important for ensuring adequate intake of macro- and micronutrients (sweet for carbohydrates, umami for proteins or amino acids, salt for minerals) and minimizing the risk of ingesting harmful substances (bitter for toxins, sour for spoiled foods) (6, 7). In modern day, taste continues to be a predominant driver of food and beverage choices (8); hence, individual differences in perception of the 5 tastes have been suggested as key individual-level drivers of food intake and, ultimately, diet quality (7, 9).

To date, much of what is known about the relations between taste perception and diet quality comes from studies that assessed each taste separately, with the majority of work focusing on bitter. In these single taste studies, individuals are often classified as non-tasters, medium tasters, or supertasters, either based on their phenotypic perception of the bitter compounds 6-n-propylthiouracil (PROP) or phenylthiocarbamide (PTC), or their genotype at the Taste 2 Receptor Member 38 (TAS2R38) bitter taste receptor gene (10, 11). When related to diet, taster status in adults has been associated with intakes of vegetables (12), sweets (13, 14), and alcohol (15), although not consistently (16, 17). Recently, studies have also begun examining whether phenotypic and genotypic variability in perception of tastes other than bitter are also related to dietary behaviors. In general, the data suggest that perception of sweet (18, 19), salt (20), sour (21), and umami (22, 23) tastes likewise play a role in determining food intake (22). On this basis, some have begun studying perception of multiple tastes as an aggregate measure or “total taste score” (22, 24). This approach has yielded valuable insights into how total taste perception may relate to obesity (24, 25) and food intake (22); however, aggregating data on multiple tastes may result in lost information about potential within-individual differences in perception of the single tastes. For example, for 2 people with similar total taste scores, one may have high perception of bitter and low perception of umami, whereas another may have low perception of bitter and high perception of umami.

Consistent with the shift towards studying dietary patterns rather than single foods and nutrients or total energy intakes (26), an additional approach to studying taste perception may be to study patterns in taste perception, or “taste perception profiles.” Similar to dietary pattern analysis, taste perception profiles can be derived using cluster analysis to capture perception of all 5 single tastes collectively. While a common criticism of cluster analysis is its reliance on subjective criteria to make key decisions (e.g., the selection of a cluster algorithm and the optimal number of clusters) (27, 28), this may be overcome by relying on quantitative criteria to inform these decisions, thereby using a more repeatable and transparent data-driven approach (29, 30). Our aims were to develop a data-driven clustering approach to derive taste perception profiles from sweet, salt, sour, bitter, and umami perception scores and to compare the ability of these taste perception profiles against total taste scores for capturing individual variability in perception of each of the 5 tastes among older subjects with metabolic syndrome.

Methods

Study design and participants

We carried out a cross-sectional study in PREDIMED (PREvención con DIeta MEDiterránea)-Plus Valencia participants (31). PREDIMED-Plus is an ongoing, multicenter, randomized controlled trial testing the effect of an energy-restricted Mediterranean diet with physical activity intervention for the primary prevention of cardiovascular disease, compared with a control (“usual care”) group. The trial was registered in 2014 at the International Standard Randomized Controlled Trial registry as number 89898870 (http://www.isrctn.com/ISRCTN89898870). Participants were recruited to be community-dwelling older women (60–75 y) and men (55–75 y), with elevated BMI (kg/m2; ≥27 and ≤40) and ≥3 criteria of the metabolic syndrome. Of the 381 eligible participants at the University of Valencia center who completed the taste perception substudy at baseline (24), 2 were excluded for implausible total energy intakes (>4000 and >4500 kcal/d for women and men, respectively) and 12 for incomplete data for some taste perception tests, resulting in a sample size of 367 participants for this secondary analysis.

All participants provided written informed consent in accordance with ethical standards of the Helsinki Declaration. Study protocols were approved by the Human Research Ethics Committee of Valencia University, and the current secondary analysis was approved by the Tufts University Institutional Review Board.

Demographic, clinical, and lifestyle measures

Demographic and lifestyle measures, including sex, age, smoking status and history, and medication use, were collected via validated questionnaires (24). In addition, a validated REGICOR (REgistre GIroní del COR) questionnaire was used to estimate total physical activity in metabolic equivalents (METs) per week (32) and a validated food frequency questionnaire was used to estimate habitual food intake over the past year (33). Anthropometric measures, including weight and height and blood pressure were measured by trained staff in accordance with PREDIMED-Plus operations protocol (24). BMI was calculated from weight and height as kg/m2. Biochemical parameters, including fasting blood glucose, HDL-cholesterol, and triglyceride concentrations were measured after a 12-h overnight fast, as previously described (24). Type 2 diabetes status was defined as having a previous clinical diagnosis of Type 2 diabetes, use of hypoglycemic medications, or fasting blood glucose ≥126 mg/dL (in duplicate).

Taste perception assessment

Taste perception was assessed by trained technicians in a standardized environment, as previously reported in detail (24). Sucrose, NaCl, citric acid, PTC, and monopotassium l-glutamate (MPG) tastants (all from Sigma-Aldrich, Milan, Italy) were used to represent sweet, salt, sour, bitter, and umami tastes, respectively. Tastants were diluted in distilled water and provided in liquid form, except for PTC, which was provided on Whatman No. 1 filter paper (Whatman plc, Little Chalfont, UK) that had been dipped into the solution heated close to boil, dried, and cut into strips. Tastant solutions were placed on the tongue using a wooden cotton bud applicator (sucrose, NaCl, citric acid, MPG) or filter paper (PTC). Participants were instructed to mix the tastant solutions with their saliva for 30 s, remove, then use a 6-point categorical scale to rate the intensity of each solution: 0 = “no taste,” 1 = “weak,” 2 = “moderate,” 3 = “strong,” 4 = “very strong,” and 5 = “extremely strong.” To minimize contextual effects, tastants were provided in a standardized order (sucrose, NaCl, citric acid, MPG, PTC). While multiple concentrations were used for each tastant in the parent study (24), since the individual variability in scores was highest for the most concentrated solutions (400 mM sucrose, 200 mM NaCl, 34 mM citric acid, 200 mM MPG, 5.6 mM PTC), only those data were used in this analysis to derive the taste perception profiles. A “total taste score” was also constructed by summing the perception scores for all 5 tastants into a single aggregate measure (potential range: 0 to 25).

