Figure 1 presents a schematic of the experimental protocol. Twelve cohorts of eight 23-day old male Sprague Dawley rats (Envigo Laboratories, Madison, Wi, United States) were used for this experiment (N = 96). A total of 13 animals did not complete the experimental protocol and were thus eliminated from all analyses, for a total N = 83. No explicit power analysis was used, the sample size was selected to ensure an adequate number of biological replicates for the primary outcome measure: polysomnographic sleep recording (target of N = 10–12/experimental group). The experiment consisted of eight experimental groups in a 2 × 2 × 2 design (control diet vs prebiotic diet, ad libitum sleep vs sleep disruption, no social defeat vs social defeat). Rats were pair-housed until electroencephalographic (EEG) and electromyographic (EMG) implant surgery at 7 weeks of age, after which they were individually housed until the end of the experiment.

After placement of rats in cages on the day of arrival, diet groups (control diet vs prebiotic diet) were randomly assigned to the different cages. After EEG/EMG surgery, rats were assigned to further experimental groups (ad libitum sleep vs sleep disruption and no social defeat vs social defeat) randomly, with an effort to ensure prior cagemates were in different groups. Male Long Evans rats (Envigo Laboratories) were used as aggressors in the social defeat model (see below). All rats were maintained on a 12:12 light:dark cycle at room temperature (23 ± 2°C) with food and water available ad libitum throughout the experiment. All protocols were approved in advance by the Northwestern University Institutional Animal Care and Use Committee. Zeitgeber time (Zt) is defined as the number of hours after the onset of the light period (light onset = Zt0).

Upon arrival to the facility, rats were started on ad libitum control or prebiotic diets as previously described (Thompson et al., 2021). The control diet was Envigo Teklad diet TD.110883 (Envigo Teklad, Madison, WI, United States). The prebiotic diet consisted of the control diet supplemented with galactooligosaccharides [GOS, 21.23 total g/kg (7.00 active g/kg); FrieslandCampina, Zwolle, Netherlands] and polydextrose (PDX, 7.69 total g/kg (7.00 active g/kg); Danisco, Terre Haute, IN, United States). The prebiotic diet was custom-made by Envigo Teklad (TD.110889). The control and prebiotic diets are isocaloric and contain similar macronutrient, vitamin, and mineral levels, as previously described (Thompson et al., 2016, 2021).

Four weeks after arrival, and 14 days prior to baseline sleep (i.e., on day –14), rats were implanted with electroencephalographic/electromyographic (EEG/EMG) sleep recording devices (Pinnacle Technologies, Lawrence, KS, United States). Surgical procedures were performed using a rat stereotaxic apparatus with standard aseptic techniques in a ventilated, specially equipped surgical suite. Anesthesia was induced by isoflurane gas. The EEG/EMG headmount consisted of a plastic 6-pin connector attached to four EEG electrodes and two EMG electrodes. Four stainless steel screws serving as two EEG leads and grounds were screwed into the skull with one lead located 5 mm anterior to bregma and 2 mm lateral to the central suture, another 1 mm anterior to bregma, 2 mm lateral to the central suture, and the other two at 1 mm anterior to lambda and 2.5 mm lateral to each side of the central suture. The exposed ends of two stainless steel Teflon-coated wires serving as EMG leads were then inserted into the nuchal muscles using a pair of forceps. The entire headmount was then sealed by dental acrylic and, at the front and the back of the implant, sutures were used to close the incision. Indirect heat support was provided until recovery from anesthetic by placing a heating pad underneath half of the cage the animals were returned to after surgery. Subcutaneous injection of analgesic meloxicam (2 mg/kg; Norbrook Laboratories, Northern Ireland) was given to the animals at the time of surgery and once more on the following day.

After surgery, rats were moved into cylindrical sleep recording cages (Pinnacle Technologies) within individual acoustically-isolated and Faraday-shielded chambers. Two days before baseline sleep, the headmount was connected to the transmission tether. Cages had corncob bedding as well as food and water available ad libitum. Sleep was recorded for a 24-h baseline, then recordings were begun at the start of sleep restriction protocol (ZT6), in which rats were permitted 4 h of interrupted sleep opportunity per 24 h. At the end of the fifth 20-h sleep disruption session, rats were unplugged from their EEG/EMG tethers for social defeat (see below). Upon return to home cages, sleep recording resumed for 24 h (ZT7-ZT7) until rats were unplugged from their EEG/EMG tethers for evaluation in the object location memory (OLM) test. Data were collected using Pinnacle Acquisition software (Pinnacle Technologies), then scored as non-rapid eye movement sleep (NREM), rapid eye movement sleep (REM), or wake in 10 s epochs using machine learning-assisted sleep scoring program as described previously (Gao et al., 2016).

