The microbiome stabilizes circadian rhythms in the gut

To address circadian regulation of the microbiome, we started by asking if flies fed ad libitum show changes in the microbiome at different times of day. Anticipating that the microbiome would cycle, in which case addressing the clock regulation of it would be important, we assayed wild-type flies (Iso31) as well as flies that lack a key circadian clock gene, period (per). We collected feces from male and female flies at Zeitgeber Times (ZT) 0, 4, 8, 12, 16, and 20 (ZT0 = lights on and ZT12 = lights off) and subjected these to 16S rRNAseq for analysis of the microbiome (SI Appendix, Fig. S1A↗ and Dataset S1 A and B↗). Using JTK cycle analysis, cycling was not detected for three separate measurements of alpha diversity of the microbiome (Faith's phylogenetic diversity, Shannon diversity, and richness) in either genotype with ad libitum feeding. Similarly, relative abundances of specific species that comprise most of the fly microbiome did not show changes over the course of a day:night cycle (Fig. 1 A–D, AF condition, which reflects microbiome-containing flies fed ad libitum).

Although flies typically display a circadian rhythm of feeding, we speculated that this might not be robust enough to drive rhythms of the microbiome, so we subjected them to a rhythmic feeding paradigm, such that food was only available from ZT0 to ZT10 (TF condition, i.e., flies carrying a microbiome subjected to TF, in Fig. 1). Similar analysis as above revealed no circadian fluctuation in alpha diversity or taxa abundance in wild-type flies (Fig. 1 A–C, TF condition). On the other hand, Lactobacillaceae and Gluconacetobacter both showed periodicity in per¹ flies with TF, suggesting that host clocks interfere with the effects of TF on the microbiome (Fig. 1D). Aside from circadian rhythms, the overall relative abundance of Lactobacillaceae and Acetobacter was lower in per¹ flies compared with wild type, regardless of feeding conditions (SI Appendix, Fig. S1B↗).

Although the microbiome was not found to cycle in flies, even with TF, there was reason to believe, from studies in mammals, that it would impact host cycling. To determine if the microbiome, with and without TF, regulates rhythmic gene expression, we depleted female flies, chosen because of the larger amount of material available from them, of their gut microbiome (SI Appendix, Fig. S2A↗) and compared circadian gene expression, via RNAseq, in the guts of these animals with that of controls that contained microbiomes. Each condition (sterile or microbiome-containing) was additionally tested under two different feeding paradigms, ad lib and TF, to yield four groups—ad lib sterile (AS), TF sterile (TS), ad lib microbiome (AM), and TF microbiome (TM) (Fig. 2A) (Dataset S1 C–F↗).

Principal component analysis (PCA) of the top genes expressed in sterile (AS and TS) and microbiome-containing (AM and TM) flies showed a clear separation between the four groups assayed, and indicated an effect of time-of-day, in particular in the groups maintained on TF (Fig. 2B). Thus, the circadian transcriptome was distinct for each of the tested conditions. Comparison of these differentially expressed genes also revealed that transcripts enriched in sterile flies are more likely to be expressed rhythmically with TF (clusters C and D in SI Appendix, Fig. S2B↗). Nevertheless, several genes were unique cyclers in microbiome-containing flies. Regardless of the statistical cutoff for cycling, the pattern of overlap across the four groups was similar, with some genes expressed cyclically in only one group and some in multiple (Fig. 2C and SI Appendix, Fig. S2 C and D↗). Results with a less stringent cutoff for cycling are shown in SI Appendix, Fig. S2 E–H and J–M↗.

The presence of a microbiome, surprisingly, dampened circadian cycling of gut transcripts, although it did not change the overall distribution of phases for the cycling transcriptome (Fig. 2D). Enhanced cycling in sterile flies consisted of an increased number of genes expressed rhythmically and also increased amplitude of cycling (Fig. 2 E and F), and was evident in comparisons of AM versus AS and TM versus TS. Genes that gained cycling in sterile flies, under ad lib conditions, tended to encode metabolic proteins, in particular those involved in amino acid and peptide metabolism (Fig. 2G). However, less cycling was observed for genes involved in development and differentiation. Transcripts encoding proteins involved in energy metabolism cycled in sterile and microbiome-containing flies (Fig. 2G). Similar results were obtained with less stringent cutoffs for cycling (SI Appendix, Fig. S3A↗).

