Fragile X syndrome (FXS), a leading inherited cause of intellectual disability and autism, is frequently accompanied by sleep and circadian rhythm disturbances. In this study, we comprehensively characterized these disruptions and evaluated the therapeutic potential of a circadian-based intervention in the fragile X mental retardation 1 (FMR1) knockout (KO) mouse. The Fmr1 KO mice exhibited fragmented sleep, impaired locomotor rhythmicity, and attenuated behavioral responses to light, linked to an abnormal retinal innervation and reduction of light-evoked neuronal activation in the suprachiasmatic nucleus. Behavioral testing revealed significant deficits in social memory and increased repetitive behaviors in the mutants, which correlated with sleep fragmentation. Remarkably, a scheduled feeding paradigm (6-hour feeding/18-hour fasting) significantly enhanced circadian rhythmicity, consolidated sleep, and improved social deficits and repetitive behaviors in the Fmr1 KO mice. This intervention also normalized the elevated levels of some pro-inflammatory cytokines, including IL-12 and IFN-γ, in the mutants’ blood, suggesting that its benefits extend to inflammatory pathways. These findings highlight the interplay between circadian disruption, behavior, and an inflammatory response in FXS, and provide compelling evidence that time-restricted feeding may serve as a promising non-pharmacological approach for improving core symptoms in neurodevelopmental disorders.

Animals and Time-restricted feeding (TRF) paradigm

All experimental procedures were approved by the UCLA Animal Research Committee and conformed to guidelines from the UCLA Division of Laboratory Animal Medicine (DLAM) and the National Institutes of Health (NIH). Wild-type (WT) (JAX ID: 000664) and Fmr1 KO mice (Fmr1^tm4Cgr) on the C57BL/6J background (JAX ID: 003025) were acquired from the Jackson laboratory (Bar Harbor, ME). These mutants have a neomycin resistance cassette replacing exon 5 of the fragile X mental retardation syndrome 1 (Fmr1) gene. The studies were carried out in male adult mice (3 to 5 months old) singly housed in light-tight ventilated cabinets in temperature- and humidity-controlled conditions, with ad libitum access to food and water unless otherwise stated. Cages were equipped with either running wheel or infrared motion sensors to monitor the sleep/wake behavior.

TRF was conducted as previously described (Wang et al 2018; Whittaker et al 2018; 2023). Briefly, WT and Fmr1 KO mice were singly housed and entrained to a 12h:12h light/dark (LD) cycle (300 lx vs 0 lx respectively) for a minimum of 2 weeks, then randomly assigned to one of two feeding conditions: food (standard chow) available ad libitum (ad lib) or available for 6 hr during the middle of the active phase from Zeitgeber Time (ZT) 15 to ZT 21. By definition ZT0 and ZT12 are, respectively, the onsets of lights turning on and off. Scheduled feeding was achieved by manually adding and removing food from the mouse cages, and a careful examination was carried out to ensure no small food fragments were dropped and remained in the cages. Food consumption was measured by weighing food at the beginning and the end of the feeding cycles (ALF: 24 hr vs TRF: 6 hr). The ALF and TRF mice groups were held in these conditions for at least two weeks (14 days) plus 3 additional days to measure immobility-based sleep behavior (Suppl. Fig. 1). 

This study used two different cohorts of mice for a total of 135 mice. The experiments shown in Fig. 1 through 6 were designed to compare the circadian phenotype of the Fmr1 KO mice to the WT and were carried out between 2017-2020 before the Covid pandemic research shutdown. After the restart of research, new cohorts of mice were obtained from Jackson Laboratory, and used to test the effects of scheduled feeding on the mutants’ behavioral and circadian deficits (Fig. 7 through 9). 

Immobility-Based Sleep Behavior

Sleep Behavior was assessed under a 12h:12h LD cycle by video recording in combination with an automated mouse tracking analysis software system (Anymaze, Stoelting Co., Wood Dale, IL) as previously described (Lee et al., 2018; Whittaker et al., 2018). Sleep was defined as the periods when 95% of the animal remained immobile for at least 40 seconds or longer, a threshold shown to correlate with >99% accuracy to EEG-defined sleep (Fisher et al., 2016). For each animal, continuous recordings were performed over 5 days and data from two consecutive days with the fewest artifact-free epochs were averaged and used for further analysis. Data were exported in 1-min bins and analyzed separately for the light and dark phases. Total sleep time was determined by summing the immobility durations in the rest phase (ZT 0-12) or active phase (ZT 12-24). Sleep fragmentation was determined by the number of sleep bouts, which were operationally defined as episodes of continuous immobility with a sleep count greater than 3 per minute, persisting for at least 60 secs.

