For this study, data were obtained from the Taiwan Children Health Study (TCHS), a prospective, nationwide cohort study focusing on obesity, pubertal development, and atopic diseases in adolescents. The TCHS had an open-cohort study design and comprised 3,229 children at age 11. The cross-sectional analysis herein mainly involved 2,819 schoolchildren aged 11–12 years in 2011–2012 with FFQ assessment and pubertal outcomes available (Figure 1). The detailed data collection in the TCHS was previously documented (32). Data regarding pubertal development were collected at age 11, 12, and 18. In addition, a food frequency questionnaire (FFQ) was administered at age 11. All participants received inform consent to participate in this study. The study protocol was approved by the Institutional Review Board of National Taiwan University Hospital and complied with the principles of the Declaration of Helsinki.

Puberty oucomes were assessed at age 11, 12, and 18. We used two types of children self-filled questionnaire to define pubertal outcomes: the Chinese version of the puberty development scale (PDS) (33) and the Tanner derived composite stage (TDCS) (34). Early puberty outcome was determined by the TDCS, and age at menarche or voice breaking was assessed by PDS. The TDCS contains drawings of the five stages of secondary sexual characteristics (pubic hair and breast development for girls and pubic hair and genital development for boys). The two sex characteristics (pubic hair for both genders, genitalia for boys and breast development for girls) were combined and averaged to create a single Tanner score (33). Furthermore, early puberty was defined as reaching a certain pubertal stage earlier than the median age for that stage (35), referencing large-scale population-based Chinese studies (36, 37). We validated the Chinese version of the TDCS, and the consistency between Tanner stages from children self-reports and physical inspection by pediatric endocrinologist was revealed to be high in our pilot study (weighted kappa = 0.85) (38). Age at menarche and age at voice breaking were defined according to questions in the PDS at age 11–12. Most of the children had completed pubertal growth at age 18, therefore, age at menarche and age at voice breaking were confirmed again using the PDS through participants' recall.

We assessed diet using a self-administered, semi-quantitative 71-food-item FFQ in 2011 (39). It was a modified and shorter version of the Nutrition and Health Survey of Taiwan (40) and had been used in adolescents in one previous study (41) to investigate SSB intake. The original Chinese FFQ used in nationwide survey of Nutrition intake had been proved reproducible and of high validity (39). Participating children were instructed by a research dietitian for 20 min regarding how the portion size of each food item was measured in the FFQ. Participants were asked how frequently (on average in a week), they consumed a typical portion size of a certain food during the preceding month. A typical serving size was specified as 240 ml (equal to one cup) for SSBs, fresh juice, fresh milk, flavored milk, soy milk, yogurt, and fermented probiotic drinks. SSBs include all carbonated soft drinks, sweetened milk tea, and all other sugar sweetened beverages, such as flavored milk and concentrated fruit juice. Fresh juice, fresh milk, soy milk, etc… which contained no added sugar, were not included in the SSB category. Fermented probiotic drinks included Yakult, fermented milk, and other probiotic drinks. The probiotics contents in those fermented probiotic drinks mainly included Lactobacillus spp. The common probiotics contents within the yogurt include Streptococcus thermophiles, Bifidobacterium lactis, Lactobacillus bulgaricus, and Lactobacillus acidophilus. Detailed contents of fermented probiotic drinks and yogurt are listed in Supplementary Table S3. The simple sugar content in each type of drink was calculated by using a Taiwanese food composition table as the nutrient database (Taipei, Taiwan) (42). Total sugars were classified as natural sugar and added sugar. We classified natural sugars as lactose and non-lactose. Non-lactose sugars included glucose, fructose, galactose, sucrose, and maltose. Added sugar was defined as any sugar added to the drink during its production. After summing up the sugar intake by frequency and portion size to produce grams per day, we then categorized distinct sugar intake from drinks by using percentiles (≤25, 25−75%, and ≥75%).

