Erratum: Alternate Day Fasting Improves Physiological and Molecular Markers of Aging in Healthy, Non-obese Humans (Cell Metabolism (2019) 30(3) (462–476.e6), (S1550413119304292), (10.1016/j.cmet.2019.07.016))

Main Text. (Cell Metabolism 30, 462–476.e1–e5; September 3, 2019) Through discussions with the scientific community and with the editors, we felt it necessary to provide additional and more detailed information on our study. In principle, concerns were raised that the primary endpoint as registered on clinicaltrials.gov (NCT02673515, registered 11/2015) was not reported in the original publication, that the statistical analysis and study design were not sufficiently explained, and that the safety aspects interpreted from the data needed to be more clearly outlined. Here, we now take the opportunity to explain and amend these issues. Importantly, none of the amendments and additional data included herein alter the conclusions of the original publication. The study protocol was approved by the ethics committee in February 2015 with the primary aim “to elucidate in which extent alternate day fasting (and thereby intermittent fasting) influences human physiology in healthy individuals in both short and long term.” Study visits occurred between April 2015 and March 2017, and collection of the registered primary endpoint data, “insulin sensitivity,” was finished as planned with the last visit. Throughout the second year of our study and before the last participant's last visit (this means also before all clinical data, including the primary endpoint of the randomized controlled trial [RCT], were collected), we decided to add mechanistic and explorative studies to the investigation. Consequently, after the last visit, we started molecular analyses of the samples by using unbiased proteomics and metabolomics in order to obtain molecular insights. Although we reported one of the registered secondary endpoints (“blood pressure”) in the published paper, we were unable to finish the analysis of the remaining registered secondary outcome measure. The analyses of the secondary endpoints “mTOR,” “Ras pathway,” and “cell cycle proteins” are currently ongoing but need more time to be finished. We plan to analyze “telomere length,” “double strand breaks,” and “mitochondrial damage” in the future. Analysis of the remaining endpoint “apoptosis” had to be skipped due to technical reasons. We erroneously did not include the originally registered primary endpoint “Difference in the changes of insulin sensitivity” in the original version of our paper. As we fully appreciate now, the primary endpoint as registered should have been reported in the original publication. We apologize for this omission. We here correct this issue and present comprehensive data on insulin sensitivity measurements. The calculation of insulin sensitivity indices has been described in the designated study protocol paper in detail (Tripolt et al., 2018). Briefly, the following indices were calculated: The Matsuda Index (IS_OGTT), [Formula presented] the HOMA-IR (homeostasis model assessment for insulin resistance), [Formula presented] the QUICKI, [Formula presented] and the insulin sensitivity index (ISI), [Formula presented] The RCT data showed no statistically significant differences (neither at baseline nor at study end) in the different parameters used to assess insulin sensitivity (HOMA-IR, QUICKI, ISI, and Matsuda) when comparing the 4-week ADF (alternate-day fasting) with the control group (which received no intervention apart from the study visits and kept their normal ad libitum lifestyle) (Table S3). This accounted for all analyses, irrespective of whether non-completers were excluded from the analysis or not, or whether non-parametric or parametric statistical tests were performed (insulin sensitivity data presented here were analyzed as outlined in the statistics section of the main manuscript and in this correction). Moreover, the in-group changes from baseline to study end at 4 weeks within the ADF group were also not significantly different (Table S3). In the second arm of the trial, we additionally analyzed differences in insulin sensitivity between a cohort of long-term fasters and the control group (represented by the baseline data assessed during the RCT arm of our study). Here, we did not detect clinically and statistically relevant differences in these parameters between these two groups either (Table S4). The herein newly provided data on insulin sensitivity (Tables S3 and S4) were gathered concurrently with the other parameters assessed in the trial, which were published in the original version of the article. Please note that the new insulin sensitivity data presented here have been peer-reviewed under the direction of the editorial team. Despite a statistically significant body weight reduction in the ADF versus the control group in the 4-week intervention trial, no changes of insulin sensitivity were observed. It seems plausible to assume here that in healthy people who are already highly insulin sensitive at baseline, ADF does not further improve parameters of insulin sensitivity. Notably, the measurements were taken after a 12-h fasting period both at baseline and study end to assess effects of the 4-week intervention, rather than differences in insulin sensitivity between a 12-h and a 36-h fasting period. Further, critique was raised concerning the cohort sizes and underlying power analysis, which we want to clarify here: we had previously published a study protocol paper (Tripolt et al., 2018), which we cited in the published paper. As stated in this protocol paper, we did not compute a power analysis prior to the study because our study was designed as an explorative, hypothesis-generating trial. This was due to the lack of previously published studies on strict human ADF, which would have been necessary for robust power