Locomotor cage activity rhythms

Methods used to characterize the sleep/wake cycles have been previously described (Lee et al., 2018; Wang et al., 2018; Whittaker et al., 2018). Single housed WT and the Fmr1 KO mice were habituated to a 12h:12h LD cycle for two weeks to ensure entrainment to the appointed LD schedule before data collection. For the phenotypical comparisons between WT and Fmr1 KO, cage activity was recorded with running wheel sensors for at least 14 days in LD, then, after the assessment of sleep behavior (3 days), the animals were released for two additional weeks into constant darkness (12h:12h dark-dark, DD) to obtain measures of the endogenous rhythms (free-running activity, Suppl. Fig. 1). To investigate the effects of TRF, cage activity rhythms were monitored using overhead passive infrared (IR) motion sensors for two weeks (Wang et al., 2018; Whittaker et al., 2018).

Activity data were collected via the VitalView data recording system (Mini Mitter, Bend, OR) and ten days of recordings were used to obtain the period (tau), the rhythmic power as well as the waveform presentations. The analysis was conducted using the El Temps (A. Diez-Nogura, Barcelona, Spain) and ClockLab (Actimetrics, Lafayette Instruments, Lafayette, IN) programs as previously described (Lee et al 2018; Wang et al., 2018; Whittaker et al., 2018). The periodogram generated by the El Temps uses c² test with a threshold of 0.001 significance, from these the amplitude of the periodicities is calculated at the circadian harmonic to obtain the power of the rhythmicity. The rhythmic power, or percentage of variation (%V), provides a measure of the strength of the mouse’s periodicity corrected for the activity amount and normalized to the percentage of variance derived from peak significance (P=0.05). The amount of cage activity over a 24hr period was averaged over 10 days and reported here as the arbitrary units (a.u.)/h. Fragmentation and imprecision of the daily onset of sleep/wake cycles were determined using ClockLab. Fragmentation was determined by the number of activity bouts per day, with one bout counted when activity was separated by a gap of 21 min or more (maximum gap: 21 min; threshold 3 counts/min). Imprecision, a measure of cycle-to-cycle variability or onset variability, was determined by calculating the daily variability in the time of activity onset from a best-fit regression line drawn through 10 days of activity in both LD and DD conditions using the ClockLab program (Actimetrics, Wilmette, IL, United States).

Photic Regulation of Circadian Behavior

Cohorts of age-matched WT and Fmr1 KO mice were single housed in cages with running wheels under a 12h:12h LD cycle (300 lx, 4500K) for two weeks and, after stable entrainment, exposed to four behavioral assays to test the photic regulation of their circadian system.

(1) Photic suppression on nocturnal activity (negative light masking): mice were exposed to 1 hr of white light (300 lx, 4500K) at ZT 14 to determine their response to light exposure during the dark period (masking of nocturnal activity). The animals’ activity level, measured as the number of wheel revolutions, during the light pulse was compared to the level at the same hour (ZT 14 to 15) on the day before the exposure. Data are reported as percent of suppression= % [(rev during the 1-hr light exposure) – (rev during the dark baseline)]/(rev during the dark baseline).

 (2) Re-entrainment to a 6hr advance of the LD cycle: the mice underwent a 6-hr phase advance. The activity levels were continuously monitored by the wheel-running system and the activity onset determined with VitalView. The ability of the animals to re-entrain was quantified by the difference between the new ZT 12 (lights-off) and the onset of the running activity for each subsequent recording day. A mouse was considered fully re-entrained when the activity onset aligned to the new lights-off time for the consecutive 5 days.

(3) Skeleton Photoperiod (SPP): the mice were transferred from the 12h:12h LD cycle to an SPP consisting of 1hL:11hD:1L:11hD cycles. Mice were exposed to the 1-hr light (300 lx) at the beginning (ZT 0) and the end (ZT 11) of the original light phase. The mice were kept in this photic condition for at least 2 weeks and activity was recorded and analyzed to determine rhythm power, period (tau) and other basic activity rhythm parameters using ClockLab.