Confounders in the statistical models were a priori confounders based on previous research (23) with data relevant to both puberty outcomes and SSBs. Body height was measured to the nearest 0.1 cm using a wall-mounted stadiometer (MAGATA, BW-120) at each school. BMI was calculated as weight divided by (height)² [kg/m²]. Body fat was measured using a tetrapolar multi-frequency bioelectrical impedance analysis machine (IOI 353, Jawon Medical, Korea). Participants with missing covariate information were included in the model by using missing indicators (43). Moreover, all models were adjusted for BMI, parental educational level, family income, and household smoking.

Continuous variables were expressed as mean ± standard deviation and categorical variables were expressed as frequencies and percentages. The correlations between basic demographics and different SSB and drink consumption groups were examined by chi-square test (frequency with percentage) and analysis of variance (mean with standard deviation). To evaluate the repeated measurements of pubertal outcomes at age 11 and 12, we used generalized estimating equations (GEEs) (44)-Time lag model, to examine the interrelations among SSB intake and pubertal outcomes (Figure 1B). The primary advantage of a GEE is that it accounts for within-individual variations. The correlation structure assumed for repeated measurements was a working exchangeable correlation. In the Time-lag Model, predictors (SSB at age 11) were modeled using data from surveys that preceded the outcome variables (puberty outcome at age 11 and 12). Additionally, the time-lag model accounts for the temporal sequence in a possible cause-and-effect relationship (45). In addition to selected covariates, we adjusted all enlisted beverages and total energy with the use of variance inflation factor to test the collinearity among variables. Odds ratios (OR) and 95% confidence intervals (CIs) represented the effect of each drink on early pubertal development outcomes. Trend testing was performed to present the dose responsiveness. Missing covariate information of the participants was also included in the model by using missing indicators (43). All tests were two-sided with a 5% significance level. All analyses were performed using R version 3.3.2 (R Foundation for Statistical Computing).

Among 2,819 children analyzed in this study (Table 1), 472 (16.7%) children consumed 0 cup/day of SSB, while 628 (22.3%) children consumed ≥3 cup/day. The majority (1,719 children, 61.0%) of the children consumed 1–2 cups of SSB per day. One hundred sixty-one males (11.5%) and 139 females (9.8%) experienced early puberty. Compared with female participants, males tended to consume more SSBs with nearly 25.9% drinking ≥3 cup/day. Those consuming high quantities of SSB (≥3 cup/day) were more likely to hail from households with higher exposure to cigarette smoking, less parental education, and lower family income. However, no difference in BMI, body fat, and physical activity levels was observed across different SSB consumption groups. Moreover, intake of total energy, carbohydrates, additional sugar, and additional fat increased significantly with higher consumption frequency of SSBs, indicating that unhealthy eating habits might accompany higher levels of SSB consumption. Furthermore, intake of protein decreased as SSB consumption increased, suggesting that appropriate protein intake might be replaced by SSBs. However, parental educational level, family income, and household smoking revealed significantly differences in distribution between three SSB consumption groups, while they were adjusted in all models.

Using GEE modeling, we observed that a 1 cup/day increment of SSB intake was associated with voice breaking at an earlier age (β = −0.18; 95% CI =-0.29, −0.06) in male participants (Table 2). However, consuming yogurt (≥2 cups/day) was a protective factor against early puberty (OR = 0.78; 95% CI = 0.73, 0.83). By contrast, consuming fermented probiotic drinks was associated with voice breaking at an earlier age (β= −1.23; 95% CI = −3.19, −0.72). Although fresh milk seemingly did not influence the timing of puberty among boys, consuming 1 cup/day of flavored milk increased the risk of early puberty (OR = 1.05; 95% CI 1.00, 1.02) and was associated with voice breaking at an earlier age (β = −0.41; 95% CI = −0.81, −0.01).

