Kefir ameliorates specific microbiota-gut-brain axis impairments in a mouse model relevant to autism spectrum disorder

Autism spectrum disorder (ASD) is one of the most severe developmental disorders, affecting on average 1 in 150 children worldwide. There is a great need for more effective strategies to improve quality of life in ASD subjects. The gut microbiome has emerged as a potential therapeutic target in ASD. A novel modulator of the gut microbiome, the traditionally fermented milk drink kefir, has recently been shown to modulate the microbiota and decrease repetitive behaviour, one of the hallmarks of ASD, in mice. As such, we hypothesized that kefir could ameliorate behavioural deficits in a mouse model relevant to ASD; the BTBR T+ Itpr3tf/J mouse strain. To this end, adult mice were administered either kefir (UK4) or a milk control for three weeks as treatment lead-in, after which they were assessed for their behavioural phenotype using a battery of tests. In addition, we assessed systemic immunity by flow cytometry and the gut microbiome using shotgun metagenomic sequencing. We found that indeed kefir decreased repetitive behaviour in this mouse model. Furthermore, kefir prolonged stress-induced increases in corticosterone 60 minutes post-stress, which was accompanied by an ameliorated innate immune response as measured by LY6Chi monocyte levels. In addition, kefir increased the levels of anti-inflammatory Treg cells in mesenteric lymph nodes (MLNs). Kefir also increased the relative abundance of Lachnospiraceae bacterium A2, which correlated with reduced repetitive behaviour and increased Treg cells in MLNs. Functionally, kefir modulated various predicted gut microbial pathways, including the gut-brain module S-Adenosylmethionine (SAM) synthesis, as well as L-valine biosynthesis and pyruvate fermentation to isobutanol, which all correlated with repetitive behaviour. Taken together our data show that kefir modulates peripheral immunoregulation, can ameliorate specific ASD behavioural dysfunctions and modulates selective aspects of the composition and function of the gut microbiome, indicating that kefir supplementation might prove a viable strategy in improving quality of life in ASD subjects.


Introduction
Autism spectrum disorder (ASD) is one of the most severe and pervasive neurodevelopmental conditions, affecting on average 1 in 150 children worldwide (Lyall et al., 2017). ASD is hallmarked by impaired social communication skills and enhanced engagement in repetitive behaviours (Association, 2013). In addition, there is a growing body of evidence showing that ASD is associated with substantial differences in gut microbiota composition, as well as symptoms of gastrointestinal dysfunction, such as altered bowel habits, bloating, abdominal pain and increased intestinal permeability (D'Eufemia et al., 1996;De Angelis et al., 2013;Kang et al., 2013;McElhanon et al., 2014;Tomova et al., 2015;Luna et al., 2017;David, 2021;David et al., 2021;Fouquier et al., 2021). Overall, these findings indicate a disturbed communication between the gastrointestinal microbiota and the brain, the microbiota-gutbrain axis may be playing a role in some of the symptoms of ASD (Rhee et al., 2009;Collins et al., 2012;Cryan and Dinan, 2012;Mayer et al., 2014;Foster et al., 2017;Cryan et al., 2019;Sherwin et al., 2019). As such, the gut microbiome is increasingly being recognised as an important novel player and therapeutic target in ASD (Vuong and Hsiao, 2017). This is additionally supported by a study showing that the transplantation of standardised human gut microbiota to ASD-diagnosed children resulted in improved gastrointestinal functionality and reduced behavioural ASD scoring (Kang et al., 2019a,b).
Preclinical research has proven crucial in finding microbiota-based therapeutics for ASD (Hsiao et al., 2013;Kumar and Sharma, 2016). A preclinical animal model often used to study ASD is the BTBR T + Itpr3 tf / J mouse strain. As is the case for humans, this mouse model shows robust deficits in social interaction, increased repetitive behaviours, altered immunity, and an abnormal gut microbiota accompanied by symptoms of gastrointestinal dysfunction (Moy et al., 2007;Coretti et al., 2017;Golubeva et al., 2017;Sgritta et al., 2019;Sharon et al., 2019;O'Connor et al., 2021). A novel modulator of the microbiota-gut-brain axis is the traditionally fermented milk drink kefir (Bourrie et al., 2016;Aslam et al., 2020). This milk drink is fermented by a community of bacteria and yeasts that stably coexist through metabolic cooperation and uneven partitioning between kefir grain and milk (Blasche et al., 2021). Previous work from our laboratory has demonstrated that kefir can modulate the microbiota and decrease repetitive behaviour in the most commonly used C57BL/6 mouse strain (van de Wouw et al., 2020b). In addition, a probiotic containing kefir-derived bacteria ameliorates lipopolysaccharide-induced changes in immune responses and depression-and anxiety-like behaviours (Murray et al., 2019), while kefirderived peptides have been shown to reduce depression-like behaviours (Chen et al., 2021). As such, we aimed to investigate if kefir supplementation could ameliorate some of the ASD deficits observed in the BTBR T + Itpr3 tf /J mouse strain, such as social interaction, repetitive behaviour and gastrointestinal function. Furthermore, we aimed to interrogate the peripheral immune system as a key mediator of kefirinduced gut-brain communication (Lowry et al., 2016;Langgartner et al., 2019) and investigate if any of these changes correlated with alterations in the neuroactive capacity of the microbiome.