Data-driven clustering approach

The data-driven clustering approach utilized cluster analysis and quantitative criteria to derive the taste perception profiles. Cluster analysis was used to identify distinct subgroups (“clusters”) of individuals with similar patterns in perception scores across all 5 tastes. The goal was to identify the set of clusters (cluster solution), which minimized the variation among individuals in each cluster (within-cluster variation) and maximized the variation between individuals in each cluster (between-cluster variation). Each derived cluster was considered to represent a group of individuals with a common taste perception profile (e.g., a cluster of individuals with low perception of sweet, salt, sour, and umami and high perception of bitter would represent a “High Bitter” profile).

Two cluster algorithms and a range of number of clusters (Ck) were considered for the derivation. The 2 potential cluster algorithms were k-means (KCA), which is an optimization algorithm that derives a prespecified, nonoverlapping Ck by minimizing total within-cluster variation (Wk), and Ward's minimum variation (Ward's D), which is an agglomerative hierarchical clustering algorithm that derives any Ck by minimizing the Wk as observations are merged into a single hierarchical structure. Potential Ck included a range from 2 to 10. Quantitative criteria were used to select the cluster algorithm, determine the optimal Ck, derive the taste perception profiles, and assess their internal validity and stability, in a stepwise manner, as follows.

Step 1: Identify a reproducible cluster algorithm

Split half cross-validation (29) was performed to examine the reproducibility of cluster solutions derived by the 2 cluster algorithms that were considered for the taste perception profile derivation: KCA (R package::command is listed in italics; stats::kmeans) and Ward's D (stats::clust, method = ward.D). Each algorithm was applied to the data with a Ck from 2 to 10. For each algorithm and Ck, the data were split randomly in half into train and test sets. Cluster analysis was run on the train set to generate a train set solution, observations in the test set were assigned to the most similar clusters in the train set solution via k-nearest neighbors (class::knn) to generate a test set solution, then cluster analysis was run on the test set to generate a second test set solution. Agreement and association between the 2 test set solutions were measured using Hubert and Arabie's Adjusted Rand Index (ARI) (fossil::adj.rand.index) and Cramer's V statistic (CramV) (rcompanion::cramerV), where values closer to 1 indicated stronger agreement and association, hence, reproducibility, respectively. Finally, the train and test sets were switched to generate a second set of estimates. The entire protocol was repeated 10 times to generate a total of 20 estimates of each index, at each Ck, by each algorithm. All estimates by each algorithm were aggregated and the algorithm that derived more reproducible cluster solutions, hence taste perception profiles, overall, was selected as the best fit for the data.

Step 2: Determine the optimal number of clusters in the data

Three criteria were used to determine the Ck that minimized the within- and maximized the between-cluster variation. First, cluster solutions were derived with a Ck from 2 to 10 with the selected cluster algorithm. Using the Elbow method, the Wk was plotted as a function of Ck to identify the Ck beyond which additional clusters did not substantially reduce the Wk, identified as an “elbow” in the plot. The average silhouette and gap statistics were also computed and plotted as functions of the Ck to measure the quality of separation between clusters as the ratio of minimum between- to maximum within-cluster variation (cluster::silhouette) and to identify the Ck that minimized the Wk relative to a reference uniform distribution generated via bootstrapping (cluster::clusGap), respectively. The Ck that performed the best across all criteria was selected as the optimal Ck and was considered to best reflect the underlying Ck in the data.

Step 3: Derive and visualize the taste perception profiles

Using the selected cluster algorithm and Ck, 50 cluster solutions were generated and the cluster solution that produced the lowest Wk was selected as the taste perception profiles. Radar plots (fmsb::radarchart) were then used to visualize the mean perception scores for each taste among individuals with each taste perception profile to examine their defining characteristics.

Step 4: Assess internal validity and stability of the taste perception profiles

Internal validity of the taste perception profiles was assessed on the basis of 2 internal cluster validity indices. The Calinski-Harabasz index was used to measure the ratio of between- to within-cluster variation (NBClust::NbClust, index =   ch), where higher values indicated better fit for the data. The Davis-Bouldin index was used to measure the average ratio of within- to between-cluster variation, over all pairs of clusters in a cluster solution (NbClust::NbClust, index = db), where lower values indicated better fit for the data. For each index, values were computed and compared against all cluster solutions with a Ck from 2 to 10 derived via the selected cluster algorithm.

Internal stability of the taste perception profiles was assessed on the basis of the Jaccard index (JI). For each Ck, data were resampled via bootstrapping 100 times. Cluster analysis, using the selected cluster algorithm, was run on the original data and each bootstrapped sample, and the JI was calculated between each original cluster and the most similar cluster in each bootstrapped sample solution (fpc::clusterboot). The JI is a relative measure of similarity between 2 solutions, where a value of 1 indicates identical clusters and a value of 0 indicates clusters with no overlap. The mean JI for each original cluster was tabulated over all sample solutions as a measure of internal stability and validity. Values ≥0.85 indicate highly stable and valid clusters, values ≥0.75 indicate stable and valid clusters, values >0.60 indicate probable clusters with unclear membership, and values ≤0.60 indicate highly unstable clusters.

Statistical analyses

Data are expressed as means ± SDs or n (%) unless otherwise specified. Descriptive statistics were used to examine central tendency and dispersion of demographic, clinical, and lifestyle characteristics for the cohort, overall and stratified by sex, and to examine mean taste perception scores, overall and stratified by sex and taste perception profile. Differences in characteristics were examined using Student's t tests, ANOVA tests, and chi-square tests; where ANOVA P < 0.05, Student's t tests were used to conduct post hoc pairwise comparisons with a Bonferroni correction applied for multiple testing. To compare the ability of the taste perception profiles against total taste scores to capture individual variability in taste perception, goodness-of-fit parameters, including adjusted coefficients of determination (R2adj) and root mean square errors (RMSE), were calculated using multivariable linear regression with each measure predicting perception of each of the 5 tastes. All regression models were adjusted for age, sex, physical activity, smoking status (current/nonsmoker), medication use (cholesterol lowering, antihypertensive and/or hypoglycemic), Type 2 diabetes status, and energy intake.