The initiation of a bout of NREM, REM, or wake was defined by the occurrence of two consecutive epochs of NREM, REM, or wake (respectively). A bout was terminated when two consecutive epochs failed to match the state of that bout. For example, a NREM sleep bout was initiated by two consecutive NREM epochs and was terminated when two consecutive non-NREM epochs occurred. A brief arousal was defined as a single epoch of wake within a sleep bout. The delta power band was defined as 0.5–4 Hz, theta as 4–8 Hz, alpha as 8–11 Hz, sigma as 11–15 Hz, and beta as 15–30 Hz. Relative power was calculated as the raw power (μV²) in a particular band divided by the total power in all bands.

After baseline sleep recordings, half of the rats were tested in the sleep disruption protocol. Sleep disruption was achieved using a commercially available system integrated into the chambers (Pinnacle Technologies), which simulates the gentle handling technique via a rotating metal bar (22 cm in length) attached to a post at the center of the cage. For sleep disruption days, the rotation speed of the bar was set at seven rotations per minute with reversals of rotation direction (i.e., clockwise vs. counterclockwise) set to occur at semi-random intervals of 10 ± 10 s. The bar was programmed to rotate for 20 h per day (ZT6-ZT2) and was stationary for 4 h per day (ZT2-ZT6), for 5 days total. Experimenters visually inspected rats at regular intervals during the sleep disruption windows to ensure that the bar mechanism was functioning properly and that the sleep-disrupted rats were awake. Control animals were placed in identical cages with bars that remained stationary throughout the experiment.

Male Long Evans rats were singly housed in large (44 cm × 24 cm × 21 cm) polycarbonate cages and screened for aggressive behavior before the experiment. Rats that began to injure their opponents by harmful bites during screening were not used for the social defeat procedure. On the day of the acute social defeat stress exposure, half of the experimental rats were introduced into the cage of an aggressor (testing done at ∼ZT6). As soon as the aggressor rat attacked and defeated the intruder rat, the intruder was covered with a 25 cm × 15 cm × 15 cm metal mesh cage while still inside the aggressor cage and left in place for 1 h. Control animals not receiving social defeat were unplugged from the recording tether and placed in a clean cage in a quiet room.

The object location memory (OLM) task is a hippocampal-dependent memory task (Murai et al., 2007) that is sensitive to stress exposure (Cazakoff et al., 2010; Howland and Cazakoff, 2010). All rats underwent testing in the OLM task the day after social defeat, beginning at ZT7. Rats were placed in a dimly lit (∼50 lux) 53 cm × 53 cm × 30 cm arena with no objects and allowed to explore for 5 min. Approximately 30–40 min later, they were returned to the chamber, this time containing two identical cylindrical objects (100 mL pyrex bottles with caps) on the same side of the arena. Rats were allowed to explore the arena for 5 min and after 90 min in their home cage were allowed to explore the arena again, with one object moved. Exploration of the moved object for longer than the non-moved object is considered evidence of successful acquisition of contextual memory (Murai et al., 2007). This is denoted by a “location index” expressed as a percentage of time [100 × (time exploring moved object/total time exploring either object)], with values significantly greater than 50% representing evidence of retained contextual memory (Ennaceur et al., 1997). In this experiment, location index was quantified during the 5 min of the testing session. LimeLight (Actimetrics, Wilmette, IL, United States) behavioral software was used to track the path of locomotor activity of each animal within the open field over time. De-identified video files were scored by two experimenters and average location indices were reported.

Fecal samples were collected at four timepoints: (T1) the day of arrival; (T2) one to 2 days before surgery (4 weeks on diet); (T3) at baseline sleep (6 weeks on diet); and (T4) at the end of the experiment (see Figure 1A). Each collection occurred on days where clean cages were provided, so rats were placed into a clean chamber with fresh bedding and food and monitored closely until at least two fresh fecal pellets from each cage were collected. Only spontaneously voided pellets were collected. Samples were placed into individual 1.5 mL microfuge tubes, and frozen at −80°C until microbiome and metabolome analysis, at which point one sample from each animal (or two from each cage for timepoints T1 and T2) was cut in half. One half was used for fecal microbiome analysis and the other half was used for fecal metabolome analysis. At each collection timepoint, duplicate samples of bedding, water, food, and blank tubes were also collected.