Animals
This study used male BTBR T + Itpr3 tf /J mice (5-6 months of age at the start of the experiment, inhouse bred). Animals were housed as three mice per cage, with one cage containing two animals. As such, treatment groups were divided into Milk control (n = 11), and Kefir UK4 (n = 9). The holding room had a temperature of 21 ± 1 • C and humidity of 55 ± 10% with a 12-hour light/dark cycle (lights on at 7:00 am). Food and drinking water were provided ad libitum throughout the study. Bodyweight was monitored weekly. All experiments were conducted under the project authorization license B100/3774 in accordance with the European Directive 86/609/EEC and the Recommendation 2007/526/ 65/EC and were approved by the Animal Experimentation Ethics Committee of University College Cork. All efforts were made to reduce the number of animals used and to minimise the suffering of the animals.

Experimental timeline and behavioural testing
Animals received daily kefir administration by oral gavage for three weeks prior to the assessment of their behavioural phenotype using various tests, which were formed in order of least stressful to most stressful to reduce the likelihood of prior behavioural tests influencing subsequent ones (Fig. 1). In addition, there was a minimum of 36-hours between tests. Behavioural testing started at 9 a.m. The order of testing was as follows: 1) Open field test, 2) Marble burying test, 3) Assessment of self-grooming, 4) 3-Chamber social interaction test, 5) Elevated plus maze, 6) Tail-suspension test, 7) Stress-induced hyperthermia test, 8) Intestinal motility test, 9) FITC-Dextran intestinal permeability test, 10) Fear conditioning, 11) Forced swim test. At the end of the study, animals were sacrificed by decapitation.

Kefir culturing and administration
Kefir UK4 was chosen based on its ability to decrease repetitive behaviour and modulate adaptive immunity (van de Wouw et al., 2020b). Kefir grains were cultured in Irish whole full fat cow's milk (2% w/v) at 25 • C and milk was renewed every 24 h using a sterile Buchner funnel and sterile Duran bottle, as previously described (Dobson et al., 2011;Walsh et al., 2016). Grains were rinsed with sterile deionised water prior to the renewal of milk. The fermented milk (i.e., kefir) collected after the culturing, or milk control, was administered to the mice within one hour by oral gavage (0.2 mL). Daily kefir administration was performed between 4.00 and 7.00 p.m, which was after the behavioural test if one was performed that day.

Open field test
Mice were assessed for locomotor activity and response to a novel environment in the open field test, which was conducted as previously described . Animals were placed in an open arena (40×32×24 cm, L × W × H) and were allowed to explore the arena for 10 min. Animals were habituated to the room 30 min prior to the test. Testing was performed under dim light (60 lx). The open field test box was cleaned with 70% ethanol in-between animals. Experiments were videotaped using a ceiling camera and were analysed for time spent in the virtual centre zone (defined as 50% away from the edges) and total distance travelled using Ethovision version 13 software (Noldus).

Marble burying test
Mice were tested for repetitive and anxiety-like behaviour with the marble burying test, which was conducted as previously described . Animals were individually placed in a novel Plexiglas cage (35×28×18.5 cm, L × W × H), which was filled with sawdust (5 cm) and had 20 equally spread marbles placed on top (5×4 rows). After mice had spent 20 min in the cage, the number of buried marbles was counted by two researchers and averaged. A buried marble was defined as 2/3 of the marble not being visible anymore. In addition, all experiments were videotaped using a ceiling camera and the time the mice showed burying behaviour was scored. Sawdust was renewed, and marbles cleaned with 70% ethanol in-between animals.

Self-grooming test
Mice were tested for repetitive and anxiety-like behaviour by assessing self-grooming. Animals were individually placed in a 250 mL glass beaker for 30 min. A hard-plastic cover was put on the beaker to prevent the mice from escaping and paper covers were put between the beakers containing test mice to block visual contact. This test was performed under dim light (60 lx). All experiments were videotaped using a camera and scored for grooming time. Beakers and the hard-plastic cover were cleaned after each test mouse with 70% ethanol and left to dry for a few minutes.