The threshold for statistical significance was P < 0.05. All analyses were conducted in R Statistical Environment (version 3.6.2) and the relevant R code for the data-driven clustering approach along with a simulated dataset have been made available as a repository on GitHub (34).

Results

Participant characteristics

Participants were 65 ± 5 y old (mean ± SD) with metabolic syndrome and had a BMI of 32.3 ± 3.6; 55% were female, 42% had Type 2 diabetes, and a majority reported taking antihypertensive and/or cholesterol lowering medications (79% or 65%, respectively) (Table 1). Compared with men, women were older based on the inclusion criteria (P = 0.0002) and were less likely to be current smokers (P  = 0.04). Women tended to have higher mean taste perception scores than men, with the differences reaching statistical significance for salt (2.9 vs. 2.3, respectively; P < 0.0001) and sour (2.7 vs. 2.2, respectively; P = 0.0006) (P = 0.20, P  = 0.25, and P = 0.23 for sweet, bitter, and umami perceptions, respectively) (Supplemental Table 1).

TABLE 1.

Demographic, clinical, and lifestyle characteristics of the study participants, overall and stratified by sex1

Overall (n = 367) Women (n = 202) Men (n = 165) P
Age, y 65 ± 5 66 ± 4 64 ± 5 0.0002
Body weight, kg 84.5 ± 13.4 77.8 ± 9.7 92.6 ± 12.8 <0.0001
BMI, kg/m2 32.3 ± 3.6 32.4 ± 3.7 32.3 ± 3.4 0.80
Fasting blood glucose, mg/dL 117 ± 32.7 116 ± 35.9 117 ± 28.2 0.77
HDL cholesterol, mg/dL 48.6 ± 11.0 51.1 ± 10.9 45.5 ± 10.4 <0.0001
Triglycerides, mg/dL 165 ± 82.9 158 ± 74.7 174 ± 91.7 0.09
Systolic blood pressure, mm Hg 140 ± 16.8 139 ± 17.6 142 ± 15.6 0.03
Diastolic blood pressure, mm Hg 80.0 ± 9.04 78.6 ± 8.52 81.7 ± 9.38 0.001
Type 2 diabetes 154 (42) 83 (41) 71 (43) 0.82
Medications
 Hypoglycemic2 118 (32) 62 (31) 56 (34) 0.58
 Antihypertensives 289 (79) 159 (79) 130 (79) >0.99
 Cholesterol lowering 240 (65) 130 (64) 110 (67) 0.72
Physical activity, MET-min/wk 1800 ± 1670 1610 ± 1430 2030 ± 1890 0.02
Current smoker 37 (10) 14 (7) 23 (14) 0.04
Energy intake, kcal/d 2390 ± 523 2260 ± 451 2550 ± 561 <0.0001
Open in a new tab
1

Values are means ± SDs or n (%). MET, metabolic equivalent.

2

Hypoglycemic medications included metformin and/or insulin.

Data-driven clustering approach

Selection of the cluster algorithm and optimal number of clusters

The reproducibility of cluster solutions derived via KCA and Ward's D were examined using boxplots (Figure 1 and Supplemental Figure 1). When aggregating across all values of Ck (2 to 10), cluster solutions derived via KCA had higher reproducibility (ARIs and CramVs closer to 1) than those derived via Ward's D. On this basis, KCA was selected to derive the taste perception profiles.

FIGURE 1.

FIGURE 1

Open in a new tab

Distribution of ARI reproducibility index values for KCA and Ward's D for deriving a range of number of clusters from sweet, salt, sour, bitter, and umami taste perception scores, among PREDIMED (PREvención con DIeta MEDiterránea)-Plus Valencia participants (n = 367). Twenty ARI estimates were generated via split half cross-validation, at each number of clusters, from 2 to 10 (range = negative to 1; optimal = 1). ARI values generated by KCA were higher than those generated by Ward's D, indicating that KCA is the more reproducible cluster algorithm. ARI, Adjusted Rand Index; KCA, k-means clustering; Ward's D, Ward's minimum variance.

Six was selected as the optimal Ck in the data (Figure 2). In the elbow plot, <6 clusters resulted in steeper declines in Wk, indicating desirable decreases until 6 clusters, while additional clusters resulted in smaller declines in Wk, indicating smaller decreases when parsing into >6 clusters (average slope of Wk decline: 224.8 vs. 67.2, respectively). In the average silhouette and gap statistic plots, local (0.21) and global (0.53) maximums (optimal values), respectively, were observed at 6 clusters, indicating that 6 clusters best minimized the within- and maximized the between-cluster variation in the data (Supplemental Figure 2).

FIGURE 2.

FIGURE 2

Open in a new tab

Elbow plot of the total within-cluster variation over a range of number of clusters derived by KCA from sweet, salt, sour, bitter, and umami taste perception scores, among the study participants (n = 367). Six was selected as the optimal number of clusters in the data based on the relatively steeper declines in total within-cluster variation before 6 and the slower declines after 6; this “elbow” is amplified in the plot inset. KCA, k-means clustering.

Derived taste perception profiles

All 6 taste perception profiles had distinct characteristics relative to the overall cohort means (Figure 3). One profile had mean perception scores of all 5 tastes ∼1 SD below the cohort means—hence, was labeled “Low All” (n = 85, 23%). Two profiles had mean perception scores of either bitter or umami ∼1 SD above the cohort means—hence, were labeled “High Bitter” (n = 59, 16%) and “High Umami” (n = 61, 17%), respectively. Another profile had mean perception scores of bitter and umami ∼0.5 SD below the cohort means while sweet, salt, and sour were slightly above—hence, was labeled “Low Bitter & Umami” (n = 72, 20%). Finally, 2 profiles had mean perception scores of 4 tastes ∼1 SD above the cohort means while each of bitter and umami, respectively, were nearer to the cohort means—hence, were labeled “High All But Bitter” (n = 49, 13%) and “High All But Umami” (n  = 41, 11%), respectively. ANOVA tests indicated there were significant differences in perception scores for all 5 tastes across the taste perception profiles (all F-test P  < 0.0001), indicating the success of the KCA algorithm in identifying distinct and relatively homogenous clusters of individuals with shared patterns in taste perception (Supplemental Table 2).

