Effect of Intermittent versus Chronic Calorie Restriction on Tumor Incidence: A Systematic Review and Meta-Analysis of Animal Studies

Both chronic calorie restriction (CCR) and intermittent calorie restriction (ICR) have shown anticancer effects. However, the direct evidence comparing ICR to CCR with respect to cancer prevention is controversial and inconclusive. PubMed and Web of Science were searched on November 25, 2015. The relative risk (RR) [95% confidence interval (CI)] was calculated for tumor incidence, and the standardised mean difference (95% CI) was computed for levels of serum insulin-like growth factor-1 (IGF-1), leptin, and adiponectin using a random-effects meta-analysis. Sixteen studies were identified, including 11 using genetically engineered mouse models (908 animals with 38–76 weeks of follow-up) and 5 using chemically induced rat models (379 animals with 7–18 weeks of follow-up). Compared to CCR, ICR decreased tumor incidence in genetically engineered models (RR = 0.57; 95% CI: 0.37, 0.88) but increased the risk in chemically induced models (RR = 1.53, 95% CI: 1.13, 2.06). It appears that ICR decreases IGF-1 and leptin and increases adiponectin in genetically engineered models. Thus, the evidence suggests that ICR exerts greater anticancer effect in genetically engineered mouse models but weaker cancer prevention benefit in chemically induced rat models as compared to CCR. Further studies are warranted to confirm our findings and elucidate the mechanisms responsible for these effects.

Inclusion and exclusion criteria. The inclusion criteria comprised the following: 1) the study was an intervention study; 2) the study used mice or rats as subjects; 3) the sample size in each group was at least 8; 4) the trial duration was at least 2 weeks; 5) the study examined the anticancer effect of ICR (intermittent fasting/ alternate-day fasting) vs. CCR; 6) the study with any effect size for which 95% confidence intervals (CIs) were provided or such information could be derived; and 7) the primary endpoints of the study were tumorigenesis rate or number of tumors, tumor weight, age at detection, insulin-like growth factor-1 (IGF-1), leptin or adiponectin.
The exclusion criteria included the following requirements: 1) the study was an in vitro or a human study; 2) the study was an animal study but the subject was neither mice nor rats; 3) the study was not an intervention study; 4) the study combined CCR/ICR with other factors such as exercise, nutrition supplements, radiation or pharmaceuticals, etc.; 5) the trial duration of the study was less than 2 weeks; or 6) the study did not report any effect size with its 95% CI or such information could not be derived.
Study selection and data extraction. The titles and abstracts of the obtained studies were first reviewed independently by two investigators (Y.C. and L.L.) to determine whether they met all of the inclusion criteria. Then, the full texts of the potentially included studies were investigated independently with reference to the inclusion and exclusion criteria. Two reviewers (Y.C. and L.L.) independently appraised each included article according to the Systematic Review Centre for Laboratory Animal Experimentation's RoB tool, which is based on the Cochrane RoB tool and specifically designed for animal intervention studies 40 . This tool contains 10 entries (Supplementary Table 1) related to the following 6 types of bias: selection bias, performance bias, detection bias, attrition bias, reporting bias and other biases. Higher quality scores represent a lower risk of bias. Scores of 0-3, 4-7 and 8-10 represent high, moderate, and low risks, respectively. This study followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. The checklist can be found in Supplementary Table 2. Data from the included studies were extracted by the two investigators (Y.C. and L.L.) using standardised and piloted design formats. Discrepancies in the process of study selection and data extraction were resolved through a group discussion with two other authors (P.X. and G.X.).