To evaluate the impact of the prebiotic diet on the microbiome and to assess for correlations between our primary outcome of interest, the response to sleep disruption, and changes to the microbiome, 16S rRNA gene sequencing was used on a total of 334 fecal and 63 environmental samples. DNA was extracted from fecal samples and the V4 region of the 16S rRNA gene was amplified using the 515f/806rB primer pair with the barcode on the forward read (Apprill et al., 2015) and sequenced as previously described (Caporaso et al., 2012) using an Illumina MiSeq. Sequence data were processed using Deblur v1.1.0 (Amir et al., 2017), trimming to 150 nucleotides to create sub-operational-taxonomic-units (sOTUs). These were then inserted into the Greengenes 13_8 (Mcdonald et al., 2012) 99% reference tree using SATe-enabled Phylogenetic Placement (SEPP) (Mirarab et al., 2012). SEPP uses a simultaneous alignment and tree estimation strategy (Liu et al., 2009) to identify placements for sequence fragments within an existing phylogeny and alignment. Taxonomy was assigned using an implementation of the Ribosomal Database Project (RDP) classifier (Wang et al., 2007) as implemented in QIIME2 (Caporaso et al., 2010). Microbiome data were generally analyzed using the Qiita (Gonzalez et al., 2018) and Quantitative Insights Into Microbial Ecology 2 (QIIME2, version 2018.4) bioinformatics software packages (Caporaso et al., 2010; Bolyen et al., 2019).

Microbial diversity analysis was performed at a rarefied depth of 9,000 reads, resulting in the removal of 21 fecal samples that did not have 9,000 reads. Beta diversity, which measures microbial similarity and dissimilarity between populations of samples, was assessed using weighted and unweighted UniFrac distance matrices as previously described (Lozupone et al., 2011). These matrices were used to generate principal coordinate analysis (PCoA) plots and to perform permutational multivariate analysis of variance (PERMANOVA) in QIIME2. Alpha diversity, which measures microbial taxonomic richness and evenness within a single sample, was calculated using scikit-bio 0.5.1 as implemented by QIIME2. Relative differential abundance was assessed at the OTU, genus, and species levels using analysis of the composition of microbiomes (ANCOM) (Mandal et al., 2015) as implemented in QIIME2. Count numbers of a taxon of interest were centered log ratio transformed in order to perform correlational analysis with physiological variables (Gloor et al., 2017).

We inferred the microbial gene content from the taxa abundance using the software package Phylogenetic Investigation of Communities by Reconstruction of Unobserved States (PICRUSt2¹; v2.1.4-b) (Langille et al., 2013). This tool allows assessment of functional capacity of a microbiome using 16S rRNA gene sequencing data. To identify differentially abundant functional pathways and enzymes, DESeq2 (version 1.14.1) was performed using the Bioconductor R package in RStudio (version 1.2.1335, RStudio Inc.).

A total of 334 fecal and 63 environmental samples were processed for fecal metabolome analyses. A clean stainless-steel bead (Qiagen Catalog# 69989) and 1.5 mL chilled extraction solvent (50% MeOH) was added to each sample. The samples were then homogenized for five min at 25 Hz using a TissueLyser II system (Qiagen Catalog# 85300) and allowed to incubate for 20 min at −20°C. The fecal homogenates were then centrifuged at 14,000 rpm for 15 min at 4°C. 1.2 mL aliquots were then transferred into Nunc 2.0 mL DeepWell plate (Thermo Catalog# 278743) and frozen at −80 °C prior to lyophilization using a FreeZone 4.5 L Benchtop Freeze Dryer with Centrivap Concentrator (Labconco). Wells were resuspended with 200 μL of resuspension solvent (50% MeOH spiked with 2.0 μM sulfadimethoxine), vortexed for 30 s, and centrifuged at 2,000 rpm for 15 min at 4°C. 150 μL of the supernatant was transferred into a 96-well plate and maintained at 4°C prior to LC-MS analysis. A resuspension solvent QC and a six standard mix QC (50% MeOH spiked with 1.0 μM sulfamethazine, 1.0 μM sulfamethizole, 1.0 μM sulfachloropyridazine, 1.0 μM amitriptyline, and 1.0 μM coumarin 314) was run every 12th sample to assess sample background, carry over, chromatography behavior, peak picking, and plate effects.