TF affected the distribution of phases in the rhythmic transcriptome, in addition to increasing the number of cycling transcripts and the overall amplitude of cycling (Fig. 3 A–F). This was evident in flies containing a microbiome (Fig. 3 A–C) and in sterile flies (Fig. 3 D–F). The change in phase was unexpected as the protocol was designed to match the endogenous rhythm (23). Given that feeding rhythms can vary a bit from strain to strain (24, 25), we assayed the feeding patterns of iso31 flies used here. We detected a peak in the morning, which is consistent with previous reports (23, 26) and suggests that phase differences between AF and TF are not due to altered phase of feeding (SI Appendix, Fig. S4A↗). We saw no differences between sterile flies and those that carried a microbiome, indicating that transcriptomic differences under these conditions are also not due to altered feeding (SI Appendix, Fig. S4B↗).

Transcripts that gained cycling with TF, or that cycled under both ad lib and TF conditions, encoded metabolic proteins. Interestingly, transcripts involved in mitochondrial/energy metabolism tended to lose cycling under TF conditions (Fig. 3 C and F). Heat maps using a less stringent cutoff for cycling are shown in (SI Appendix, Fig. S3 B and C↗).

Consistent with the lack of an effect of the microbiome on the distribution of phase, comparison of sterile flies with microbiome-containing flies under TF conditions did not reveal any differences in phase although more genes cycled in the sterile flies and with higher amplitude (SI Appendix, Fig. S5 A and B↗). In general, with TF, sterile flies gained cycling of metabolic genes and lost cycling of transcripts implicated in translation and transport (SI Appendix, Figs. S3D and S5C↗).

To address the overlap between the effects of the microbiome and TF, we asked if the genes that gain cycling in TM flies, relative to AM, also gain cycling in sterile conditions. Of 1,183 genes that show enhanced cycling in TM, 269 displayed enhanced cycling in sterile flies, suggesting partial but not complete overlap (Dataset S1G↗). Given some independent effects of a TF paradigm and the loss of a microbiome, a combination of these two regimens (sterile flies maintained on TF) resulted in maximal cycling. This was reflected in a very high number of cycling genes, and also higher amplitude cycling of genes that cycled across all conditions (SI Appendix, Figs. S2I, S3E, and S5 D and E↗). Notable in these shared cycling transcripts were those encoding proteins of small molecule and carboxylic acid metabolism. Changes in cycling brought about by loss of the microbiome or by TF were independent of changes in average expression levels of the affected genes (SI Appendix, Fig. S6 A–D↗), which is consistent with the findings of Thaiss et al. (7).

As both loss of a microbiome and TF predominantly affected metabolic genes, we closely examined the daily expression of a selection of metabolic genes under AM, TM, AS, and TS conditions (Dataset S1H↗). Increased amplitude with TF or with loss of the microbiome resulted either from higher peak levels or lower trough levels (SI Appendix, Fig. S7 A–P↗). For instance, Prx2540-2 showed higher peak levels with TF in both sterile and microbiome-containing flies (TS and TM), while increased amplitude of Sodh1 rhythms resulted largely from a decrease in trough levels. In general, loss of a microbiome tended to increase overall levels, which often resulted in an increase in amplitude e.g., Ninad and Sodh1.