The data were independently examined and adjudicated after being extracted and assessed; several values were extracted from the results of original graphs in studies in which data were not provided directly in the text or tables using GetData Graph Digitizer 41,42 . The major outcomes and conclusions were extracted from each study using preset data recording forms. Baseline characteristics of the included studies are given in Supplementary  Table 3. The information includes animal type, tumor type, feeding regimen, trial length and body weight at the end of the follow up. Tumor characteristics, including the number of subjects with tumors, age at detection, number of tumors per animal and tumor weight in each model and are shown in Tables 1 and 2. In addition, three tumor-related factors, including the hormones IGF-1, leptin, and adiponectin, were extracted. Statistical analysis. Tumor incidence was compared between ICR and CCR group in genetically engineered and chemically induced models, respectively. Based on the information extracted from each included study, the relative risk (RR) of developing a primary tumor was calculated as p 1 /p 0 , and the corresponding 95% CI was calculated as 43  where p 1 and p 0 are tumor incidences in ICR and CCR group, respectively; and n 1 and n 0 are the related sample size in each group. Then the RRs (95% CIs) were transformed into their natural logarithms to stabilize the variances and normalize their distributions. The pooled RRs with theirs 95% CIs were calculated by a random-effects model weighting for the inverse of the variance 44 .
The heterogeneity among studies was tested by Cochran's Q test and quantified by the I 2 statistic. An I 2 value of < 25%, 25-< 50%, 50-< 75% and ≥ 75% represents very low, low, moderate, and high heterogeneity, respectively 45 . A P-value of ≤ 0.10 was considered statistically significant. Subgroup or meta-regression analyses were used to identify possible sources of heterogeneity. Publication bias was assessed by Egger's regression asymmetry test (when the numbers of studies was ≥ 3) or Begg's asymmetry test (when the numbers of studies was < 3) with a significance level of 0.10. The Duval and Tweedie nonparametric "trim and fill" method was used to adjust for publication bias if needed 46 .
In addition, the standardised mean differences (SMDs) and 95% CIs were computed to compare the differences between the two groups (ICR vs. CCR) in the levels of serum IGF-1, leptin, and adiponectin that may be involved in the development of the tumor. Other major characteristics including tumor weight, number of tumors/animal, and age at detection, were compared using a similar approach between two groups.
In the sensitivity analyses, the influence of each included study on the pooling was examined by omitting one study at a time, and a random-effects model was replaced with a fixed-effects model to evaluate whether the model selection substantially affected the pooled results.
All analyses were performed using STATA (Version 14.0; STATA Corporation LP, College Station, Texas, USA). A two-sided P value of ≤ 0.05 was considered statistically significant, if not otherwise specified.

Results
Eligible studies. The flow of the search strategy followed PRISMA and is shown in Fig. 1. A total of 2,673 studies were identified, 1,166 from Pubmed and 1,507 from Web of Science. In addition to the 904 duplicated studies, 1,646 studies were excluded after reviewing the title and abstract, and the details are documented in Fig. 1. Moreover, a total of 109 studies were excluded after full-text reading according to the inclusion criteria for one of the following reasons: human studies (n = 2), reviews, letters to editor, abstracts (n = 8), not a direct comparison of ICR with CCR (n = 79), or no concrete measures for cancer (n = 20). Furthermore, two studies were found from the relevant reference lists. Thus, a total of 16 animal studies were included in this meta-analysis. The score of quality assessment of the 16 studies using the Systematic Review Centre for Laboratory Animal Experimentation's RoB tool ranged from 5 to 7 and all represent moderate quality and risks except for an unpublished study.
As presented in Tables 1 and 2 and Fig. 2, a total of 908 mice (469 in the ICR group and 439 in the CCR group) with 374 events (161 in the ICR group and 213 in the CCR group) were included in the genetically engineered models. Comparing ICR to CCR, the pooled RR (95% CIs) of tumor incidence was 0.57 (0.37, 0.88) with high heterogeneity (I 2 = 89.7%, P < 0.01). Because strong evidence of publication bias was observed (Egger's test: P < 0.01), the adjusted pooled association was 0.66 (0.50, 0.88) using the Duval and Tweedie method.
In chemically induced models, a total of 379 rats (216 in the ICR group and 163 in the CCR group) were included, and the total events were 128 and 73 for the ICR and CCR group, respectively. The pooled RR (95% CIs) of tumor incidence comparing ICR to CCR was 1.53 (1.13, 2.06), with moderate heterogeneity (I 2 = 55.2%, P = 0.06). Because publication bias was documented (Egger's test: P = 0.052), the pooled association was adjusted as 1.33 (1.02, 1.74) using the Duval and Tweedie method.