FIGURE 3.

FIGURE 3

Open in a new tab

Six taste perception profiles derived via a data-driven clustering approach among the study participants (n = 367): “Low All” (n = 85), “High Bitter” (n = 59), “High Umami” (n = 61), “Low Bitter & Umami” (n = 72), “High All But Bitter” (n = 49), and “High All But Umami” (n = 49). Mean perception of each taste for each profile is represented by solid black lines; mean ± 1 SD perception of each taste for the cohort is represented by the dark-gray lines and shaded gray area, respectively. Perception scores for each taste ranged from 0 to 5; 0 is the innermost pentagon and 5 is the outer most pentagon.

Internal validity and stability of the taste perception profiles

Although there was some discrepancy among internal cluster validity indices, the minimum (optimal) value of the Davies-Bouldin index, which reflects the lowest average ratio of within- to between-cluster variation, was obtained at 6 clusters, indicating that the 6 taste perception profiles were valid and well fitted to the data (Supplemental Table 3).

For internal stability, all 6 taste perception profiles had JIs >0.75 (range: 0.78 to 0.91) and 4 profiles (High Bitter, Low All, High All But Bitter, High All But Umami, in descending order) had values ≥0.85, indicating that all 6 taste perception profiles were either stable or highly stable, respectively, valid, and had well-defined cluster membership (Supplemental Table 4). The 6 taste perception profiles were also among the most stable cluster solutions of all derived with a Ck from 2 to 10, further demonstrating their relative validity and fit compared with other potential cluster numbers and solutions.

Comparison of taste perception profiles and total taste scores

The mean ± SD total taste scores according to taste perception profile were 5.1 ± 1.7 for Low All, 9.6 ± 2.0 for High Bitter, 11.4 ± 2.1 for High Umami, 10.5 ± 1.8 for Low Bitter & Umami, 17.2 ± 2.1 for High All But Bitter, and 15.7 ± 1.8 for High All But Umami. Examination of the distribution of total taste scores across taste perception profiles showed high degrees of overlap, such that individuals with different taste perception profiles had similar total taste scores (Figure 4). Accordingly, total taste scores were similar for those with Low Bitter & Umami or High Umami profiles (P = 0.15) and those with Low Bitter & Umami or High Bitter profiles (P = 0.11). Likewise, the magnitude of differences in total taste scores were <2 points for individuals with a High All But Bitter or High All But Umami profile and those with a High Bitter or High Umami profile; however, the differences were statistically significant (P = 0.008 and P  < 0.0001, respectively).

FIGURE 4.

FIGURE 4

Open in a new tab

Distribution of total taste scores for individuals with each derived taste perception profile among the study participants (n = 367). The high degrees of overlap suggest that individuals with different taste perception profiles can have similar total taste scores.

Results from the goodness-of-fit evaluations indicated that the taste perception profiles substantially outperformed total taste scores for capturing the individual variability in bitter and umami perception, whereas both measures yielded comparable results for capturing the individual variability in sweet, salt, and sour perception (Table 2). Accordingly, after adjusting for age, sex, physical activity, smoking status, medication use, Type 2 diabetes status, and energy intake, taste perception profiles explained a substantially higher amount of the variability in bitter and umami perception than total taste scores (R2adj for bitter: 0.63 vs. 0.19, respectively; R2adj for umami: 0.65 vs. 0.40, respectively) and yielded lower average prediction errors than total taste scores (RMSE for bitter: 0.87 vs. 1.23, respectively; RMSE for umami: 0.82 vs. 1.06, respectively).

TABLE 2.

Comparison of model fit parameters for taste perception profiles and total taste scores predicting perception of each of the 5 tastes among the study participants1

Adjusted R2  2 Root mean square error3
Taste perception profiles Total taste scores Taste perception profiles Total taste scores
Sweet 0.35 0.43 1.03 0.96
Salt 0.67 0.65 0.81 0.83
Sour 0.52 0.58 0.89 0.83
Bitter 0.63 0.19 0.87 1.23
Umami 0.65 0.40 0.82 1.06
Open in a new tab
1

Values were obtained from multivariable linear regression models, adjusting for age, sex, physical activity, smoking status, medication use (cholesterol lowering, antihypertensive, or hypoglycemic), Type 2 diabetes status, and energy intake.

2

A higher adjusted R2 indicates that more variability is explained in the outcome given the number of predictors—hence, better model fit.

3

A lower root mean square error indicates lower average prediction errors for each outcome—hence, better model fit.

Discussion

The present study demonstrates the successful use of a data-driven clustering approach to derive taste perception profiles from sweet, salt, sour, bitter, and umami perception scores. This approach identified 6 distinct taste perception profiles among a cohort of older, overweight adults with metabolic syndrome—including Low All, High Bitter, High Umami, Low Bitter & Umami, High All But Bitter, and High All But Umami—which were well-fitted to the data and stable when assessed by several internal validity and stability indices, respectively. Collectively, the taste perception profiles outperformed total taste scores for capturing individual variability in bitter and umami perception and were comparable for sweet, salt, and sour perception. Taste perception profiles also captured the differential perception of each taste within individuals, whereas these distinct patterns were lost when using total taste scores.

A key advantage of the current approach is the ability to capture perception of all 5 tastes simultaneously as one collective measure. This may provide a more comprehensive representation of how myriad taste compounds in foods and beverages are detected and perceived and may allow for the elucidation of patterns in the data that are difficult to study using single taste perceptions or total taste scores. From a methodological perspective, accounting for the differential perception of all 5 tastes within individuals may also reduce residual confounding (22). For example, in the present cohort, individuals with a Low All or High Bitter profile had similar mean sweet perception scores (1.4 vs. 1.7, respectively; P  = 0.60) but different bitter perception scores (0.4 vs. 2.9, respectively; P < 0.0001) and individuals with a High Bitter, Low Bitter & Umami, or High Umami profile had similar total taste scores (9.6 vs. 10.5 vs. 11.4, respectively), despite having different single taste perceptions. When linking taste perception to diet quality, if only sweet perception was considered, relations may be confounded by not controlling for differences in bitter perception and relations that involve both tastes may be misattributed to a single taste. Likewise, as total taste scores are unable to capture the within-individual differences in perception of the 5 tastes, their ability to capture relations with food intake and diet quality may be limited.

