In healthy males, microbial richness was significantly higher in the TRF group than in the control group in the end of the 25-day intervention period (linear regression p < 0.005) (27). In obese adults, operational taxonomic units (OTU) Richness, which is the sum of unique OTUs found in each sample, was not significantly different between beginning of the baseline period (B1) (436 ± 105), first day of the intervention (W1) (459 ± 115), and at 12 weeks of TRF (W12) (460 ± 119) (26). Shannon Diversity was not significantly different between B1 (3.81 ± 0.40), W1 (3.88 ± 0.41), and W12 (3.97 ± 0.41), either (26).

Among healthy Chinese participants, comparisons between pre- and post-Ramadan showed that the abundance-based coverage estimator (ACE) was higher in the end of Ramadan (p = 0.026), whereas the coverage index was lower in the end of Ramadan (p = 0.039) (29). Other alpha diversity indices showed no significant differences between pre- and post-Ramadan within neither Chinese nor Pakistani group (29). In another study conducted on healthy adults, unpaired t-test revealed a significant increase in microbial richness (Observed OTUs) post-Ramadan compared to baseline levels (p = 0.016), while no significant difference was observed between the two time points in terms of phylogenetic diversity (p = 0.052) and Shannon index (p = 0.121) (20).

In a healthy young cohort without a control group, there was a significant increase in Shannon (5.17 ± 0.49 vs. 5.40 ± 0.44; p = 0.02), Simpson (0.92 ± 0.04 vs. 0.94 ± 0.02; p = 0.001), Chao1 (350.9 ± 34.5 vs. 372.4 ± 55.5; p = 0.04) and ACE (354.4 ± 34.9 vs. 378.6 ± 56.4; p = 0.01) alpha diversity indices following Ramadan, whereas PF whole tree decreased (48.68 ± 11.40 vs. 37.64 ± 10.20; p < 0.001) (31). As for the Good’s Coverage alpha diversity index, there was no significant difference between pre- and post-Ramadan (0.99 ± 0.0002 vs. 0.99 ± 0.0002; p = 1) (31). In a separate healthy middle-aged cohort reported in the same paper, the Shannon index of the TRF group showed a slight upward trend, both compared with controls (values not reported) as well as compared with the pre-TRF state (5.04 ± 0.59 vs. 5.13 ± 0.59; p = 0.46), but this effect was not statistically significant. Likewise, no significant differences were observed in the other alpha diversity indices in the middle-aged TRF group after 30 days of Ramadan; Simpson (0.92 ± 0.06 vs. 0.92 ± 0.06; p = 0.70), Chao1 (352.0 ± 34. vs. 3342.9 ± 39.7; p = 0.31), ACE (350.7 ± 31.1 vs. 342.8 ± 35.4; p = 0.34), PF whole tree (21.99 ± 2.10 vs. 21. 69 ± 2.18; p = 0.62) and Good’s Coverage (0.99 ± 0.0005 vs. 0.99 ± 0.0004; p = 0.12) (31).

In the sole study that covered ADF, the authors did not measure the alpha diversity (28). The only 5:2 diet study, which was conducted on metabolic syndrome patients, showed no significant differences in the alpha diversity at baseline between the 5:2 diet group and control group (8). The number of observed species (OTUs) did not differ significantly between baseline and post-intervention within each group. No statistically significant changes in alpha diversity in the 5:2 diet group were observed with the Shannon index (p = 0.983) or Simpson index of 1-D (p = 0.977).

The study with healthy males (27) applied a principal component analysis (PCA) to illustrate microbiome similarity in the TRF and control groups at the OTU level. Samples from the two groups clustered separately, indicating that the groups had two distinct microbiome communities in the end of the 25-day intervention period (permutational multivariate analysis of variance (PERMANOVA) test, p < 0.05).

In healthy adults, microbial community structure (β-diversity) was significantly different between pre- and post-Ramadan [unweighted unique fraction metric (UniFrac) analysis p = 0.025] (20). Likewise, principal coordinate analysis (PCoA) using the Bray-Curtis model showed that the microbial community composition exhibited substantial divergence with little overlap between pre- and post-Ramadan in healthy Pakistani participants (29). However, in healthy Chinese participants, microbial community composition shifted only slightly following Ramadan (29). Subsequent PERMANOVA tests further supported the significant differences in the Pakistani group in the beginning vs. in the end of Ramadan (p = 0.0129) (29).