Effect of ICR versus CCR on levels of IGF-1, leptin, and adiponectin. Five 22,26,27,29,31 out of the 11 studies (Fig. 3) that used genetically engineered models compared levels of IGF-1 between ICR-R and CCR. All of them focused on 2 types of hormone-sensitive cancers -mammary and prostate cancer. The pooled SMD was  Only 1 study that used a chemically induced model had data available on IGF-1 levels, and none of the studies reported data on levels of leptin, and adiponectin, which limited our ability to pool them quantitatively.

Effect of ICR versus CCR on other tumor characteristics.
All of the other pooled statistical effects of tumor characteristics and tumor relative indexes are displayed in Table 3.
In the studies utilizing genetically engineered models, three studies have complete data descriptions (mean ± SE/SD) on tumor weight 22,26,34 , number of tumors/animal 22,27,34 , and age at detection 22,26,34  In the chemically induced models, two studies reported information on the number of tumors/animal 38,39 . No significant difference was found [0.63 (− 1.31, 2.57)], and publication bias (Begg's test: P = 1.00) was not evident. However, high heterogeneity was observed (I 2 = 95.2%; P < 0.01).   Table 4). Omitting 1 study each time and recalculating the pooled RRs/SMDs for the rest of the studies showed that none of the single studies substantially influenced the pooled RR for tumor incidence or the pooled SMDs for the other continuous outcomes (Supplementary Table 5).

Discussion
The main findings of this study indicate that ICR showed a greater anticancer effect in genetically engineered mouse models but a weaker cancer prevention benefit in chemically induced rat models as compared to CCR. The decreased IGF-1 and leptin and increased adiponectin in genetically engineered models supported our main findings.
Compared to AL intake, anti-tumor benefits from CCR and ICR have been reported for breast, colon, liver, skin, and lung tumors in rodent models 32,[47][48][49] . CCR of 30% or greater energy reduction consistently reduces tumor incidence in spontaneous 50 , chemically induced 47 and radiation-induced tumor models 51 . However, no general conclusion could be drawn regarding the tumor inhibition of ICR compared with CCR.
It is worth noting that our findings (i.e., a greater anticancer effect in genetically engineered models and a weaker benefit on cancer prevention in chemically induced models relative to CCR) are consistent with those suggested by Thompson et al. 52 . The decreased IGF-1 and leptin and the increased adiponectin levels in genetically engineered models reflect the significantly superior tumor inhibition of ICR compared to CCR. Opposite inhibition effects can be observed when comparing ICR to CCR in the 2 animal models, indicating that CR may have different mechanisms when different tumor models are applied.
Although the exact mechanisms of the anticancer effect of CCR are debatable, it is widely believed that CCR prevents tumorigenesis by decreasing metabolic rate 53 and promoting protective mechanisms that allow DNA damage to be prevented 54 . Simone et al. 55 suggests that the mechanism behind ICR is relatively simple: it postpones tumor growth by starving tumors from glucose for a short period of time. A modified diet of increased protein and fat and decreased carbohydrates in the ICR group (similar to ketogenic diets) may account for a large proportion of the effects 56 . However, in addition to the widely studied dysregulated glucose metabolism to fuel tumor cell growth, accumulating evidence suggests that utilisation of amino acids and lipids also contributes significantly to cancer cell metabolism 57,58 . Whether these factors play similar roles in tumor inhibition with different models or whether they produce different suppression effects remains unclear.
Tumor suppression in genetically engineered models. For genetically engineered models, a review by Varady et al. 59 in 2007 suggested a protective effect of ICR on cancer risk, which supports our findings. However, we found no statistically significant difference in the number of tumors per animal and tumor weight comparing ICR to CCR group, which suggests that ICR may play a protective role only in the early stage of tumorigenesis. We speculate that this limitation is partially due to the more severe late tumor development in the ICR group.
Research shows that the IGF-1 receptor and the insulin receptor may differ mechanistically in subjects undergoing ICR compared with those undergoing CCR and this may result in greater reductions in hepatic and visceral fat stores, IGF-1 levels, leptin and cell proliferation, and increase insulin sensitivity and adiponectin levels 26,31,60 .
