Overall response to intermittent fasting
We prospectively enrolled diabetic adults who expressed the intention to fast during 2018 Ramadan (approximately 16 hours for 30 days in the present study) (Fig. 1a). Participants were kindly requested to restrain from overeating or binge eating at each meal during the study. Finally, thirty-six patients who completed the fasting were included (Extended Data Fig. 1a). When taken in toto, following analysis of various metabolic parameters, we observed that 30 days of intermittent fasting was associated with an overall significant decrease in body weight, body mass index (BMI), SBP (systolic blood pressure), blood glucose, blood Haemoglobin A1c and the activity of liver enzymes including ALT (alanine transaminase), AST (aspartate transaminase) and GGT (Gamma-glutamyl transferase) (Extended Data Table 1). There was, however, no significant change in serum levels of TCHOL (total cholesterol), HDL (high-density lipoprotein), LDL (low-density lipoprotein) and TRIG (triglycerides). Thus overall, intermittent fasting had a beneficial if temporary effect on metabolism in type 2 diabetes.
Definition of a responding and non-responding phenotype following intermittent fasting
The expectation was that a fraction of the cohort would show benefit from the intermittent fasting (responders), whereas another fraction would show no benefit (non-responders) allowing dissection of those lifestyle and microbiome factors that drive success in this respect. To address the issue of whether in our cohort there was a dichotomy: responders versus non-responders to intermittent fasting, we performed a post-hoc analysis on seven aforementioned parameters in which our overall cohort showed significant changes (body weight, BMI, blood glucose, SBP, ALT, AST and GGT). A Markov analysis driven by whether a parameter improved or did not improve (irrespective of the size of the effect) was performed and transformed into a dichotomous data set. A visualisation is shown in the heatmap with an agglomerative hierarchical clustering (Fig. 1b). Two evident clusters emerge: the right-side one of the heatmap is composed of patients who improved in all parameters (n = 8) or glucose plus two of liver enzymes (n = 6) or at least all three liver enzymes (n = 5) and this group was as the responder cluster of participants, whereas the left-side cluster (n = 17) showed little improvement and thus was denominated as non-responders.
When responders and non-responders were compared with respect to changes of metabolic parameters (Fig. 1c), improved insulin resistance and liver function are the most important effects and typically represent the remission from diabetes as well as one of its extremely common morbidities - metabolic dysfunction-associated fatty liver disease. The value of these parameters for each individual is provided in Extended Data Fig. 1b. The most straightforward explanation that differences in lifestyle during the fasting period drove the success or failure of the intervention.
To confirm adherence to the intermittent fasting protocol, serum creatinine was measured and was considered as a fasting-sensitive biomarker as it has been showed to be increased almost in all fasted volunteers but not in those without fasting18. As expected, creatinine level was substantially increased in both responders (p < 0.0001) and non-responders (p < 0.0001) (Extended Data Fig. 1c), demonstrating that adherence to the fasting protocol our cohort does not explain differences between responders and non-responders.
Green tea consumption drives success of intermittent fasting
An obvious potential key determinant driving success of failure is alternative food consumption during the time that food intake occurred during the study. To identify such potential key dietary factors influencing the response to intermittent fasting, monthly dietary intake was assessed using a modified food frequency questionnaire (FFQ), identical to the methodology described elsewhere17. A total of 33 food items were semi-quantified in this study. They were, according to their popularity during intermittent fasting, classified into three categories: most popular (consumed by over 94% patients during intermittent fasting), less popular (consumed by over 80–33% patients during intermittent), and sporadically consumed food items. As the later were only being consumed by less than one third of subjects, they were discarded from further analysis.
To identify possible factors explaining the differences between the groups, we carried out a longitudinal analysis and it reveals that only green tea consumption was highly divergent between responders and non-responders, while the change in vegetables, beef and sheep, and wheat-based food kept the same between groups (Fig. 2a). The success of intermittent fasting corresponded to a 76 mL increase in average green tea consumption in responders versus a 662 mL decrease in non-responders (p = 0.008) (Extended Data Fig. 2a). For these less popular food items (Fig. 2b), the prevalence of fruit, chicken, tuber, and yogurt were reduced among non-responders, but not responders, during intermittent fasting compared to that of before and after fasting, while fish and nuts were reduced only in responders. The amount of rice and tofu intake was reduced in both groups. Notably, the change of cumulative energy intake as well as cumulative liquid consumption (Extended Data Fig. 2b) was essentially identical in both groups and thus did not account for the differences in reaction to intermittent fasting.
Logistical regression curve analysis without control for covariates showed that the change of green tea consumption during intermittent fasting was significantly associated with the likelihood of being a responder (absolute change: R = 0.45, p = 0.0068) (Fig. 2c). Multivariable-adjusted logistic regression model further confirmed the association (OR = 1.82 [95% CI: 1.17, 2.84], p = 0.008) (Fig. 2d), suggesting an association independent of covariates (such as anti-diabetic medication use). These associations reversed upon the discontinuation of intermittent fasting (Extended Data Fig. 2c). Overall, these data show that success of intermittent fasting in patients with type 2 diabetes is associated with high consumption of green tea.