To put these observations into context of the broader literature focusing on bitter taster status (10), post hoc analyses were conducted to examine whether taste perception profiles differed among those regarded as non-tasters, medium tasters, or supertasters. Individuals were classified into the 3 taster groups using their bitter perception score, and histograms were used to evaluate the frequency of taste perception profiles among taster groups (Supplemental Figure 3). Among non-tasters (n = 136), 43% had low perception of all 5 tastes (Low All), while 52% had moderate to high perception of at least 1 of sweet, salt, sour, or umami (27% Low Bitter & Umami, 18% High Umami, and 12% High All But Bitter). Likewise, among supertasters, 45% had a High Bitter profile, indicating high bitter perception only, while 41% had a High All But Umami profile, indicating high perception of sweet, salt, and sour in addition to bitter. These data point to the possibility of clusters within taster groups that may differ on the basis of perception of the other tastes. Furthermore, while non-tasters and supertasters tended to have different taste perception profiles, the heterogeneity within taster groups challenges the notion that bitter taster status may serve as a surrogate measure of perception of all 5 tastes (35).

Applications of this work may help facilitate a better understanding of how taste perception varies across diverse populations. In the present cohort, 6 taste perception profiles were identified, which were characterized by high variability in bitter and umami perception and relatively less variability in sweet, salt, and sour perception. However, given observed population divergence in taste perception genotypes and phenotypes (36), it is unclear whether different taste perception profiles may be identified in different populations or whether certain taste perception profiles may be generally conserved. Understanding these differences could provide valuable insights into regional- or cultural-level drivers of food intake and diet quality (37); and the ability to classify individuals into taste perception profiles may help identify subgroups of individuals at elevated risk for certain diet-related chronic diseases.

While cluster analysis has been widely used for studying dietary patterns, only 2 prior publications were identified that used cluster analysis to study patterns in taste perception (38, 39). One study examined individual differences in taste perception (for all 5 tastes) among Finnish adults (38). The other examined the relations between “taste-intensity patterns” (for sweet, salt, sour, and bitter) and longitudinal changes in health outcomes among US adults (39). In both studies, Ward's D was the only algorithm used to identify patterns in taste perception, and in 1 study the number of clusters in the data was selected a priori. In contrast to these reports, the current approach was more data driven. Considering 2 cluster algorithms and a range of number of clusters minimized the need to make decisions a priori. Instead, quantitative criteria identified from studies outlining statistical approaches to limit subjectivity in cluster analysis (29, 30, 40) were relied upon to identify the most reproducible cluster algorithm, determine the optimal number of clusters in the data, and to assess the internal validity and stability of the derived taste perception profiles. This resulted in a relatively repeatable and transparent approach, which required fewer assumptions to be made about the underlying data structure.

Nevertheless, the inability to consider all possible cluster algorithms or number of clusters for the derivation may have introduced some subjectivity in the approach. However, KCA and Ward's D were selected as they are the most widely used cluster algorithms for deriving dietary patterns (28), and a range of 2 to 10 clusters was considered to ensure representativeness of the derived profiles. Although there was some discrepancy across the criteria that informed the final number of clusters to select, this is common in cluster analysis (27, 28) and highlights the importance of using multiple criteria to inform these decisions. While a primary index was selected for each major decision (e.g., ARI for the algorithm selection and Elbow plot for the number of clusters in the data), when discrepancies occurred, consensus was reached by examining all relevant criteria for a common trend: for example, 6 clusters was selected as the best fit for the data based on the consensus among the Elbow and gap statistic plots and the Davies-Bouldin internal validity index, and because 6 clusters yielded a local optimal value of the average silhouette statistic. While it is possible these differences are indicative of underlying instability stemming from the small sample size, the 6 taste perception profiles were among the most stable of all solutions derived with 2 to 10 clusters (Supplemental Table 4), which supports this selection.

Limitations of the study include the lack of generalizability of the derived taste perception profiles, as the present cohort was comprised of older, overweight adults of European ancestry, with metabolic syndrome, who were living in a Mediterranean country. However, the data-driven clustering approach to derive the taste perception profiles is generalizable and would require limited a priori knowledge to perform. The taste perception data were also not standardized prior to clustering, in light of evidence suggesting standardization may dilute underlying variability or give undue influence to minor clusters in a dataset (41). For comparison, however, profiles were also derived from z-score standardized data and the results were highly similar (data not shown). Also, while underlying sex effects may have contributed to the taste perception profiles, sex-stratified analyses were not conducted given the lack of statistical power. For the goodness-of-fit comparisons, since total taste scores are not designed to capture information on the individual tastes, they may not have provided the most meaningful comparison; however, they are the only widely used approach for studying multiple taste perceptions as a single measure—hence, they served as the most relevant comparison. Limitations inherent in collecting the taste perception data include the use of a 6-point categorical scale for taste perception assessments rather than the general labeled magnitude scale with non-taste sensations as standards (42). The tastant concentrations used to assess taste perception may also not reflect usual tastant concentrations in foods and beverages (43). However, the 6-point scale is suggested to be more suitable for older adults because of its simplicity, which minimizes several contextual effects (44), the tastant concentrations were based on previous reports (3), and the taste perception assessments were carried out under highly controlled conditions (24).