Fecal extracts were analyzed using an ultra-high performance liquid chromatography system (Vanquish, Thermo) coupled to a hybrid quadrupole-Orbitrap mass spectrometer (Q-Exactive, Thermo) fitted with a HESI probe. Reverse phase chromatographic separation was achieved using a Kinetex C18 1.7 μm, 100 Å, 50 × 2.1 mm column (Phenomenex) held at 40 °C with a flow rate of 0.5 mL/min. 5.0 μL aliquots were injected per sample/QC. The mobile phase used was (A) 0.1% formic acid in water and (B) 0.1% formic acid in acetonitrile. The elution gradient was: 5.0% B for 1 min, increased to 100% B in the next 8 min, held at 100% B for 2 min, returned to 5.0% B in 0.5 min, equilibrated at 5.0% B for 2 min. Positive electrospray ionization parameters were: sheath gas flow rate of 52 (arb. units), aux gas flow rate of 14 (arb. units), sweep gas flow rate of 3 (arb. units), spray voltage of 3.5 kV, capillary temperature of 270 °C, S-Lens RF level of 50 (arb. units), and aux gas heater temperature of 435 °C. Negative electrospray ionization parameters were: sheath gas flow rate of 52 (arb. units), aux gas flow rate of 14 (arb. units), sweep gas flow rate of 3 (arb. units), spray voltage of 2.5 kV, capillary temperature of 270 °C, S-Lens RF level of 50 (arb. units), and aux gas heater temperature of 435 °C. MS data were acquired using a data dependent acquisition method with a resolution of 35,000 in MS¹ and 17,000 in MS². An MS¹ scan from 100–1,500 m/z was followed by an MS² scan, produced by collision induced disassociation, of the five most abundant ions from the prior MS¹ scan.

The orbitrap files (.raw) were exported to mzXML files using MSConvert (Chambers et al., 2012). Feature detection of the MS¹ data was performed using MZmine2 (Pluskal et al., 2010).

The resultant feature tables contained 12,570 features (fecal). Feature tables were also generated for samples of the control and prebiotic diets, containing 2,379 features. All these features were removed from the fecal feature table, resulting in a table of 10,229 non-dietary fecal metabolites. To annotate features with a metabolome standard initiative (MSI) level 1 level of confidence, mass and retention time were aligned and MS/MS fragmentation pattern was compared between features and 20 purified bile acid reference standards. Annotated features were normalized to an internal standard followed by a row sum (total bile acid ion count) normalization.

All graphs depict the mean ± SEM unless otherwise stated. Initial analysis of sleep data used in this study revealed that while 8/8 variables passed heteroscedasticity testing, only 4/8 passed multiple normality tests. Therefore, we elected to use either non-parametric testing or statistical approaches like mixed effects modeling that have been shown to be fairly robust in the setting of mild-moderate violations of assumptions (Schielzeth et al., 2020; Knief and Forstmeier, 2021). All PCoA plots (Supplementary Figure 1) were generated using the EMPeror visualization tool as implemented in QIIME2 (Vazquez-Baeza et al., 2013). Microbiome data processing and analysis, including microbiome PERMANOVA, were performed in QIIME2 as outlined above. Wilcoxon Rank-Sum tests and linear mixed effect modeling of metabolome data with Benjamini-Hochberg correction for multiple comparisons (Table 1), linear mixed effect modeling of body weight data (Supplementary Figure 3), DESeq2 analysis with Benjamini-Hochberg adjustment of PICRUSt2 data (Supplementary Figure 2), and microbe/metabolite Spearman correlation networking (Figure 7) were performed or generated in RStudio (version 1.2.1335, RStudio Inc., Boston, MA, United States). In order to account for the factorial design of the study and to allow interaction terms while also using a non-parametric test, aligned rank transform ANOVA (Conover and Iman, 1981; Wobbrock et al., 2011) was used in lieu of a regular 2-Way ANOVA to analyze post-sleep disruption sleep (Figure 5). Mixed-effects models with Bonferroni or Benjamini-Hochberg post hoc testing for alpha diversity data (Supplementary Figure 1), baseline sleep data (Figure 3), during sleep-disruption sleep data (Figure 4), as well as Spearman correlations of sleep/microbiome data (Figure 6), along with generation of all other graphs/figures, were performed using GraphPad PRISM (version 9.2.0; GraphPad Inc., San Diego, CA, United States).