calculations. However, our cohorts are of comparable size to those commonly used in other human fasting and/or caloric restriction (CR) trials (in our case n = 30 per group). For instance, Trepanowski et al. (2017) studied CR and intermittent fasting in cohorts of 35 and 34 participants, respectively (no-intervention control group, 31 participants). Redman et al. (2018) used cohorts of 34 (CR) and 19 (control) participants for a CR intervention study (ca. 15% CR). Further, Sutton et al. (2018) studied the effects of early time-restricted feeding in a cohort of 8 completers (cross-over design). Moreover, we feel it is best to describe our trial design in more detail. In summary, we performed a cross-sectional analysis of a long-term ADF cohort of healthy adults (who had been performing ADF for more than 6 months on their own, prior to the enrollment in the study) and a healthy control group (none of them had any history performing ADF). After the completion of the cross-sectional study arm, all subjects of the control group were then enrolled in an RCT, where they were randomized either to ADF for 4 weeks or to continue their current ad libitum eating behavior. This means that the 60 participants that served as the control group for the long-term fasters subsequently entered a 4-week RCT. To illustrate this point more clearly, we now provide a more detailed version of the study diagram in Figure 1. We further wish to explain the following points in more detail: (1) We have replaced Figure 1 and revised the legend to provide more details to the figure and give detailed information on the exclusion criteria of participants during eligibility screening as well as the reasons for participants discontinuing the RCT.(2) We have added one supplemental figure (Figure S6) and two supplemental tables (Tables S3 and S4) to the Supplemental Information file to provide additional data.More detailed information on the analyses performed: (1) Participants in the RCT did not have any history of ADF. The main reasons for exclusion during the eligibility screening period were BMI not within the aimed for range (22.0 to 30.0 kg/m²), people had a smoking habit, or people were taking high blood pressure medication or thyroid hormone substitution.(2) RCT dropouts (2 in the control group, 1 in the ADF intervention group) were excluded from the analyses of the RCT. However, their baseline data are included in the analysis when comparing the long-term ADF group to a control group (the participants of this control group were then randomized into control and intervention groups of the RCT).(3) Other missing values that occurred, e.g., due to technical problems during one of the visits, were not imputed. In baseline and follow-up calculations, all available data (except for the three dropouts mentioned in 3) were included. Due to some additional missing values during baseline or follow-up, the n(delta) may be smaller in comparison to n(baseline) or n(follow-up), which explains the slight discrepancy of some absolute values compared to the deltas in Table 1 of the original publication.(4) The deltas were calculated as the average of (value[follow-up] − value[baseline]) for every study participant with paired measurements available (i.e., baseline and follow-up).(5) The duration of ADF performance of the participants in the long-term cohort before the enrollment was in median 19 months (IQR 7 to 44 months), now shown in Figure S6.(6) Participants in CR or fasting trials represent a selected population, as they need to be willing to participate in an RCT with harsh dietary interventions. Thus, caution is required when translating the study results to the entire general population (especially regarding motivation and adherence to the ADF regimen).With regard to the statistical data analyses performed, corrections for multiple comparisons were applied for data presented in Figures 6 and S5 (metabolomics and proteomics). Instead, no corrections were performed for data presented in Figures 1–5 and S1–S4 and Table 1 because it is often acknowledged that exploratory studies represent a special case, in which adjustments for testing of multiple parameters are not required per se (Bender and Lange 2001; Althouse 2016; Li et al., 2017). However, it should be emphasized that any explorative finding needs to be verified in subsequent controlled clinical trials, which also applies to our findings in the originally published paper. Finally, we would like to point to our concluding words in the original publication. “Even healthy adults should not perform ADF without consultation by clinicians to rule out adverse effects due to critical medical conditions. Importantly, although not directly assessed in this study, a wholesome and balanced diet is likely crucial to foster the beneficial effects caused by ADF. Thus, appreciable clinical support and a generally healthy lifestyle should be considered before starting ADF.” This is especially important with regard to the results obtained from the long-term cohort, as we cannot exclude that many more people may have originally started ADF and stopped due to unknown reasons. Obviously, these people could not be assessed in our cross-sectional study. Moreover, we have not assessed the personal reasons for starting ADF of the cross-sectional trial participants, which could have introduced an additional bias. Hence, overall safety and efficacy of ADF should be assessed by performing future large-scale RCTs with long-term follow-up. Nonetheless, those participants performing ADF for more than 6 months did not show major side effects in our study. Additionally, the first name of author Jörn Dengjel was mistakenly written as “Jöm.” Again, we want to emphasize that the newly reported details and explanations do not affect the conclusions of the published paper. The authors apologize for any confusion the original publication may have caused.