In the heathy young cohort, the structure of gut microbiota differed significantly between day 0 (the beginning of Ramadan) and day 30 (the end of Ramadan) (analysis of similarities (ANOSIM) test, p < 0.001) (31). Similarly, in the healthy middle-aged cohort TRF group, gut microbiota structures differed significantly between T1 (the beginning of Ramadan; day 0) and T2 (the end of Ramadan; day 30) (ANOSIM test, p < 0.001) (31). Unlike the young cohort, the middle-aged cohort additionally included a 30-day ad libitum follow-up period after the cessation of fasting. Interestingly, the gut microbial community showed a significant trend (ANOSIM test, p < 0.001) of return toward baseline conditions after the discontinuation of fasting (T3; 1 month after the end of Ramadan), indicating that the effects of TRF are reversible (31). As for the non-fasting control group of the middle-aged cohort, no significant differences were found between T1, T2 and T3, i.e., microbiome composition did not change during the study period (31). The authors pointed out that this observation agreed with the notion that gut microbiomes tend to be stable when lifestyles are not changed.

In the sole study that covered ADF, the authors did not measure the beta diversity (28). In the 5:2 diet study with metabolic syndrome patients (8), PCoA of unweighted UniFrac distances based on OTU data from the phylotype sequencing run showed a significant shift in the microbial community compositions in the end of the intervention within the 5:2 diet group (5:2 diet baseline vs. 5:2 diet 8 weeks, p = 0.005). Conversely, PCoA of weighted UniFrac distances based on OTUs data and the relative abundances showed no significant alteration in the composition of the gut microbiota within either the 5:2 diet group or the control group.

Statistically significant changes in the gut microbiota composition are presented in Supplementary Table S1A (the beginning vs. the end of trial comparisons of fasting groups), in Supplementary Table S1B (the beginning vs. the end of trial comparisons of control groups), and in Supplementary Table S1C (the end of trial comparisons of fasting vs. control groups). Guo et al. (8), Gabel et al. (26), Ozkul et al. (20), and Ali et al. (29) as well as both the young and the middle-aged cohort of Su et al. (31) compared the gut microbiota compositions of fasting groups in the beginning vs. in the end of the intervention period. Studies that compared the gut microbiota compositions of control groups in the beginning vs. in the end of intervention period included Guo et al. (8) and the middle-aged cohort of Su et al. (31). Both studies by Zeb et al. (27, 30) compared the gut microbiota compositions of fasting groups to those of control groups in the end of the intervention period. Lastly, Cignarella et al. (28) made all three kinds of comparisons.

There was a lot of heterogeneity in the compositional changes observed in fasting groups (Supplementary Table S1A) and the bacteria which were significantly affected by IF were mostly different in each study. Additionally, if a significant change in certain bacterial taxa’s abundance was observed in multiple studies, the changes were sometimes opposite in direction. At phylum level, for example, Bacteroidetes were found increased in Ozkul et al. (20) and the Pakistani group of Ali et al. (29), but conversely, decreased in the Chinese group of Ali et al. (29) and the young cohort of Su et al. (31). Similarly, contradicting results were found regarding Firmicutes, as they were found increased in the young cohort of Su et al. (31), but decreased in Ozkul et al. (20) and the Pakistani group of Ali et al. (29). Likewise, there was some discrepancy concerning Proteobacteria, as they were increased in the Chinese group as well as the total participants (Chinese+Pakistani) of Ali et al. (29) and in the young cohort of Su et al. (31) but decreased in Ozkul et al. (20). The study by Gabel et al. (26) is not mentioned in Supplementary Table S1A, as it showed no significant differences in community composition between B1, W1, and W12.

By considering only bacteria whose abundance significantly increased/decreased in three or more fasting groups, the abundance of Proteobacteria (at phylum level), Gammaproteobacteria (at class level), Clostridiales (at order level) and Faecalibacterium (at genus level) were increased. Meanwhile, Negativicutes (at class level), Selenomonadales (at order level) and Veillonellaceae (at family level) were decreased.

Regarding potential compositional changes in control groups (Supplementary Table S1B), in Guo et al. (8) a linear discriminant analysis (LDA) showed that some bacterial taxa were differentially abundant between baseline and 8 weeks within the control group (LDA-score > 3.00). Contrastingly, in the middle-aged cohort of Su et al. (31), LDA coupled with effect size measurements (LEfSe) showed that there were no differentially abundant taxa in the non-fasting control group either immediately after or one month after the cessation of fasting (LDA-score threshold value 4.00), i.e., no taxa were significantly changed during the study period.

As shown in Supplementary Table S1C, both studies by Zeb et al. (27, 30) found that the fasting and control groups had significant differences in their gut microbiota compositions in the end of the study period. However, since in both these studies the stool samples were collected only once in the end of the trial, it remains unclear whether the groups might have had significant differences in their gut microbiota compositions to begin with.

Cignarella et al. (28) compared the relative abundances of four major phyla, Bacteroidetes, Firmicutes, Actinobacteria and Verrucomicrobia, between the beginning and the end of the intervention period. No bacteria were significantly different at day 15 between the ADF and control group, but the abundance of Faecalibacterium, Lachnospiracea incertae sedis and Blautia showed an increasing trend after 15 days of ADF.