We first sought to characterize the effect of the prebiotic diet on the fecal microbiome over time. Rats arrived at postnatal day 23, at which time a baseline fecal sample was collected (timepoint T1). Samples were also collected at the start of the fourth week on diet (T2), the sixth week on diet (T3), and at the end of the experiment, corresponding to the seventh week on diet (T4, see Figure 1), and processed for 16S rRNA gene microbiome analysis. Unweighted and weighted UniFrac revealed significant differences in microbial community structure due to the diet at T2, T3, and T4 (Supplementary Figures 1A,B). This was accompanied by a reduction in alpha diversity in the prebiotic diet-fed rats compared to control diet-fed rats. Both Faith’s phylogenetic diversity (Supplementary Figure 1C) and the total number of observed OTU (Supplementary Figure 1D) were reduced at T2 and T3 in the prebiotic diet group compared to control, but this difference was no longer present at T4. Furthermore, Pielou’s evenness was not affected by diet condition throughout the experiment (Supplementary Figure 1E).

To characterize prebiotic diet-induced changes to the microbiome, we performed ANCOM at each timepoint to determine differentially abundant taxa. At the OTU level, ANCOM detected zero differentially abundant features at T1, 55 differentially abundant OTUs at T2, 32 differentially abundant OTUs at T3, and 30 differentially abundant OTUs at T4 (Figure 2A). Because 16S rRNA gene microbiome analysis is generally more reliable at accurate taxonomic prediction at the genus level than the species level (Gilbert et al., 2018; Knight et al., 2018), we then performed ANCOM on features that had taxonomic assignment at the genus level. We found four genera to be differentially abundant due to diet at one or more timepoints (Figures 2B–E).

Ruminococcus was increased at T2 in the prebiotic diet group, but not at the other timepoints (Figure 2B). The genus Coprobacillus was also increased due to the prebiotic diet, but only at T3 and T4 (Figure 2C). Conversely, an unknown genus within the order Bacteroidales was lower compared to control diet-fed rats at T2, T3, and T4 (Figure 2D). Interestingly, the genus Parabacteroides was greatly increased at all non-baseline timepoints, by a factor of 6.5 at T2, a factor of 7 at T3, and a factor of 5 at T4 (Figure 2E). Due to the particularly marked increase in genus Parabacteroides due to the diet, and the fact that there is a growing body of literature describing the role of a particular species within Parabacteroides (Parabacteroides distasonis) in host physiology (Lathrop et al., 2011; Dziarski et al., 2016; Valles-Colomer et al., 2019; Wang et al., 2019), we then performed ANCOM again at each timepoint at the species level to investigate whether P. distasonis was the driving factor behind the increase in genus Parabacteroides. Indeed, we found that P. distasonis was the dominant species within the genus, and that it was significantly increased at T2, T3, and T4 (Figure 2F). There were 26 OTUs that were assigned to P. distasonis, and the average confidence score of the assignments was 0.9755 (95% CI: 0.9459–1.005).

In order to assess whether the diet changed the function of the microbiome, we performed PICRUSt2 (Langille et al., 2013), which uses taxonomy based on 16S rRNA gene data to predict potential functional gene content of the microbiome based on reference sequences of each taxa. Then, we used DESEq2 to identify differentially abundant genes and pathways due to diet at each timepoint. We found that, using an FDR cutoff of 0.1, there were 1,114 differentially abundant genes (525 increased, 589 decreased) due to diet at T2, 882 differentially abundant genes (387 increased, 495 decreased) at T3, and 250 differentially abundant genes (134 increased, 216 decreased) at T4 (Figure 2G). Altered pathways followed a similar pattern, with 243 differentially abundant pathways (113 increased, 130 decreased) at T2, 160 differentially abundant pathways (73 increased, 87 decreased) at T3, and 22 differentially abundant pathways (11 increased, 11 decreased) at T4 (Figure 2H).