In conclusion, as emphasis shifts from population-wide to more personalized nutrition guidance, it is critical to better understand the individual-level drivers of food intake. Although taste has become a prime candidate, data linking individual differences in taste perception to food intake are inconsistent. Taste perception profiles, like dietary patterns, may offer an alternative approach to address this issue by providing a more comprehensive measure of perception of all 5 tastes. The present study demonstrated that a data-driven clustering approach could be used to derive taste perception profiles from sweet, salt, sour, bitter, and umami perception scores. By applying this approach to a cohort of older adults with metabolic syndrome, 6 distinct taste perception profiles were identified, which all had high internal validity and stability. Compared with total taste scores, taste perception profiles captured relatively more of the individual variability in bitter and umami perception, as well as the differential perception of all 5 tastes within individuals. On this basis, data-driven approaches that capture perception of all 5 tastes collectively as a taste perception profile may be informative for studying how individual differences in sweet, salt, sour, bitter, and umami perception collectively relate to food choices and intakes. Such understanding may, in turn, facilitate the development of customized strategies to improve diet quality and reduce diet-related chronic disease risk.

Supplementary Material

nxab160_Supplemental_File
Click here for additional data file. (254.9KB, pdf)

Acknowledgments

The authors’ responsibilities were as follows—JEG, AHL, KKHC, JM, DC, and OC: designed the analytical approach; DC, MF, OC, and JVS: designed and managed the PREDIMED-Plus Valencia study; RF-C, RB, JVS, and JIG: collected the PREDIMED-Plus Valencia data; JEG: analyzed the data and performed statistical analyses; JEG and AHL: drafted the manuscript; AHL and DC: have primary responsibility for final content; and all authors: critically evaluated the manuscript and read and approved the final manuscript.

Notes

This research was supported in Spain by the Ministry of Health (Instituto de Salud Carlos III) and the Ministry of Science and Innovation–Fondo Europeo de Desarrollo Regional (FEDER) (grants CIBER 06/03, PI06/1326, PI13/00728, PI16/00366, PI19/00781, SAF2016-80532-R, PID2019-108858RB-I00); the University Jaume I (grants UJI-B2018-69 and COGRUP/2016/06); the Rei Jaume I Award for Medical Research 2018; and the Generalitat Valenciana-Conselleria de Innovación, Universidades, Ciencia y Sociedad Digital (grants PROMETEO 2017/017 and APOSTD/2019/136) (OC, RF-C, JVS, RB, MF, JIG, DC).

This research was also supported by USDA Non-Assistant Cooperative Agreement 58-80-50-9-004 (AHL); the Gerald Cassidy Student Innovation Award, Jean Mayer USDA Human Nutrition Research Center on Aging, Tufts University, and Stanley N Gershoff Student Scholarship, Friedman School of Nutrition Science and Policy, Tufts University (JEG); and the National Heart, Lung, and Blood Institute Career Transition Award 1K22HL135075-01 (JM).

Author disclosures: The authors report no conflicts of interest. Any opinions, findings, conclusions, or recommendations expressed in this publication are those of the author(s) and do not necessarily reflect the views of the USDA or NIH.

Supplemental Tables 1–4 and Supplemental Figures 1–3 are available from the “Supplementary data” link in the online posting of the article and from the same link in the online table of contents available at https://academic.oup.com/jn.

DC and AHL contributed equally as last authors.

Abbreviations used: ARI, Adjusted Rand Index; Ck, number of clusters; CramV, Cramer's V statistic; JI, Jaccard index; KCA, K-means clustering; MPG, monopotassium l-glutamate; PREDIMED, PREvención con DIeta MEDiterránea; PTC, phenylthiocarbamide; RMSE, root mean square error of approximation; Ward's D, Ward's minimum variation; Wk, total within-cluster variance.

Contributor Information

Julie E Gervis, Cardiovascular Nutrition Laboratory, Jean Mayer USDA Human Nutrition Research Center on Aging, Tufts University, Boston, MA, USA.

Kenneth K H Chui, Department of Public Health and Community Medicine, Tufts University School of Medicine, Boston, MA, USA.

Jiantao Ma, Department of Nutrition Data Science, Friedman School of Nutrition Science and Policy, Tufts University, Boston, MA, USA.

Oscar Coltell, Department of Computer Languages and Systems, University of Jaume I, Castellón, Spain; CIBER Fisiopatología de la Obesidad y Nutrición, Instituto de Salud Carlos III, Madrid, Spain.

Rebeca Fernández-Carrión, CIBER Fisiopatología de la Obesidad y Nutrición, Instituto de Salud Carlos III, Madrid, Spain; Department of Preventive Medicine and Public Health, University of Valencia, Valencia, Spain.

José V Sorlí, CIBER Fisiopatología de la Obesidad y Nutrición, Instituto de Salud Carlos III, Madrid, Spain; Department of Preventive Medicine and Public Health, University of Valencia, Valencia, Spain.

Rocío Barragán, CIBER Fisiopatología de la Obesidad y Nutrición, Instituto de Salud Carlos III, Madrid, Spain; Department of Preventive Medicine and Public Health, University of Valencia, Valencia, Spain.

Montserrat Fitó, CIBER Fisiopatología de la Obesidad y Nutrición, Instituto de Salud Carlos III, Madrid, Spain; Department of Preventive Medicine and Public Health, University of Valencia, Valencia, Spain; Cardiovascular Risk and Nutrition Research Group, Hospital del Mar Medical Research Institute (IMIM), Barcelona, Spain.

José I González, CIBER Fisiopatología de la Obesidad y Nutrición, Instituto de Salud Carlos III, Madrid, Spain; Department of Preventive Medicine and Public Health, University of Valencia, Valencia, Spain.

Dolores Corella, CIBER Fisiopatología de la Obesidad y Nutrición, Instituto de Salud Carlos III, Madrid, Spain; Department of Preventive Medicine and Public Health, University of Valencia, Valencia, Spain.

Alice H Lichtenstein, Cardiovascular Nutrition Laboratory, Jean Mayer USDA Human Nutrition Research Center on Aging, Tufts University, Boston, MA, USA.