The microbiome and TF also altered the cycling of clock genes, although probably less so than they did clock-controlled genes (SI Appendix, Fig. S8 A–F↗), and the effect varied from one clock gene to another. Loss of the microbiome increased the amplitude of cycling of tim, pdp1, and Clk, but not of cyc (SI Appendix, Fig. S8 B, C, E and F↗). Changes in the amplitude of per were not significant with the four time points assayed in the RNAseq experiment, but were evident in qPCR experiments where six time points were assayed (SI Appendix, Fig. S9A↗). TF increased the amplitude of cycling of per, cry, and cyc in sterile flies, but in microbiome-containing flies it only enhanced cry cycling (SI Appendix, Fig. S8 A, D and E↗). Clk and tim showed no effect of TF and, interestingly, pdp1 cycling showed a decrease in amplitude with TF of sterile flies (SI Appendix, Fig. S8C↗).

As changes in clock genes were variable and did not always parallel changes in the transcriptome as a whole, we asked if microbiome and/or TF-induced alterations in cycling genes were mediated by altered cycling of other transcription factors. Many transcription factors cycled in one or more of the conditions assayed—AM, TM, AS, TS—and, as in the case of other transcripts, cycling was more robust in sterile flies and also with TF (SI Appendix, Fig. S9 C–F↗ and Dataset S1 I–L↗). Examination of the distribution of phases revealed that the microbiome had little effect on phase distribution, but TF altered phases of transcription factor cycling in sterile and microbiome-containing flies (SI Appendix, Fig. S9G↗). This matches what was observed for the transcriptome as a whole (Figs. 2 and 3 and SI Appendix, Fig. S9H↗), and is consistent with other reports of changes in transcription-mediating effects of feeding (27, 28). In addition, specific metabolism-relevant transcription factors responded to TF and loss of a microbiome in ways that parallel effects on other transcripts (SI Appendix, Fig. S9 I–N↗). We suggest that global changes in cycling genes are downstream of changes in specific transcription factors effected by the reported manipulations, in particular TF.

Given the profound effects of the microbiome and TF on host gene cycling, we asked what mechanisms might mediate these effects. Histone acetylation is implicated in the impact of the mouse microbiome on host gene expression (15), so we considered it as a mechanism here. We found that TF enhanced the cycling of several histone deacetylases (HDACs) and also drove cycling of acetylated histone H4 (H4ac) in the guts of sterile flies (Fig. 4 A and B). TF had less of an effect on flies that contain a microbiome, although it still drove significant cycling (Fig. 4 A and B). Acetylated H4 was not detectably different between sterile flies and microbiome-containing flies (Fig. 4B), so we focused on effects of TF.

To determine whether histone acetylation could account for the enhanced cycling seen with TF, we injected ad lib fed iso31 flies with histone deacetylase inhibitors and examined gene expression in fly bodies, focusing on genes that showed stronger cycling with TF. The HDAC inhibitor sodium phenyl-butyrate (PBA) increased expression of H4ac (Fig. 4C) and also increased expression of many of the genes expressed cyclically with TF, implicating TF-driven rhythms of histone acetylation in the cycling of these genes (Fig. 4D). A different HDAC inhibitor, Valproic acid (VA), did not consistently increase H4ac, but still increased levels of many genes, including the transcription factor CrebA and clock genes per, tim, and Clk (SI Appendix, Fig. S10 A–C↗), perhaps through effects on other histones. Some genes were preferentially affected by one HDAC inhibitor, PBA or VA, and not the other (SI Appendix, Fig. S10D↗). Given the effects of TF on histone acetylation, together with enhanced expression of genes targeted by TF with HDAC inhibitors, we propose that histone acetylation is a major mechanism mediating effects of TF on the circadian transcriptome.