(4) Light-induced phase shift of activity rhythms: Mice were released into DD for 10 days. The circadian time (CT) of their free-running activity rhythms was determined using VitalView and El Temps software. The time of activity onset under DD was defined as CT 12. On day 11, the mice were exposed to light (300 lx, 4500K) at CT 16 for 15 min, and after the light pulse, they were held in DD for an additional 10 days. The best-fit lines of the activity onsets of sleep/wake cycles before and after light exposure were measured and compared.

Behavioral Assays

All these tests were conducted under dim red light (<2 lx) during the dark (active) phase between 2 and 4 h after lights- off (ZT 14–16) to avoid sleep disruptions after the sleep and activity recordings had been completed (Suppl. Fig. 1). Animals were habituated to the testing environment for at least 30 minutes prior to the behavioral assays.

The Stereotypic Behavior was assessed using two tests:

(1) The marble burying test was used to evaluate repetitive digging behavior (Yrigollen et al., 2019). An array of 4 by 6 marbles was placed in the testing arena over a layer of 4-cm deep shavings. The testing mouse was introduced to the arena from the corner and its behavior recorded for 30 mins. After that, the testing mouse was carefully removed from the arena with special precaution to not move the marbles and returned to the home cages. The number of marbles buried more than ⅔ of their area was counted, and these were cleaned with 70% ethanol before the next use. Time spent on digging was manually scored by an observer masked to the experimental groups, whilst the distance travelled was derived from the automatic mouse-tracking system (Anymaze software, Stoelting Co., Wood Dale, IL).

(2) The grooming test was conducted as in our previous work (Wang et al 2020). The mouse was introduced to the testing arena and its behavior recorded by a camcorder for 30 mins. The time spent on self-grooming, defined as the cleaning, licking, or washing of the limbs, tail, and body surface areas, typically from a head to tail direction, but excluding bouts of scratching, was manually scored by an observer masked to the experimental groups. The distance travelled was derived from the automatic mouse-tracking system (Anymaze software).

            Social Behavior was assessed using two tests:

 (1) The three-chamber test was performed as previously described (Wang et al 2020). The testing mice were allowed to freely explore an arena with three chambers where the central chamber remained empty. When being habituated to the three-chamber arena, both the WT and the mutants explored the arena evenly with no preference toward the left or the right chamber (left/right ratio: WT: 0.95 ± 0.14; P = 0.43 by paired t-test. Fmr1 KO: 0.82 ± 0.1; P = 0.085 by paired t-test). The three-chamber test consisted of two parts: the first stage assessed the preference of social approach toward the stranger mouse, and the second stage assessed the ability of social discrimination of the testing mouse. In the first stage (social approach), an up-turned metal-grid pencil cup was placed in the side chambers: one was left empty as the novel object (the object chamber), while a never-met stranger mouse matched by sex, age, and genotype with the testing mouse was placed in the second up-turned cup (the social chamber). Thus, the testing mouse was tested for its preference between the object chamber and the social chamber in this first testing stage. In the second stage (social discrimination), the first stranger mouse and the cup remained the same, while a second stranger mouse matched by sex, age, and genotype with the testing mouse was placed in the second up-turned cup. In other words, the social chamber in the first stage became a familiar chamber in the second stage, and the object chamber of the first stage a novel chamber in the second stage. Thus, the preference between the familiar chamber and the novel chamber of the testing mouse was assessed in this second stage. Time spent in each chamber and the distance travelled were derived from the automatic mouse-tracking system (Anymaze software).

(2) The 5-trial social test was conducted as previously described (Mineur et al., 2006). The testing mouse was first habituated to the arena for 30 mins and then introduced to a never-met stranger mouse for 4 trials. The testing mouse was allowed to explore and interact with the stranger mouse for 2 mins in each trial, then it was introduced to a second stranger mouse during the 5th trial for 2 mins. The resting interval between trials was 5 mins. Active social behaviors such as physical contacts (e.g. crawling over, social grooming), nose-to-nose sniffing, nose-to-anus from the testing mouse to the stranger mice were scored manually. Testing mice as well as stranger mice that showed aggressive behavior were withdrawn from the experiment.