More than half of the differentially abundant pathways at T2 and T3 overlapped (144/257; Figure 2I). However, only 8 pathways were differentially abundant due to prebiotic diet at all three timepoints (Figure 2I and Supplementary Figure 2). The abundance of two pathways involved in lipopolysaccharide (LPS) synthesis were significantly increased with prebiotic diet, and two pathways relating to metabolism of exogenous molecules such as aromatic amines and nitrates were decreased due to prebiotic diet at all timepoints (Supplementary Figure 2). Furthermore, three pathways involved in sugar metabolism were altered at all time points due to the prebiotic diet (Supplementary Figure 2). Finally, abundance of a pathway involved in pyrimidine deoxyribonucleotides de novo biosynthesis was increased due to prebiotic diet at all three timepoints (Supplementary Figure 2). Together, these results indicate that the GOS/PDX prebiotic diet had a strong impact on the structure and function of fecal microbiome throughout the experiment, which was characterized by a 5–7-fold increase in the bacterium P. distasonis.

Baseline sleep was assessed using EEG/EMG recording before the sleep disruption protocol (see Methods, Figure 1). We did not observe an effect of 4 weeks exposure to the prebiotic diet on NREM sleep (Figure 3A), NREM EEG delta power (Figure 3B), REM sleep (Figure 3C), or brief arousals (a measure of sleep fragmentation, Figure 3D). These measures were significantly impacted by time of day, demonstrating the circadian rhythm of these sleep parameters.

We measured sleep during the sleep disruption protocol to validate the efficacy of the motorized sleep disruption unit and to examine whether the prebiotic diet influenced the response to sleep restriction. As expected, we found that the sleep disruption protocol significantly reduced NREM sleep on all days of the sleep disruption protocol (Figure 4A), and nearly completely deprived rats of REM sleep, particularly on the first 3 days of the protocol (Figure 4B). The sleep disruption protocol appeared to become slightly less effective over time, as the amount of NREM and REM sleep obtained during the 20 h of sleep disruption increased over time in both groups (Figure 4). Interestingly, this process occurred more quickly in the prebiotic diet-fed rats: on days 2–4 of the sleep disruption protocol, rats in the prebiotic diet group obtained more NREM sleep during the 20-h sleep disruption period than in the control diet-fed rats (Figure 4A). The prebiotic diet-fed rats also obtained more REM sleep during the protocol than control diet-fed rats on days 4 and 5 (Figure 4B).

To assess the effect of the prebiotic diet on recovery from the 5-day sleep disruption protocol, we examined the first 24 h of recovery sleep, which began after completion of the 1-h acute social defeat stress exposure in the experimental groups or an equivalent amount of time in a clean cage in the control groups. Whereas there was no difference in total sleep due to sleep disruption in the control diet-fed rats, there was a significant increase in total sleep in the prebiotic diet-fed, sleep-disrupted rats compared to prebiotic diet-fed controls (Figure 5A). However, NREM delta power, a well-accepted measure of sleep intensity and sleep homeostatic drive (Meerlo et al., 2001; Kamphuis et al., 2015), was not changed in any group (Figure 5B). Examination of NREM sleep architecture revealed that the sleep deprived, prebiotic diet-fed rats had a trend for more NREM than non-sleep disrupted, prebiotic diet-fed rats (Figure 5C), and that this NREM sleep was more consolidated into longer bouts (Figures 5D,E). REM sleep was significantly increased during recovery in both the control diet-fed and prebiotic diet-fed groups, and this increase was due to an increase in the number of bouts without an increase in median bout length (Figures 5F–H). Thus, the prebiotic diet increased total amount of recovery sleep as compared to non-sleep disrupted animals, and promoted consolidation of recovery NREM sleep after repeated sleep disruption.

We also sought to assess the physiological impact of the sleep disruption protocol. We regularly weighed the rats throughout the experiment and found no overall effect of diet or social defeat stress on body weight (Supplementary Figure 3). Though the sleep disruption protocol did not cause a reduction in body weight, it did reduce the rate of weight gain as indicated by an overall effect of sleep disruption on body weight, as well as a sleep disruption by time interaction (Supplementary Figure 3), consistent with prior studies demonstrating increased energy expenditure in the setting of sleep restriction (Mchill and Wright, 2017). The prebiotic diet did not ameliorate this effect. We investigated whether repeated sleep disruption impacted stress-induced changes in performance in the OLM task. At the end of the sleep disruption protocol, half of the rats were exposed to 1 h of social defeat stress, while the other half were transferred to a quiet room for 1 h (see Methods). Twenty-four hours later, all animals were subjected to an OLM task in which a location index significantly above 50% is considered to indicate recognition of the moved object and retained contextual memory (see Methods). In control diet-fed rats, groups exposed to the social defeat stressor did not achieve significantly greater than 50% location index, although the mean location index for these groups was overall similar to non-stressed groups (Supplementary Figure 4). In contrast, all prebiotic diet-fed groups exhibited learning indices significantly greater than 50% (Supplementary Figure 4).