Potential confounders were identified based on both existing knowledge of the literature and discussions between the members of our review group. Using the largest publicly available 16S dataset of the gut microbiota [from the American Gut Project (AGP)], Vujkovic-Cvijin and colleagues have listed the following potential confounders which have a strong association with the composition of the gut microbiota: BMI, sex, age, geographical location, frequency of alcohol consumption, bowel movement quality, and dietary intake frequency of various food types (39). In addition to these, ethnicity, use of pre- or probiotics, use of antibiotics and socioeconomic status were identified as potential confounders.

Because some confounding domains may not be directly measured, investigators can measure specific variables in an attempt to fully or partly adjust for these confounding domains (40). In case of socioeconomic status, income and education can be used to adjust for it, as it cannot be directly measured. As diet is a central factor that affects gut microbiota composition (32, 33), socioeconomic status was included on the list of confounding domains with the assumption that it influences dietary choices.

In this review ‘diet’ was defined as a confounding domain by the following measurable variables: energy intake, macronutrient intake (fat, carbohydrate, protein, dietary fiber), and the intake of different foods/food groups. It should be noted that these variables are often measured with questionnaires, interviews or food diaries, i.e., with methods that are based on self-report and, if retrospective, rely on memory. This might lower the reliability of these measures.

Regarding confounding due to diet, there are a few things to note. To avoid baseline confounding, it would be best if the IF and control group had similar diets, so that the groups would be as comparable as possible to begin with. Furthermore, during the intervention, it would be important that the groups maintain a similar diet compared to baseline and also compared to each other. This would facilitate distinguishing the independent effect of IF itself on the gut microbiota, which may be confounded by between-group differences or within-group changes in diet composition/quality.

One central difficulty in attempting to draw conclusions about the causal relationship between IF and gut microbiota richness, alpha and beta diversity, and composition based on the studies included in this review, is the bias due to confounding. As stated earlier, majority of the studies in this review were quasi-experimental. Despite this, the authors of the original studies rarely addressed concerns related to confounding, i.e., what confounders they controlled for, if any, or what method they used for this, hence making it difficult from our side to judge whether an appropriate analysis method that controlled for all the important confounders was used. Thus, it is very difficult to evaluate to what extent bias due to confounding may be present in the studies, and whether the observed intervention effects truly arose from the intervention itself or if they were at least partly explained by confounding factors.

In many—although not all—studies included in this review, IF induced changes in gut microbiota richness, alpha and beta diversity, and composition. It is thus important to aspire to understand the underlaying mechanism(s) explaining these changes. We propose that the changes could be, at least in part, driven by weight loss or dietary changes that may happen during an IF intervention, as both weight and diet substantially affect gut microbiota.

Body weight and adiposity have been reported to affect the Firmicutes-Bacteroides ratio with individuals with obesity having a greater Firmicutes/Bacteroidetes ratio (41, 42). It has also been found that individuals with obesity have other compositional differences compared with lean individuals (41). Diet and dietary components can strongly influence the gut microbiota composition and are among the most important contributors to its alteration (32, 33).

Based on the studies included in this review, it is difficult to conclude to which extent IF-induced changes in the gut microbiota may be explained by changes in body weight and/or BMI. In future IF studies analyses need to be performed to examine whether weight loss has any significant mediating effect on the changes in gut microbiota richness, alpha and beta diversity, and composition.

As for dietary information, apart from Gabel et al. (25) and Ali et al. (29), the studies did not report pre- and post-intervention dietary measurements in sufficient detail to be able to evaluate whether diet composition/quality changed during IF, and if this could at least partly explain the observed changes in gut microbiota. Furthermore, as the observations by Ali et al. (29) point out, substantial changes in the foods/food groups consumed are not necessarily reflected on the macronutrient intakes, as even very similar macronutrient intakes can consist of very different foods. Thus, it would be advisable that future IF studies collect and report not only energy and macronutrient intakes of the participants, but also the intakes of specific foods and/or food groups.

Further research is also required about possible gender differences regarding IF interventions and possible resulting weight loss and gut microbiota changes. In a study by Domaszewski et al. (43) it was seen that TRF may have gender-related effects on body composition and that gender differences could be driven by sex specific differences in insulin and adrenaline secretion after fasting. A recent study by Khan et al. (7) evaluated the impact of TRF on BMI and gut microbiota outcomes in groups of overweight/obese, normal weight and underweight males and females. Interestingly, female subjects who were underweight gained weight during TRF intervention, showing for the first time that IF regimen could potentially normalize body weight. More research about the relation of gender and gut microbiota changes in response to IF is needed.