In girls, SSB consumption was associated with an increased risk of early puberty in a dose–response manner (Table 3). Girls consuming 1 cup/day of yogurt or fermented probiotic drink exhibited an increased risk of early puberty compared with those who rarely consumed those drinks [OR = 1.03 (95% CI = 1.00, 1.05) for yogurt and OR = 1.16 (95% CI = 1.11, 1.33) for fermented probiotic drinks]. By contrast, yogurt consumption (≥2 cups/day) was a protective factor against early puberty (OR = 0.92; 95% CI = 0.89, 0.95). Similarly, consuming fermented probiotic drinks (≥2 cups/day) was associated with a reduced risk of early puberty (OR = 0.96; 95% CI = 0.94, 0.99) as well as a delayed age at menarche (β = 0.41; 95% CI = 0.24, 1.42).

To investigate which component of the selected drinks might contribute to the aforementioned association with early puberty, we examined the effects of total sugar and added sugar from selected drinks on early puberty risk after adjustment for BMI, parental educational level, family income, and household smoking (Figures 2, 3). Notably, among girls, both the intake of total sugar and added sugar predicted an increased likelihood of early puberty in a significant dose–response manner. However, neither total sugar nor added sugar could fully explain the observed associations between more frequent consumption of SSBs and early puberty among boys. Detailed associations between consumption of several sugars and puberty outcomes are listed in Supplementary Tables S1, S2. The intake of total sugar, lactose from natural food, and added sugar each increased the risk of early puberty among girls in a dose–response manner.

In this study, we identified SSB-related risk and protective factors of pubertal timing among Taiwanese adolescents. To our knowledge, this is the first study to reveal that SSB consumption is significantly associated with early puberty in both sexes; probiotic drinks to a higher amount may mitigate this risk. Our findings indicate that health care providers should increase awareness of the effects of SSBs on adolescent health and pubertal development. The novel finding that fermented probiotic drinks or yogurt have a protective effect against early puberty suggests that early puberty might be a gut–brain disease. Further experimental studies are warranted to prove the aforementioned associations.

This is the first study to identify an independent association between SSB consumption and early puberty in boys. In girls, several reports have consistently indicated the association between SSB consumption and early menarche (23, 31, 46). SSB consumption might be associated with metabolic and weight status changes that potentially affect menarcheal timing. The link between obesity and early puberty in girls is well known (38). However, the relationship between obesity and early puberty among boys varies among different ethnicities (35, 47). Several large Chinese studies have reported that obesity also promotes the earlier onset of puberty in boys (47, 48). Furthermore, recent evidence has indicated that SSBs may have a direct effect on sexual maturation in addition to increasing body weight (23). Excessive sugar consumption per se is associated with an increase in the production of insulin and IGF-1, which could enhance cell proliferation and growth (49). Evidence also indicates that IGF-1 has a direct effect, through intracellular signal transduction cascades, on the proliferation and differentiation of ovaries as well as testes (50). In summary, excessive SSB consumption may cause fat accumulation and thus increase the risk of obesity, which is a known risk factor for early puberty. Moreover, increased insulin and IGF-1 as a result of excessive sugar intake contribute to dysregulation of the reproductive axis via downregulation of SHBG. Further research may be required to identify the key regulator involved in pubertal initiation and how it is linked to sugar metabolism.