A growing body of evidence suggests that insulin and IGF-1 receptors regulate cell proliferation, differentiation, apoptosis, glucose transport, and energy metabolism by regulating downstream signalling cascades through insulin receptor substrate molecules. CR in rodents reduces IGF-1/insulin-phosphatidylinositol-3 kinase-Akt-mammalian target of rapamycin complex 1 signalling, which has been shown to be correlated with significant tumor growth delay 61,62 .
Nevertheless, although CR generally reduces the levels of IGF-1 and leptin, their absolute values were higher in the ICR group than in the CCR group in most of our studies 23,28,32,34 , which indicates an inverse correlation between the levels of IGF-1 and leptin and tumor occurrence. From several studies that reported information stratified by period (ICR-R vs. ICR-RF) 22,26,27,29,31 , we were able to explain the above results. Although the levels of IGF-1 and leptin during the ICR-R period were much lower compared to the CCR group, they increased substantially during the ICR-RF period and became much higher than those in the CCR group. This could explain the higher mean levels of IGF-1 and leptin in the ICR group than those in the CCR group (Table 1).
The sharply reduced serum IGF-1 and leptin and the elevated adiponectin and adiponectin/leptin ratio (data not shown) were associated with the protective effects of ICR in genetically engineered models. Furthermore, previous studies have characterised IGF-1 and leptin as mediators of the anticancer effects of CR 63 .  Table 3. Standardised mean differences in other tumor relative indexes comparing ICR to CCR in two animal models. All the pooled estimates were obtained using a random-effects model. CCR: chronic calorie restriction; CI: confidence interval; ICR: intermittent calorie restriction; SMD: standardised mean difference.
Scientific RepoRts | 6:33739 | DOI: 10.1038/srep33739 Leptin is an activator of cell proliferation and anti-apoptosis in several cell types and an inducer of cancer stem cells; its critical roles in tumorigenesis are based on its oncogenic, mitogenic, proinflammatory, and proangiogenic actions 64 .
Leptin enhances proliferation of human cancer cell lines 65,66 . In contrast, adiponectin reduces cancer cell proliferation 67,68 , which is verified by the results of our study. Findings from two human studies 68 and one in vitro study that evaluated the impact of different adiponectin/leptin ratios on human breast cancer cell proliferation 69 suggested that the adiponectin/leptin ratio may be more important in determining how these two proteins together affect mammary tumor development than either one alone. Two studies of ICR-R vs. CCR found that ICR can promote a better adiponectin/leptin ratio than CCR (15.5 ± 5.6 vs. 3.8 ± 0.8 31 ; 7.96 ± 2.6 vs. 2.85 ± 0.3 27 ). The reduced serum leptin and elevated adiponectin/leptin ratio were associated with the protective effect of ICR 63 .
Tumor suppression in chemically induced models. Chemically engineered models are induced by chemical carcinogens such as polycyclic aromatic hydrocarbon (PAH), N-nitrosamines and mycotoxin 70 . Most of the studies included in this meta-analysis used the carcinogen DMBA, a type of PAH known to cause mammary tumors in rats. The formation of PAH-DNA adducts (DNA binding products), a necessary step in PAH-initiated carcinogenesis, has been widely studied in experimental models and has been documented in human tissues 71 . According to previous studies, several nutrients such as vitamin A 72,73 , C 74 , D 75,76 and E 77,78 protect against the carcinogenic effects of DMBA exposure. The relatively low intake of these nutrients during the ICR-R period could cause a reduced protective effect, which may eventually lead to a relatively high tumor occurrence. More importantly, tumor cells have evolved the ability to utilize different carbon sources due to the limited supply of nutrients. For example, glutamine, the most abundant amino acid in the plasma, has long been recognized as an alternative fuel 57 .

Advantages and disadvantages.
There are several strengths in this research that should be highlighted.