As stated in the previous sections, different forms of IF seem to induce changes in gut microbiota richness, alpha and beta diversity, and composition, and might also cause changes in body weight/BMI and energy, macronutrient and food/food group intakes. However, based on the evidence included in this review, it cannot be said to which extent the latter may explain the former.

One question that arises is whether IF would still affect gut microbiota-related outcomes if the subjects’ weight and diet composition remained the same, and which could be the mechanism behind this. At least for TRF studies it could be possible that the change in feeding-fasting pattern would be enough to independently provoke changes in the gut microbiota. A mice study has demonstrated that TRF affected the gut microbiota composition even when mice were fed with the same diet (44). In this study, mice received the same feed with either ad libitum access or during an 8-h feeding window. Future studies are needed to confirm this observation in humans.

In Western culture, the normal daily meal distribution is three to five meals, spread from breakfast to late dinner (45), which means that energy and nutrients are constantly available for the gut microbiota. Furthermore, it has been reported that an overnight fast of 8-10 h is normal for most people (46). In TRF, the fasting interval – whether it occurs at day or night – is considerably longer, e.g., 16 h. Hypothetically, this could favor the growth of certain bacteria. Ozkul et al. (20) found that the species Akkermansia muciniphila was enriched in the end of Ramadan fasting. It was stated that as a mucin degrading bacterium, A. muciniphila strongly adheres to the mucus layer, and can resist environmental changes, such as decreased food intake and changes in the intestinal flow rate. In the middle-aged cohort of Su et al. (31), the family Lachnospiraceae were increased in the end of Ramadan fasting. The authors reported that many members of this family appear to have the capacity to ferment mucins and suggested that this ability provides Lachnospiraceae a competitive advantage during IF when other carbohydrates are unavailable as an energy source to the gut microbiota for extended time periods.

Concerning circadian rhythms, Ozkul et al. (20) point out that the feeding-fasting pattern in Ramadan fasting is not compatible with human circadian rhythms. Future human studies are needed to investigate the role of either maintaining or disrupting daily circadian rhythms in TRF interventions, and whether changes in the gut microbiota may result from changes in circadian rhythms.

Alpha diversity, the microbial diversity of an ecological community, is the most common indicator for assessing gut microbiota health in adults (48). In general, a high diversity provides the ecosystem with strong stability as well as its ecological function (49). Higher alpha diversity has also been associated with numerous health indicators suggesting that gut microbiota heterogeneity may play a role in responses to dietary and lifestyle interventions (50). Meanwhile there is mounting evidence of lower diversity levels being observed in several acute and chronic illnesses (51). A higher diversity microbiota typically includes large proportions of anaerobes, whereas when diversity is reduced, facultative anaerobes, including phyla such as Proteobacteria and Bacilli, increase (52).

For compositional changes, the bacteria with most consistent evidence are focused herein, i.e., the bacteria whose abundance significantly increased/decreased in three or more fasting groups. As previously mentioned in the results section, the abundance of Proteobacteria (at phylum level), Gammaproteobacteria (at class level), Clostridiales (at order level) and Faecalibacterium (at genus level) were increased. Meanwhile, Negativicutes (at class level), Selenomonadales (at order level) and Veillonellaceae (at family level) were decreased.

Of the eight studies included in this review, only Guo et al. (8) measured gut microbial metabolites. In this study three circulating gut-derived metabolites were analyzed: lipopolysaccharide (LPS), short-chain fatty acids (SCFAs), and trimethylamine N-oxide (TMAO). It was shown that the 5:2 diet improved plasma SCFAs and plasma LPS but not plasma TMAO.

In future research, it would be important to place more focus on measuring various gut microbiota-derived metabolites, as they are known to have several different local and systemic effects (76, 77). SCFAs, for example, modulate different processes including cell proliferation and differentiation, hormones secretion and activation of immune/inflammatory responses (78). Butyrate, one of the three major SCFAs produced by gut bacteria along with acetate and propionate (79), has been reported to improve gut barrier function by facilitating the assembly of tight junctions (80).

In the studies included in this review, the intervention durations ranged from 15 days to 12 weeks. It would be interesting to examine how longer-term IF interventions affect the gut microbiota, or in general, if the effects on gut microbiota differ depending on the IF duration. Furthermore, the resiliency of IF-induced gut microbiota modifications should be studied more. Out of the eight studies in this review, only the middle-aged cohort of Su et al. (31) examined this by including a 30-day ad libitum follow-up period after the cessation of fasting. The results of this study suggested that the gut microbiome composition returns to baseline upon cessation of TRF. Thus, it should be further investigated if one has to adopt IF as a continuous lifestyle in order to gain beneficial health effects from it, or if not, how often one should fast (e.g., for one month a few times a year). The role of maintaining or disrupting daily circadian rhythms and the significance of the feeding-window’s timing (in daytime vs. in night-time) also require further investigation, as previously discussed.