We next sought to determine the functional impact of the microbiome and TF on physiology, asking also whether beneficial effects of these were linked. For instance, TF paradigms promote metabolic health, and so we wondered if the microbiome was required for the health benefits. To establish a paradigm to visualize beneficial effects of TF, we focused on responses to stress, as these constitute a good measure of fitness. Thus, we exposed 5 to 7-d-old flies to ad lib feeding conditions or TF for 4 or 21 d (SI Appendix, Fig. S11A↗) and then treated them with different stressors—starvation, injection with bacteria, heat shock—and assayed survival. The 21-d protocol was based on a previous study where this duration was required for beneficial effects of TF (29). Surprisingly, while the 4-d TF had no effect on the response to starvation, TF for 21 d actually decreased survival (SI Appendix, Fig. S11 B and C↗). In our initial experiments, TF consisted of a 14-h fast (food availability from ZT0 to 10, as above), but we found that decreasing the duration of the fast to 10 h did not change the outcome of the experiment. As with the 14-h fast, the 10-h fast actually decreased survival in the 21-d protocol (SI Appendix, Fig. S11B↗).

To determine whether the TF paradigm affected the health of the flies in obvious ways, especially after 21 d, we first measured the weight of female flies after 4 and 21 d of each of the three TF paradigms used. While the weight of flies maintained on TF was a little lower than that of ad lib fed flies, the difference was the same at 4 and 21 d, indicating that changes in weight did not account for the different responses at these two time points (SI Appendix, Fig. S12 A and B↗). We also measured locomotor activity of flies subjected to the different feeding conditions, using multibeam Drosophila Activity Monitors (DAM) that allow high-resolution monitoring of activity. The activity of females, but not males, was higher in some TF flies than those fed ad lib, but it was not always with the same TF paradigm nor was it associated with the duration of the fast. Also, the data for this parameter were similar with 4- and 10 d of TF (SI Appendix, Fig. S12 C and D↗). Thus, decreased survival with TF is not due to obvious impairments in health/function and likely results from changes in endogenous metabolism.

TF also decreased survival upon bacterial infection or heat shock in both sexes. In these experiments, we tested effects of 10-h, 12-h, or 14-h fasts, in each case initiating feeding at ZT0. However, in general, fasting was deleterious to the survival of flies in response to infection or heat shock (Fig. 5 A–L). We infer that while TF promotes metabolic health, it may not be optimal for responses to stress.

As the fitness paradigm we used above did not show benefits of TF, we could not use it to determine if the microbiome mediates beneficial effects of TF. Instead, we considered other ways to assess the functional impact of the microbiome. Although, the microbiome is well studied in terms of its impact on metabolism and immunity (1–3), little is known about its physiological relevance to the circadian system, which is also critical for health. Flies with a microbiome show reduced amplitude circadian cycling, relative to sterile flies, raising the question of whether the dampened cycling is beneficial to the host. To determine if resetting of the gut clock by different environmental cycles is affected by the microbiome, we subjected sterile and microbiome-containing flies to a circadian phase shift paradigm where every 2 d they received a 6-h delay in the light: dark (LD) cycle (Fig. 6A). On the ninth day, they ended up back on the original LD cycle, and during this day we examined circadian expression of the genes per and tim in heads and guts and compared it with that of controls that had remained in the original LD cycle throughout (Fig. 6 B–E).

On day 9, per and tim expression in heads of experimental flies was different from that of unshifted controls, indicating that it had not shifted back to the original cycle (Fig. 6 B and C, SI Appendix, Fig. S13A↗). We noted that, although the presence of the microbiome reduces amplitude of circadian gene expression in the gut, it appeared to increase cycling of per and tim in heads. In guts of microbiome-containing flies subjected to light:dark changes, per and tim expression was similar to that in heads, i.e., it was different from unshifted controls and not resynchronized to the LD cycle on the ninth day (Fig. 6 D and E). Surprisingly, per and tim expression in sterile flies subjected to LD changes showed a profile indistinguishable from that of controls, indicating that these flies adapted very quickly to the LD cycle on Day 9 (Fig. 6 F and G). tim expression in these flies did not shift as quickly, but even so, sterile flies showed a significant shift to match the phase of unshifted controls.