We next evaluated whether any of these prebiotic diet-induced changes to sleep were associated with the observed changes in the fecal microbiome. We focused on P. distasonis, as this bacterium exhibited the most pronounced increase in relative abundance in response to the prebiotic diet (Figure 2), and which has been shown in prior studies to be beneficial to host physiology (Lathrop et al., 2011; Wang et al., 2019; Thompson et al., 2021). Mean-centered log ratio (clr) transformation was then performed to convert P. distasonis relative abundance data to a suitable form for correlational analysis with physiological data, as relative abundances are susceptible to spurious correlations because they are compositional, such that the relative abundances within a single sample sum to one. Using clr transformation has been shown to at least partially ameliorate this problem (Gloor et al., 2017). Using this approach, we found that the clr transformed abundance of P. distasonis at T4 correlated positively with total sleep during the first 24 h of recovery sleep after sleep disruption in prebiotic diet-fed rats but correlated negatively with recovery sleep in control diet-fed rats (Figure 6A). P. distasonis also positively correlated with REM sleep and the number of REM bouts during recovery sleep (Figures 6F,H) in prebiotic diet-fed rats, both of which were altered by sleep disruption but not by the prebiotic diet (see Figure 5). P. distasonis did not significantly correlate with NREM delta power (Figure 6B), NREM parameters (Figures 6C–E), or median REM bout duration (Figure 6G).

A proposed mechanism by which intestinal bacteria may influence host physiology is by generation of microbially-modified metabolites including those originating from dietary sources and host bile acids (Furusawa et al., 2013; De Vadder et al., 2014; Kuipers et al., 2014; Govindarajan et al., 2016; Stilling et al., 2016; Yanguas-Casas et al., 2017). Interestingly, a recent study of P. distasonis demonstrated that it exerts metabolic benefits in part via secondary bile acid production (Wang et al., 2019). We therefore evaluated the impact of the prebiotic diet on the fecal metabolome, and specifically on the bile acid pool, by performing untargeted LC/MS/MS mass spectrometry on fecal samples taken throughout the experimental protocol. We identified 20 different unconjugated and conjugated primary and secondary bile acids within the final feature tables using purified standards and investigated whether the diet altered levels of these bile acids at T2, T3, or T4 (Table 1). We found that at T2 (4 weeks on diet) there were six bile acids that were significantly altered due to diet after correcting for multiple comparisons. The primary bile acid muricholic acid was decreased, while another primary bile acid chenodeoxycholic acid was increased (Table 1). The conjugated primary bile acid taurochenodeoxycholic acid was also increased. Secondary bile acids deoxycholic acid and lithocholic acid were lower in the prebiotic diet group, and the conjugated secondary bile acid taurodeoxycholic acid was significantly increased in the prebiotic group. A number of these changes are consistent with our prior observations from independent experiments at a separate facility (Thompson et al., 2021). These diet effects were no longer present at T3 or T4 (Table 1). There was an overall effect of sleep disruption on the primary bile acid chenodeoxycholic acid and the secondary bile acid deoxycholic acid, but no other bile acids were affected by sleep disruption, and none of the 20 identified bile acids were impacted by social defeat stress exposure (Table 1).

To investigate whether the microbes we found to be most affected by the prebiotic diet were related to the fecal bile acid pool, we performed pairwise Spearman correlations of clr transformed abundances of P. distasonsis, Ruminococcus, Coprobacillus, and the unknown order within Bacteroidales (see Figure 2) with the normalized abundances of the 20 identified bile acids at T3 (time of baseline sleep). The resultant network (Figure 7) of significant (q < 0.05) correlations revealed a tightly covarying network of primary and secondary bile acids, such that P. distasonis, Ruminococcus, and the unknown order within Bacteroidales all correlated with at least one bile acid, thus integrating the observed microbial changes with the detected alterations in the fecal bile acid pool.