Assessing taste phenotypes
Volunteers involved in the study were classified, after the perception test, in different tasting categories for both PROP and stevioside. Regarding the PROP status, the distribution of PROP sensitivity showed the classical bimodal curve (Online Resource 2), with 18 individuals (20.9 %) being classified as non-tasters and 68 (79.1 %) as tasters. In order to test whether the adopted deviation from the method described in Zhao et al. (2003) could affect PROP phenotype assessment, we calculated the overall genotype–phenotype concordance (90.7 %), thus inferring the correctness of our cotton swab PROP test. After stevioside tasting, 11 individuals (13 %) were able to perceive only its bitter taste. On the contrary, 22 individuals (26 %) perceived only its sweet taste, whereas the majority of samples constituted by 53 individuals (61 %) identified a sweet taste followed by a bitter after-taste.
The 53 bitter/sweet-tasters were subsequently distinguished into 32 “bitter-low” (LMS scores up to 50) and 21 “bitter-high” (LMS from 60 to 100) tasters (Fig. 2) according to LMS scale scores.
TAS2R4 SNP regulates the ability to perceive the bitter taste of stevioside
The TAS2R4 rs2234001 (C/G) turned out to be associated with stevioside bitter status. In fact, the “bitter” and “sweet” phenotype groups showed statistically significant differences for both genotypic (adjusted P = 0.002, Fisher’s exact test) and allelic frequencies (adjusted P = 0.039, Fisher’s exact test). In particular, the GG genotype and the G allele were more frequent in bitter-tasters (n = 11), whereas genotype CC and the C allele were more frequent in sweet-tasters (n = 22). More specifically, 68.18 % of bitter-tasters carried the G allele, whereas 76.19 % of sweet-tasters carried the C allele at this locus (Table 2). The same segregation was observed between the sweet and the “bitter-low” tasters (n = 32), with the genotype GG (adjusted P = 0.008, Fisher’s exact test) and the G (adjusted P = 0.013, Fisher’s exact test) allele being more frequent in the “bitter-low” group. Analyses conducted with the SIFT software showed that rs2234001 caused an amino acidic substitution at residue 96, resulting in a valine–leucine change, without altering the secondary structure of the protein. However, this SNP was found to be in strong LD (r
2 ≥ 0.9) in European populations with two other non-synonymous SNPs (rs2227264 and rs2233998), both predicted by SIFT to alter the function of TAS2R5 and TAS2R4 proteins, respectively.
TAS2R4 SNP does not predict variations in stevioside bitterness perception
We also tested whether TAS2R4 rs2234001 (C/G) could predict stevioside bitterness or sweetness. Differently than TAS2R14 rs3741843, this SNP did not show variation among different levels of bitter and sweet perception (ANCOVA, P = 0.601 and P = 0.623, respectively).
TAS2R14 SNP predicts variations in stevioside bitterness perception
We found evidence that TAS2R14 rs3741843 (A/G) has a significant impact on bitterness perception. Through comparing the “bitter-low” and “bitter-high” groups, the allelic frequency of this SNP was found to be significantly different (adjusted P = 0.002; Fisher’s exact test). In particular, the G allele was more frequent in the “bitter-high” group (n = 21) compared to the “bitter-low” one (n = 32). Genotypic analyses confirmed this statistically significant different distribution, with genotype AA being more frequent in the “bitter-low” group (adjusted P ≤ 0.001, Fisher’s exact test). The same tests were repeated by removing the LMS modal group (LMS = 50), which is suspected to introduce a confounding effect, and obtained results confirmed our previous estimation (P = 0.003 and P = 0.001, Fisher’s exact test, respectively).
In addition, to test whether arbitrary classifications in taste phenotypes were plausible, an ANCOVA was performed on the subset of individuals who were able to perceive stevioside bitterness (n = 64), considering age, sex and BMI as covariates. As shown in Fig. 3, homozygote individuals with the AA genotype of TAS2R14 rs3741843 reported less bitterness from stevioside than heterozygote ones (P = 0.002). In contrast, there was no evidence that this allele predicts stevioside sweetness (ANCOVA, P = 0.621). We also found minimal evidence that stevioside bitterness is predictive of PROP bitterness (ANCOVA P = 0.081).