Involvement of the gut microbiome in the effect of green tea in intermittent fasting
The apparent relation between green tea consumption and intermittent fasting raises questions as to the possible mechanisms explaining this association. Previously, we showed in a cohort of healthy volunteers that intermittent fasting had profound effects on the microbiome, which appeared to be linked to metabolic parameters like GGT17,19. Thus encouraged we tried to establish a relation between alternative microbiome composition, effects of intermittent fasting in type 2 diabetes and green tea consumption. For this, we analysed faecal microbiota data derived from 16S rRNA gene sequencing of longitudinal samples collected at different time points during the study. The effects on microbiome composition at phylum level in responders during intermittent fasting were very substantial (Fig. 3a) and were dominated by a remarkable increased relative abundance of Firmicutes (from 54.8% at day 0 to 71.6% at day 30; i.e. a net increase of 16.8%, p < 0.001) and a significantly decreased Bacteroidetes content (from 25.1% at day 0 to 11.9% at day 30; a net decrease of 13.2%, p = 0.016) (Fig. 3b). By contrast, no significant change was observed in these two phyla in non-responders. Thus, the success of intermittent fasting strongly relates to specific changes in the microbiome.
A relation between these chances and green tea consumption is supported by a linear regression model. The change in Firmicutes level is closely associated with green tea consumption (Fig. 5d). No such association emerged in the absence of intermittent fasting suggesting that the interaction between fasting and green tea consumption is important here. The implication of our findings is that beneficial changes in the gut microbiome, potentially important for the effects of intermittent fasting in type 2 diabetes require the consumption of green tea, which in turn translates in an actionable life style advice for patients with type 2 diabetes who consider submitting to intermittent fasting for improving metabolic health.
Validation
To validate the association of green tea with the metabolic benefits of intermittent fasting, we analysed a second cohort of patients with hypertensive diabetic patient (n = 33, containing 50% of prediabetes) of which twenty-two performed intermittent fasting in the 2019 Ramadan (the remainder serving as non-fasting controls) and were all recruited in Linxia city, China (Fig. 4a). Detailed geographic information is provided in Extended Data Table 2. By analysing the same metabolic parameters as in the first cohort, four (weight, BMI, AST and GGT) were found to be improved during intermittent fasting. Compared to non-fasting group, 77.3% (17/22) of patients in fasting group were responders and this number is significantly higher than the 36.4% (5/11) showing improved metabolic parameters in the non-fasting group (Chi squared test, p = 0.0213) (Fig. 4b).
Analysis was hampered by the observation that only 5 out of 22 patients were non-responders in fasting group (Extended Data Fig. 4a). Thus we resorted to a comparison within fasting group between those with decreased green tea consumption (GTCdown; n = 11) during fasting and those not showing such a decrease (GTCup; n = 11) (Fig. 4c). This analysis resulted in an average change of 1305 ± 1100 (SD) mL in GTCup group (p = 0.001) and − 700 ± 489 mL in GTCdown group (p = 0.0008), both were significantly different from non-fasting group (Fig. 4d). By using the same grouping strategy, it became apparent that the number of responders in GTCup group was higher than GTCdown group (13/19 vs. 5/16 [one subject who did not drink green tea was excluded], Chi squared test, p = 0.0284), indicating that this grouping strategy based on GTC is a bonafide approach for both groups.
With above grouping strategy, three parameters, AST, GG and SBP, were significantly improved in GTCup group but not in GTCdown group and non-fasting controls (Extended Data Fig. 4b). However, the absolute change of GGT, HbA1c and TRIG were significantly lower in GTCup group compared to GTCdown group and/or non-fasting controls (Fig. 4d), whereas LDL was higher in GTCdown group compared to non-fasting group (Extended Data Fig. 4c). Noticeably, the change of GGT was negatively associated with the change of GTC, and this association become stronger (spearman r = -0.35, p = 0.035) when excluding an outlier (Extended Data Fig. 4d). These data again demonstrated that GTC is indeed a critical factor in driving the success of intermittent fasting in metabolic disease, especially in improving liver function. The increase of serum creatinine levels in all fasting subjects again indicated good the adherence to fasting protocol (Extended Data Fig. 4e).
To determine whether apart from green tea other items associate with a benefit of intermittent fasting on metabolic health, we analysed food intake using the same methodology used for the first cohort. We observed that wheat-based food, vegetables, beef and sheep, fruit, and chicken are popular food items in this cohort, as they all were consumed by over 93% of subjects. However, except green tea, no differences in these popular food items, less popular food items (consumed by 78.8%-33.3% of patients), as well as cumulative energy were detected between the two groups (Extended Data Fig. 4g, f, h), again providing robust evidence that only GTC, and no other consumed item, is the critical factor in determining the success of intermittent fasting. Lastly, we found that the change of the gut microbiome in the second cohort is similar with what observed in the first cohort, green tea consumption together with intermittent fasting driving an increase Firmicutes abundance (Fig. 4e).