References

  • 1. Bachmanov  AA, Beauchamp  GK. Taste receptor genes. Annu Rev Nutr. 2007;27:389–414. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2. Roper  SD, Chaudhari  N. Taste buds: cells, signals and synapses. Nat Rev Neurosci. 2017;18:485–97. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3. Barragán  R, Coltell  O, Portolés  O, Asensio  EM, Sorlí  JV, Ortega-Azorín  C, González  JI, Sáiz  C, Fernández-Carrión  R, Ordovas  JM  et al.  Bitter, sweet, salty, sour and umami taste perception decreases with age: sex-specific analysis, modulation by genetic variants and taste-preference associations in 18 to 80 year-old subjects. Nutrients. 2018;10:1539. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4. Dovey  TM, Boyland  EJ, Trayner  P, Miller  J, Rarmoul-Bouhadjar  A, Cole  J, Halford  JCG. Alterations in taste perception due to recreational drug use are due to smoking a substance rather than ingesting it. Appetite. 2016;107:1–8. [DOI] [PubMed] [Google Scholar]
  • 5. Pepino  MY, Mennella  JA. Effects of cigarette smoking and family history of alcoholism on sweet taste perception and food cravings in women. Alcohol Clin Exp Res. 2007;31:1891–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6. Breslin  PAS. An evolutionary perspective on food review and human taste. Curr Biol. 2013;23:R409–18. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7. Hayes  JE, Feeney  EL, Allen  AL. Do polymorphisms in chemosensory genes matter for human ingestive behavior?. Food Qual Preference. 2013;30:202–16. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8. International Food Information Council . The Food and Health Survey. [Internet]. Washington (DC): The International Food Information Council (IFIC) Foundation, 2019, [cited 2021 Feb 17]. Available from: https://foodinsight.org/2019-food-and-health-survey/. [Google Scholar]
  • 9. Reed  DR, Alhadeff  AL, Beauchamp  GK, Chaudhari  N, Duffy  VB, Dus  M, Fontanini  A, Glendinning  JI, Green  BG, Joseph  PV. NIH workshop report: sensory nutrition and disease. Am J Clin Nutr. 2021;113:232–45. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10. Hayes  JE, Bartoshuk  LM, Kidd  JR, Duffy  VB. Supertasting and PROP bitterness depends on more than the TAS2R38 gene. Chem Senses. 2008;33:255–65. [DOI] [PubMed] [Google Scholar]
  • 11. Bartoshuk  LM. The biological basis of food perception and acceptance. Food Qual Preference. 1993;4:21–32. [Google Scholar]
  • 12. Duffy  VB, Hayes  JE, Davidson  AC, Kidd  JR, Kidd  KK, Bartoshuk  LM. Vegetable intake in college-aged adults is explained by oral sensory phenotypes and TAS2R38 genotype. Chemosens Percept. 2010;3:137–48. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13. Sandell  M, Hoppu  U, Mikkilä  V, Mononen  N, Kähönen  M, Männistö  S, Rönnemaa  T, Viikari  J, Lehtimäki  T, Raitakari  OT. Genetic variation in the hTAS2R38 taste receptor and food consumption among Finnish adults. Genes Nutr. 2014;9:433. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14. Shafaie  Y, Koelliker  Y, Hoffman  DJ, Tepper  BJ. Energy intake and diet selection during buffet consumption in women classified by the 6-n-propylthiouracil bitter taste phenotype. Am J Clin Nutr. 2013;98:1583–91. [DOI] [PubMed] [Google Scholar]
  • 15. Duffy  VB, Davidson  AC, Kidd  JR, Kidd  KK, Speed  WC, Pakstis  AJ, Reed  DR, Snyder  DJ, Bartoshuk  LM. Bitter receptor gene (TAS2R38), 6-n-propylthiouracil (PROP) bitterness and alcohol intake. Alcohol Clin Exp Res. 2004;28:1629–37. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16. Drewnowski  A, Henderson  SA, Cockroft  JE. Genetic sensitivity to 6-n-propylthiouracil has no influence on dietary patterns, body mass indexes, or plasma lipid profiles of women. J Am Diet Assoc. 2007;107:1340–8. [DOI] [PubMed] [Google Scholar]
  • 17. Ooi  S-X, Lee  P-L, Law  H-Y, Say  Y-H. Bitter receptor gene (TAS2R38) P49A genotypes and their associations with aversion to vegetables and sweet/fat foods in Malaysian subjects. Asia Pac J Clin Nutr. 2010;19:491–8. [PubMed] [Google Scholar]
  • 18. Jayasinghe  SN, Kruger  R, Walsh  DC, Cao  G, Rivers  S, Richter  M, Breier  BH. Is sweet taste perception associated with sweet food liking and intake?. Nutrients. 2017;9:750. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19. Dias  AG, Eny  KM, Cockburn  M, Chiu  W, Nielsen  DE, Duizer  L, El-Sohemy  A. Variation in the TAS1R2 gene, sweet taste perception and intake of sugars. Lifestyle Genom. 2015;8:81–90. [DOI] [PubMed] [Google Scholar]
  • 20. Veček  NN, Mucalo  L, Dragun  R, Miličević  T, Pribisalić  A, Patarčić  I, Hayward  C, Polašek  O, Kolčić  I. The association between salt taste perception, Mediterranean diet and metabolic syndrome: a cross-sectional study. Nutrients. 2020;12:1164. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21. Chamoun  E, Carroll  NA, Duizer  LM, Qi  W, Feng  Z, Darlington  G, Duncan  AM, Haines  J, Ma  DWL; Guelph Family Health Study . The relationship between single nucleotide polymorphisms in taste receptor genes, taste function and dietary intake in preschool-aged children and adults in the Guelph Family Health Study. Nutrients. 2018;10:990. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22. Puputti  S, Hoppu  U, Sandell  M. Taste sensitivity is associated with food consumption behavior but not with recalled pleasantness. Foods, 2019; 8:444. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23. Choi  J-H, Lee  J, Choi  IJ, Kim  Y-W, Ryu  KW, Kim  J. Variations in TAS1R taste receptor gene family modify food intake and gastric cancer risk in a Korean population. Mol Nutr Food Res. 2016;60:2433–45. [DOI] [PubMed] [Google Scholar]