TAS2R38 SNPs predict variations in PROP bitterness but not in stevioside bitterness perception
As expected, PROP-tasters and non-tasters differed significantly for TAS2R38 alleles and haplotypes. In particular, TAS2R38 rs10246939 (T/C), rs1726866 (T/C) and rs713598 (G/C) were more frequent in PROP-tasters (n = 68) compared to PROP-non-tasters (n = 18) (adjusted P < 0.001, adjusted P < 0.001, adjusted P < 0.001, Fisher’s exact tests). In the same way, perception of PROP bitterness varied with TAS2R38 haplotypes. The proline–alanine–valine (PAV) homozygotes (n = 15) reported significantly more bitterness than the heterozygotes (n = 42) (ANCOVA, P < 0.001) or the AVI (alanine, valine, and isoleucine) homozygotes (n = 24) (ANCOVA, P < 0.001). Moreover, bitterness reported by heterozygotes was similar to that reported by PAV homozygotes (ANCOVA, P = 0.345). Rare haplotypes were excluded from the analyses because they are known to have intermediate phenotypes that differ from both each other and the common haplotypes. In this study, we observed both AAI (3) and AAI/AVI (2) individuals. No associations were instead found between PROP haplotypes and stevioside bitterness (ANCOVA, P = 0.945) or sweetness (ANCOVA, P = 0.812).
Comparison between stevioside bitterness and sweetness perception
Great variability in both stevioside bitterness and sweetness perception was found. Figure 4 shows this distribution of variation, with stevioside bitterness and sweetness ranging from 20 to 80 on a LMS. We therefore tried to test whether a covariation between them existed. By plotting the bitterness and sweetness of stevioside simultaneously, we did not observe a covariation between them (R
2 = 0.007, P = 0.776), with bitterness showing a decreasing trend toward the modal class and sweetness a decreasing trend followed by an increase after the modal class.
Population genetics of TAS2R4 rs2234001 and TAS2R14 rs3741843 SNPs
Alleles worldwide distribution
To better understand the differences observed for TAS2R4 and TAS2R14 SNPs, we examined their worldwide distribution. Figure 5 shows a map of allele frequencies of the two studied SNPs, based on 1000 Genomes Project data. Hardy–Weinberg test (Online Resource 3) and Heterozygosity test (Online Resource 4) were performed, confirming that all the examined populations are in Hardy–Weinberg equilibrium and no excess of heterozygosity was found (P > 0.05).
In order to identify the causes underlying the different distributions of rs3741843 and rs2234001, AMOVA based on the 1000 Genomes Project data was performed. As shown in Table 3, 19.62 % of variation at rs3741843 was due to differences among population groups (P = 0.002), 1.32 % was due to differences among populations within groups (P < 0.001), and 79.06 % was accounted by differences among individuals within populations (P < 0.001). rs2234001 showed a similar pattern, with 11.75 % of variation which was attributed to differences among groups (P < 0.001), with the remaining 88.25 % accounting only for differences observable within populations (P = 0.006).
Genetic differentiation analyses
Genome-wide pairwise Wright’s FST was calculated as a measure of genetic differentiation on 1000 Genomes populations and continental groups, and a set of relevant percentiles was extracted (Online Resource 5). FST obtained for rs3741843 and rs2234001 fell within the top percentiles (0.95) of most of the continent-wide distributions and turned out to be outlier values also for some population-level comparison.
The admixed nature of the American 1000 Genomes populations allowed us to test whether the studied SNPs could be differentially selected according to the ancestral components of these populations. The American, African and European genomic components of the 1000 Genomes American populations were extracted from the literature (1000 Genomes Project Consortium et al. 2012; Skotte et al. 2013) and used to infer the expected frequencies of the two SNPs, given the observed admixture. The African and European frequencies of the two SNPs were obtained from the relevant populations, while American frequencies were extracted from the Mexican one (MXL), after correcting for admixture. Such frequencies were then combined, according to their admixture proportions, to generate the expected frequencies of each of the American populations. Table 4 reports the observed and expected frequencies hence calculated, along with the observed excess. The difference between observed and expected frequencies is remarkable for rs3741843, where Colombians (CLM) show a defect of 22 % and Puertoricans (PUR) show an excess of 61 % on the expected frequencies.