As the phase of the sterile flies at the end of the experiment matched that of controls that remained in the same LD cycle throughout, we sought to confirm that the sterile flies had shifted in response to the light:dark changes earlier in the experiment. We first examined day 3 of the experiment, which represents the first day of a new LD cycle and thus is roughly equivalent to day 9, although on day 3, the shift to the new LD involves an 18-h dark period and the shift on day 9 introduces an 18-h light period. On day 3, per and tim showed transitional phases, and there was no obvious difference between sterile and microbiome flies (SI Appendix, Fig. S13B↗), suggesting that the microbiome influences responses to light and so has less of an effect in extended darkness. We then examined per and tim expression in guts on the second day after the first move to a new LD cycle, i.e., day 4 of the experiment. In this case, per expression had shifted to the new regime in sterile and microbiome-containing flies. tim expression showed a pronounced shift in sterile flies, but not in microbiome-containing flies, where it still showed a transitional phase (Fig. 6 H and I). We surmise that in the gut, the expression of tim RNA shifts slower than per RNA, but both shift faster in sterile flies than in flies that have a microbiome. Thus, the presence of a microbiome stabilizes the gut clock in the event of environmental changes.

The w¹¹⁸ iso³¹, per¹ fly lines present in the study were maintained on standard cornmeal/molasses medium at 25 °C under 12:12 LD conditions unless otherwise specified. For 16S sequencing of gut microbiomes, we first treated flies with 1 mM kanamycin (11815024, ThermoFisher) to remove Wohlbachia, as this would otherwise swamp other bacterial signals in the sequencing. Afterward, the flies were repopulated with a Wolbachia-free microbiome containing Lactobacillus and Acetobacter bacteria from medium previously occupied by other flies. Generation of sterile flies and an account of the TF protocol is provided in the (SI Appendix, SI Methods↗).

5 to 7–d old female and male flies were separately transferred into fresh sterile medium vials every 4 h. At least 35 flies were maintained per vial and allowed to defecate at 25 °C under 12:12 LD conditions coupled with AF or TF treatment method as above. Processing of fecal samples for 16S sequencing is described in the (SI Appendix, SI Methods↗).

For heat shock and infection stress tests, 4-d-old flies were maintained in 12:12 light:dark cycles and fasted for 10-h, 12-h, or 14-h each day and then tested after either 4 d or 21 d. Details of the test can be found in the SI Appendix, SI Methods↗.

For each RNAseq experiment, total RNA was extracted from at least 30 male guts by using the SV Total RNA Isolation kit (Z3105, Promega). Details of RNA isolation, preparation of libraries and RNAseq analysis can be found in the (SI Appendix, SI Methods↗).

Each of the two histone deacetylases inhibitors, sodium 4-phenylbutyrate (PBA) (HY-15654, MedChenExpress) and VA (PHR1061, Sigma), was dissolved in PBS containing 1% blue food coloring and injected subcutaneously into the female fly body. Injection concentrations for PBA and VA were 30 mM and 10 mM, respectively. PBS containing 1% blue food coloring was used as a control. Histone acetylation and target gene expression were investigated 12 h after injection of either inhibitor, by western blot or qPCR. Mortality of the flies was also monitored throughout the experiment.

Details are provided in the SI Appendix, SI Methods↗.

For the circadian phase shift assay, microbiome-containing and sterile female flies were subjected to a 6-h delay in the LD cycle every 2 d, i.e., on the third, fifth, seventh, and ninth days, such that flies returned to the original phase on Day 9. Female fly guts were dissected at ZT4, ZT8, ZT12, ZT16, ZT20, and ZT24 timepoints. Control flies were maintained in the original LD regime (9 am:9 pm) at 25 °C. For each time point, three repeat samples (>30 guts per sample) were collected for microbiome-containing and sterile female flies, respectively. All the samples were subjected to qPCR analysis.

Statistical details of experiments can be found in the figure legends. Circadian statistical analysis was performed in R using JTK_CYCLEv3.1. One-way ANOVA with Holm–Sidak post hoc correction tests were performed with GraphPad Prism 9, experimental groups were compared by unpaired t test with Welch’s correction. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Error bars represent ± SEM, unless otherwise stated in the figure legends.