We discovered that probiotic drinks appear to be a protective factor against early puberty. This finding is consistent with research (31) that revealed that yogurt delays menarche. We argue that this protection effect may be exerted through modifying dysbiosis in the gut. A paper by Cowan and Richardson (51) indicated that probiotic treatment can restore the normative timing of vaginal opening in stress-induced early-maturing rats. Researchers have also observed that probiotics reverse stress-induced learned fear behavior and stress hormone levels, thus highlighting the restorative effects of probiotics in the context of early-life stress (52). Our findings provide human evidence to further support the crucial role of the gut–brain axis in pubertal development. One prior human study measuring the urine metabolites in patients with central precocious puberty revealed significant alterations in urinary excretion levels of gut microbial-mammalian cometabolites before and after treatment with Triptorelin, suggesting that central precocious puberty may involve an alteration in symbiotic gut microbial composition (53). Given the cumulative evidence of the influence of gut microbiota on the development and functionality of the central nervous system (54), we tentatively conclude that gut microbiota may alter the function of the hypothalamus–pituitary–gonadal axis and thus the activation of puberty. From a practical perspective, changing gut microbiota through the intake of probiotics may attenuate early puberty. On the other hand, although the probiotic content within the yogurt and fermented probiotic drinks might protect children from early puberty, the high sugar contents within these drinks should still be noted. Currently, the World Health Organization (WHO) recommends that children consume foods with low sugar to <50 g/day (55). Hence, if a child consumes more than four cups of yogurt and fermented probiotic drinks, their sugar content within these drinks would exceed the WHO's recommendation. Parents and school policymakers are suggested to provide these probiotic drinks with low or no sugar content for children undergone pubertal growth.

Studies exploring the relationships between milk or diary consumptions and early puberty yielded inconsistent results. Evidence revealed that cow milk was the most commonly consumed animal product by girl, which might influence sexual maturation during adolescence (56, 57). However, a cohort study reported that milk consumption in girls aged ≥9 years failed to predict age at onset of menarche (58). No study has reported the aforementioned association in boys. Although we did not discover significant links between milk consumption and early puberty risk in female participants, consuming flavored milk increased the risk of early puberty in boys. This could be attributed to the extra sugar intake from flavored milk.

The intakes of total and added sugar were associated with an increased risk of early puberty among girls, but not boys. Carwile et al. (23) discovered that added sugar, but not total sugar, was associated with early menarche. Added sugar, such as artificial sweeteners, was characterized to influence metabolic changes and reproductive health by activation of sweet taste receptors (59). Sweet taste receptors locate not only in the mouth cavity but also in ovaries. This was proposed as a possible mechanism why sugar intake may influence the reproductive system. Moreover, routine exposure to added sugars may alter thresholds for sweet taste perception, imparting a stronger preference for sweet food, which may yield adiposity accumulation (60). The detailed mechanism of added sugars on early puberty has not been further explored. The only relevant animal studies reported that long-term in-utero exposure of added sugar to 18-day-old mice resulted in a thicker epidermal and dermal layer in the mammary glands of the mice embryos. By the age of 4 weeks after birth, these exposed mice experienced an early production of the mammary gland. This phenomenon indirectly confirmed that long-term exposure to added sugar may prompt early puberty (61).

The TCHS cohort was nationally representative, systemically sampled, and longitudinally followed; therefore, the derived findings could provide nutritional guidance for our local population. Although same-source bias may be present in a questionnaire-based research design, a detailed FFQ, used to assess food intake, allowed us to calculate nutrient consumption and examine its effect on sexual maturation. However, we did not specify the species and doses of probiotics in our population-based epidemiological study. We were also unable to conclude the direction of the causal relationship between pubertal timing and probiotics use. Given the short time interval in which the exposures and outcomes were assessed (i.e., age 11 years for SSB or probiotic consumption and age 11–12 years for puberty), it is difficult to ascertain whether the exposure truly precedes the pubertal outcome. There could be instances where children in this study had attained puberty before 11 years; this could affect the interpretation of the study findings, which may be biased by reverse causality. Moreover, we tested several regression models in Tables 2, 3 for all exposures and outcomes. We did not specifically adjust for multiple testing, because preferences for food choices were not our primary variables of interest. Despite this, the associations found in the tables remained significant after using stringent corrections for multiple testing. Lastly, the study findings may also be biased by residual confounding from unmeasured covariates, such as water consumption that was not available in the dataset. Further investigation may be required to determine a mechanistic explanation of the association revealed in our study.