First, we collected and systematically analysed the most up-to-date comprehensive evidence and performed the first quantitative meta-analysis to compare the anticancer effects between ICR and CCR. Second, all of the included studies are intervention studies, which provide stronger evidence than observational studies. Third, to our knowledge, this is the first study conducted to identify animal models (genetically engineered models vs. chemically induced models) that study the anticancer effects of ICR versus those of CCR.
When interpreting our results, a number of issues should be considered. First, the frequency of cycling (the number of cycles experienced), the duration of the cycles and the restriction regimen (the amount of CR) varied across studies and need further unification. However, we used random-effects models in accordance with the heterogeneity, and further adjusting for these factors using meta-regression did not substantially change our main findings. Second, several studies divided the ICR group into the ICR-R period and the ICR-RF period, while others did not, which might cofound our findings. However, the likelihood of this is low, because the results were generally consistent when we restricted our analysis to the ICR-R period in the studies in which such information was available. Third, although we did not find strong evidence of publication bias in most of the pooling and we adjusted the pooled association using statistical methods when publication bias existed, publication bias due to unpublished data or publications in non-English languages may exist. Fourth, all genetically engineered studies used mice and all chemically induced models used rats, and mice may respond better to ICR than rats, which might partially explain our results or at least this possibility cannot be completely ruled out. Finally, this research focused on studies conducted in rodents, which limits the border application of the findings. Thus, the results need to be further verified in other advanced animals, e.g., mammals and primates, and in human beings. It is also important to note that the human studies examined in this review are not sufficient; the direct effect of ICR vs.
CCR on cancer has been tested only in animal models. Future studies with more reasonable experimental designs are needed to answer these important questions. Much work remains to be done to translate the knowledge gained from CR research to humans for chronic disease prevention 79 . Highlights and implications. Based on the evidences herein, we propose that, for individuals carrying several of the cancer risk genes, ICR may be a more effective choice; for the chemical carcinogen-exposed population group, CCR may achieve better results. This finding also suggests that the energy control in the ICR-R period needs to be confined to a reasonable range and that supplementation of certain protective nutrients during the ICR-R period would achieve better suppression effects, which has been mentioned in one cancer chemoprevention study 80 but requires further experimental and clinical verification. Second, it needs to be emphasized that in the presence of strong carcinogens, the excessive restriction of energy and total nutrition may lead to excessive loss of several nutrients that are beneficial to anti-cancer mechanisms and this may eventually dilute the anti-cancer effect of energy restriction. Third, original, randomized controlled trials are needed to directly compare the anticancer effect of a specific ICR regimen with that of a specific CCR regimen, considering specific tumor occurrence and development and using both genetically engineered and chemically induced models. The research on ICR needs further refinement. For instance, it is necessary to develop a clearer definition of the ICR-R and ICR-RF periods and indicators of the two periods. Studies can also be designed to compare the cancer inhibition effects of ICR and CCR using specific animal species and induced models. This will help us better answer several questions about the tumor inhibition of ICR. For example, can the acquired protective effect in the restriction period be compromised by AL intake or high fat intake in the re-feeding period? If the answer is yes, can the compromised effects be modified using different animal models (genetic vs. chemical models)? Furthermore, is the attenuation effect stronger in the re-feeding period in the chemically induced model than in the genetically engineered model? Important questions remain unanswered. For example, would greater energy restriction in the ICR group attain equivalent or superior tumor inhibition in the DMBA-induced animal cancer models than CCR? This question requires further verification in both experimental and clinical studies and leads to further questions: Is it comprised by AL intake in the re-feeding period? What is the difference between ICR and CCR in inhibiting tumors when different fuel sources are used? Can ICR provide additional cancer protective effects in human tumors as compared to CCR, and what is the long-term safety of ICR? Answering these questions will provide helpful information for dietary recommendations for tumor prevention and for weight maintenance/control in normal weight/overweight individuals.
Summary. The protective effects of ICR and CCR on tumor varied according to the animal model. Compared to CCR, ICR could prevent cancer to a greater extent in genetically engineered mouse models and to a lesser extent in chemically induced rat models. The potential difference in the mechanism of the effects of ICR vs. CCR in different tumor exposure scenarios, including genetic defects and environmental exposures, warrants further elucidation, which may facilitate the adoption of ICR for human beings.