  • 24. Coltell  O, Sorlí  JV, Asensio  EM, Fernández-Carrión  R, Barragán  R, Ortega-Azorín  C, Estruch  R, González  JI, Salas-Salvadó  J, Lamon-Fava  S. Association between taste perception and adiposity in overweight or obese older subjects with metabolic syndrome and identification of novel taste-related genes. Am J Clin Nutr. 2019;109:1709–23. [DOI] [PubMed] [Google Scholar]
  • 25. Fernandez-Garcia  JC, Alcaide  J, Santiago-Fernandez  C, Roca-Rodriguez  M, Aguera  Z, Baños  R, Botella  C, Torre  R, Fernandez-Real  JM, Fruhbeck  G  et al.  An increase in visceral fat is associated with a decrease in the taste and olfactory capacity. PLoS One. 2017;12:e0171204. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26. Hu  FB. Dietary pattern analysis: a new direction in nutritional epidemiology. Curr Opin Lipidol. 2002;13:3. [DOI] [PubMed] [Google Scholar]
  • 27. Devlin  UM, McNulty  BA, Nugent  AP, Gibney  MJ. The use of cluster analysis to derive dietary patterns: methodological considerations, reproducibility, validity and the effect of energy mis-reporting. Proc Nutr Soc. 2012;71:599–609. [DOI] [PubMed] [Google Scholar]
  • 28. Edefonti  V, De Vito  R, Dalmartello  M, Patel  L, Salvatori  A, Ferraroni  M. Reproducibility and validity of a posteriori dietary patterns: a systematic review. Adv Nutr. 2020;11:1255–81. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29. Sauvageot  N, Schritz  A, Leite  S, Alkerwi  A, Stranges  S, Zannad  F, Streel  S, Hoge  A, Donneau  A-F, Albert  A  et al.  Stability-based validation of dietary patterns obtained by cluster analysis. Nutr J. 2017;16:4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30. Lo Siou  G, Yasui  Y, Csizmadi  I, McGregor  SE, Robson  PJ. Exploring statistical approaches to diminish subjectivity of cluster analysis to derive dietary patterns: the Tomorrow Project. Am J Epidemiol. 2011;173:956–67. [DOI] [PubMed] [Google Scholar]
  • 31. Martínez-González  MA, Buil-Cosiales  P, Corella  D, Bulló  M, Fitó  M, Vioque  J, Romaguera  D, Martínez  JA, Wärnberg  J, López-Miranda  J  et al.  Cohort profile: design and methods of the PREDIMED-Plus randomized trial. Int J Epidemiol. 2019;48:387–388o. [DOI] [PubMed] [Google Scholar]
  • 32. Molina  L, Sarmiento  M, Peñafiel  J, Donaire  D, Garcia-Aymerich  J, Gomez  M, Ble  M, Ruiz  S, Frances  A, Schröder  H. Validation of the Regicor short physical activity questionnaire for the adult population. PLoS One. 2017;12:e0168148. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33. Fernández-Ballart  JD, Piñol  JL, Zazpe  I, Corella  D, Carrasco  P, Toledo  E, Perez-Bauer  M, Martínez-González  MÁ, Salas-Salvadó  J, Martín-Moreno  JM. Relative validity of a semi-quantitative food-frequency questionnaire in an elderly Mediterranean population of Spain. Br J Nutr. 2010;103:1808–16. [DOI] [PubMed] [Google Scholar]
  • 34. Gervis  JE. GitHub. Taste-perception-profiles. [Cited 2021 Apr 13] [Internet]. Available from: https://github.com/julieeg/taste-perception-profiles. [Google Scholar]
  • 35. Hayes  JE, Keast  RSJ. Two decades of supertasting: where do we stand?. Physiol Behav. 2011;104:1072–4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36. Sjöstrand  AE, Sjödin  P, Hegay  T, Nikolaeva  A, Shayimkulov  F, Blum  MG, Heyer  E, Jakobsson  M. Taste perception and lifestyle: insights from phenotype and genome data among Africans and Asians. Eur J Hum Genet. 2020, 29;325–37. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37. Risso  DS, Giuliani  C, Antinucci  M, Morini  G, Garagnani  P, Tofanelli  S, Luiselli  D. A bio-cultural approach to the study of food choice: the contribution of taste genetics, population and culture. Appetite. 2017;114:240–7. [DOI] [PubMed] [Google Scholar]
  • 38. Puputti  S, Aisala  H, Hoppu  U, Sandell  M. Multidimensional measurement of individual differences in taste perception. Food Qual Preference. 2018;65:10–17. [Google Scholar]
  • 39. Fischer  ME, Cruickshanks  KJ, Schubert  CR, Pinto  A, Huang  G-H, Klein  BEK, Klein  R, Pankow  JS. The association of taste with change in adiposity-related health measures. J Acad Nutr Diet. 2014;114:1195–202. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40. Fransen  HP, May  AM, Stricker  MD, Boer  JMA, Hennig  C, Rosseel  Y, Ocké  MC, Peeters  PHM, Beulens  JWJ. A posteriori dietary patterns: how many patterns to retain?. J Nutr. 2014;144:1274–82. [DOI] [PubMed] [Google Scholar]
  • 41. Moeller  SM, Reedy  J, Millen  AE, Dixon  LB, Newby  PK, Tucker  KL, Krebs-Smith  SM, Guenther  PM. Dietary patterns: challenges and opportunities in dietary patterns research: an experimental biology workshop. J Am Diet Assoc. 2007;107:1233–9. [DOI] [PubMed] [Google Scholar]
  • 42. Bartoshuk  LM, Duffy  VB, Green  BG, Hoffman  HJ, Ko  C-W, Lucchina  LA, Marks  LE, Snyder  DJ, Weiffenbach  JM. Valid across-group comparisons with labeled scales: the gLMS versus magnitude matching. Physiol Behav. 2004;82:109–14. [DOI] [PubMed] [Google Scholar]
  • 43. Lucas  F, Bellisle  F. The measurement of food preferences in humans: do taste-and-spit tests predict consumption?. Physiol Behav. 1987;39:739–43. [DOI] [PubMed] [Google Scholar]
  • 44. Griep  M, Borg  E, Collys  K, Massart  D. Category ratio scale as an alternative to magnitude matching for age-related taste and odour perception. Food Qual Preference. 1998;9:67–72. [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

nxab160_Supplemental_File
Click here for additional data file. (254.9KB, pdf)

Articles from The Journal of Nutrition are provided here courtesy of American Society for Nutrition

RESOURCES