Retinal-SCN Connectivity: Injection, Visualization and Analyses of the Neuroanatomical tracer Cholera Toxin

WT and Fmr1 KO mice (4 months-old) received a bilateral injection of CholeraToxin (b subunit) conjugated to Alexa Fluor™ 555 Conjugate (catalog number: C34776; Invitrogen™, Carlsbad, CA). Prior to the injections, the animals received a drop of a local ophthalmic anesthetic (Proparacaine HCl, 0.5%, Sandoz, Holzkirchen Germany) and an intraperitoneal (i.p.) injection of a non-steroidal anti-inflammatory drug (Carprofen, Zoetis, Parsippany-Troy Hills, NJ). The mice were then anesthetized with isoflurane and a 30G needle (BD PrecisionGlide^TM Needle; Becton Dickinson, Franklin Lakes NJ) was inserted at a 45° angle into the sclera and the vitreous chamber, towards the base of the retina to allow leakage of vitreous humor. The Cholera Toxin (2 μg in 2μl of sterile PBS) was injected into the vitreous chamber using a 32G Hamilton syringe (Hamilton, Reno NV). The needle was left in place for ten seconds before being retracted. Seventy-two hours after the injection, the mice were euthanized with isoflurane (30%–32%) and transcardially perfused with phosphate-buffered saline (PBS, 0.1 M, pH 7.4) containing 4% (w/v) paraformaldehyde (PFA, Sigma). The brains were rapidly dissected out, post-fixed overnight in 4% PFA at 4°C, and cryoprotected in 15% sucrose. Sequential coronal sections (40-50 μm) containing the left and right SCN were mounted, and the cover-slips applied with a drop of Vectashield-containing DAPI (4’,6-diamidino-2-phenylinodole; catalog number: H-1200; Vector Laboratories, Burlingame, CA). Sections were visualized on a Zeiss AxioImager M2 microscope equipped with an AxioCam MRm and the ApoTome imaging system, and images acquired with the Zeiss Zen software and a 10x objective to include both left and right SCN. Two methods of analyses were carried out on the images of 5 consecutive sections per animal containing the middle SCN (Lee et al., 2018). First, the relative intensity of the Cholera Toxin fluorescent processes was quantified in the whole SCN, both left and right separately, by scanning densitometry using the Fiji image processing package of the NIH ImageJ software (https://imagej.net). A single ROI of fixed size (575.99 μm x 399.9 μm, width x height) was used to measure the relative integrated density (mean gray values x area of the ROI) in all the images. The values from the left and right SCN were averaged per section and 5 sections per animal were averaged to obtain one value per animal. Second, the retinal innervation of the SCN is strongest in the ventral aspect, where the retino-hypothalamic fibers reach the nuclei, hence, the distribution of the Cholera Toxin fluorescent signal was also obtained for each left and right ventral SCN separately in the same 4-5 consecutive sections per animal using the Profile Plot Analysis feature of ImageJ (Lee et al., 2018). Briefly, a rectangular box of fixed size (415.38μm x 110.94 μm, width x height, Suppl. Fig. 2) to include the ventral part of the SCN was set for each side, and a column plot profile was generated whereby the x-axis represents the horizontal distance through the SCN (lateral to medial for the left and medial to lateral for the right; Suppl. Fig. 2) and the y-axis represents the average pixel intensity per vertical line within the rectangular box. Subsequent processing of the resulting profiles was performed for left and right SCN images separately. To average the profiles of the 5 sections and obtain a single curve per animal, fifth-order polynomial curves were fit to best estimate the position of the intensity peak on the x-axis and, using this position, the original y-axis values were aligned and averaged arithmetically [1 profile per section (either left or right), 5 sections per animal]. Analyses were performed by two observers masked to the genotype of the animals. Data are shown as the average profile ± SD of 3 animals per genotype.

Photic induction of cFos in the SCN and cFos-positive Cell Counting

A separate cohort of male WT and Fmr1 KO mice (3–4 months-old) was housed in DD conditions and exposed to light (300 lx, 4500K, 15 min) at CT 16. Forty-five mins later, the mice were euthanized with isoflurane (30%–32%) and transcardially perfused with phosphate-buffered saline (PBS, 0.1 M, pH 7.4) containing 4% (w/v) PFA. The brains were rapidly dissected out, post-fixed overnight in 4% PFA at 4°C, and cryoprotected in 15% sucrose until further processing. Sequential coronal sections (40-50 μm), containing the middle SCN, were collected on a cryostat (Leica, Buffalo Grove, IL) and further processed for cFos immunofluorescence as previously described (Wang et al., 2020; 2023; Longcore et al., 2024). Briefly, free-floating coronal sections, paired on the rostral-caudal axis and containing both the left and right SCN, were blocked for 1 h at room temperature (1% Bovine Serum Albumin, 0.3% Triton X-100, 10% normal donkey serum in 1xPBS) and then incubated overnight at 4°C with a rabbit polyclonal antiserum against cFos (1:1000, RRID: AB_2247211; Cell Signaling) followed by a Cy3-conjugated donkey-anti-rabbit secondary antibody (Jackson ImmunoResearch Laboratories, Bar Harbor, ME). Sections were mounted and coverslips applied with Vectashield mounting medium containing DAPI, and visualized on a Zeiss AxioImager M2 microscope (Zeiss, Thornwood NY) equipped with a motorized stage, an AxioCam MRm and the ApoTome imaging system. Z-Stack Images (35 images; 34mm, 1.029mm interval) of both the left and right middle SCN were acquired with a 20X objective using the Zeiss Zen digital imaging software, and two observers masked to the experimental groups performed the cell counting. The boundaries of the SCN were visualized using the DAPI nuclear staining and the cells immuno-positive for cFos counted with the aid of the Zen software tool ‘marker’ in three to five consecutive sections. The numbers obtained from the left and right SCN were averaged to obtain one value per section, and those from three-five sections averaged to obtain one value per animal and are presented as the mean ± standard deviation (SD) of 4 animals per genotype.

Histomorphometrical analyses of the SCN

Photographs of DAPI-stained sections generated from the WT and Fmr1 KO mice as described above were used to estimate the area, the perimeter, height and width of the SCN as previously reported (Li et al., 2015; Lee et al., 2018). For each animal, the four measurements were performed in three consecutive sections containing the middle SCN and acquired with a 10X objective and the Zen software. Measurements (in μm) of both the left and right SCN were obtained with the auxilium of the AxioVision software (Zeiss, Pleasanton, CA, USA). Because the borders of the DAPI-defined SCN are somewhat arbitrary, measurements were performed independently by two observers masked to the genotype of the animals. The area of the SCN in the three sections was summed, whilst the perimeter, height and width were averaged to obtain one value per side. No significant differences were found between the left and right SCN, therefore the values of the left and right SCN were averaged to obtain one value per animal. Data are shown as the mean ± standard deviation (SD) of 6 animals per genotype.

Blood sampling and measurements of plasma immune molecules

Blood was collected (~0.5 mL per animal) in the beginning of the light phase via cheek puncture into microvette tubes coated with EDTA (Sarstedt, Numbrecht, Germany; Suppl. Fig. 1). Tubes were gently inverted a few times and placed immediately on wet ice. Within 1 h following collection, samples were centrifuged at 2500 rpm for 15 min at 4° C and the plasma collected into prelabeled Eppendorf tubes (Fisher Scientific, Hampton, NH), and immediately stored at  -80° C until further processing using the Luminex Multiplex Assay at the UCLA Immune Assessment Core (https://www.uclahealth.org/pathology/services-immunoassays).

Statistical Analyses

Data analyses were performed using SigmaPlot 14.5 (Grafiti LLC, Palo Alto, CA) or Prism 10 (GraphPad Software, La Jolla, CA). The impact of the loss of Fmr1 on the waveforms of sleep/wake cycles was analyzed using repeated measures two-way analysis of variance (ANOVA) with time and genotype factors. While a two-way ANOVA with genotype and treatment as factors was used for the differences in sleep bouts between light phase and dark phase. The Holm-Sidak’s multiple comparisons test was applied to determine significant differences among the groups. The effect of TRF on the activity waveforms was analyzed using a three-way ANOVA with genotype, feeding regimen, and time as factors followed by Holm-Sidak’s multiple comparisons test. The datasets were examined for normality (Shapiro–Wilk test) and equal variance (Brown–Forsythe test); genotypic differences in the behavioral tests were determined by Student t-test or the Mann-Whitney test. Correlations between circadian/sleep parameters and other behaviors were examined by applying the Pearson correlation analysis. Normal distribution of the histomorphological datasets and the relative intensity of the Cholera Toxin fluorescent processes in the whole SCN were assessed using the Shapiro-Wilk test, and since these did not pass the normality test, a two-tailed Mann-Whitney test was employed to identify significant differences between groups. The effect of the loss of FMRP** on the photic induction of cFos cells was assessed by one-way ANOVA followed by Bonferroni’s multiple comparisons test. Values are reported as the mean ± standard error of the mean (SEM) or mean ± standard deviation (SD). Differences were determined significant if P < 0.05.