Three-chamber social interaction test
The three-chamber sociability test was used to assess social preference and recognition and was conducted as previously described (Desbonnet et al., 2014). The testing apparatus was a three-chambered, rectangular box. The dividing walls between each chamber (20×40×22 cm, L × W × H) had small circular openings (5 cm diameter), allowing for access to all chambers. The two outer chambers contained wire cup-like cages (10 cm bottom diameter, 13 cm height), allowing for auditory, olfactory and visual, but no physical contact. The test consisted of 10-minute three phases: 1) Habituation, 2) Social preference, 3) Social recognition. In the first phase (Habituation), mice were allowed to explore the entire box with both wire cup-like cages left empty to allow for habituation to the novel environment. In the second phase (Social preference), one wire cup-like cage contained a novel, agematched, conspecific, male mouse, whereas the other cage contained an object (rubber duckie). In the third phase (Social recognition), the mouse of the previous trial was left in the wire cup-like cage (Familiar mouse), while the object was replaced with a conspecific mouse (Novel mouse). The test mouse was held in the middle chamber while the conspecific mouse and object were placed in the cup wire-like cages. The location of the conspecific mice and object were systemically altered inbetween test mice. The three-chamber test apparatus and wire cup-like cages were cleaned with 70% ethanol after each test mouse and left to dry for a few minutes. The three-chamber test apparatus was not cleaned with 70% ethanol in between testing phases. To reduce potential anxiogenic factors, all mice were habituated to the testing room 40 min before the test, the floor of the testing arena was covered with sawdust and testing was performed under dim light (60 lx). All experiments were videotaped using a ceiling-mounted camera and were scored blinded for the time interacted with the wire cup-like cages. The discrimination index was calculated as follows: Time spent interacting with object or mouse/Total time spent interacting *100%.

Elevated plus maze
The elevated plus maze test was used to assess anxiety-like behaviour and was conducted as previously described . The elevated plus maze apparatus was elevated 1 m above the ground and consisted of a grey cross-shaped maze with two open arms and two closed arms (50×5 cm with 15 cm walls in the closed arms and 1 cm walls in the open arms). Mice were allowed to explore the maze for 5 min. Mice were habituated to the room 30 min prior to the test. Experiments were conducted in red light (5 lx). The elevated plus maze apparatus was cleaned with 70% ethanol in-between animals. Experiments were videotaped using a ceiling-mounted camera and videos were scored blinded for time spent in the open arms, which was defined as all paws in the open arm.

Tail-suspension test
The tail-suspension test was used to assess depressive-like behaviour and was conducted as previously described . Mice were hung by their tail using adhesive tape (2 cm from the tip of the tail) to a 30 cm-elevated grid bar for 6 min. Experiments were videotaped using a numeric tripod-fixed camera and videos were scored blinded for the time mice spent immobile.

Stress-induced hyperthermia test
The stress-induced hyperthermia test was used to assess stressresponsiveness, which was conducted as previously described . Body temperature was determined at baseline (T1) and 15 min later (T2) by gently inserting a Vaseline-covered thermometer 2.0 cm into the rectum. The temperature was noted to the nearest 0.1 • C after it stabilised (~10 s). Mice were restrained by scruffing during this procedure which was the stressor. Animals were habituated to the testing room 1 h prior to the test. The difference between T1 and T2 reflected stress-induced hyperthermia.

Intestinal motility assay
Gastrointestinal motility was assessed as previously described (Golubeva et al., 2017). Briefly, mice were single-housed at 8.00 a.m. with ad libitum access to food and drinking water. Three hours later, 0.2 mL of non-absorbable 6% carmine red in 0.5% methylcellulose dissolved in sterile phosphate-buffered saline was administered by oral gavage, after which drinking water was removed. The latency for the excretion of the first red-coloured faecal pellet was subsequently timed as a measure of gastrointestinal motility.

Assessment of faecal water content and weight
Mice were single-housed for one hour during which faecal pellets were collected (±9 per animal). Pellets were subsequently weighed, incubated at 50 • C for 24 h and weighed again. The average weight per pellet and percentage of faecal water content was calculated. In addition, on a separate day, animals were single housed prior to the start of the active-phase for 24-hours. Faecal pellets from the entire cage were subsequently collected and animals were placed back into their original housing cages.

FITC-Dextran intestinal permeability test
Gastrointestinal permeability was assessed using the FITC-dextran Fig. 1. Experimental design. After three weeks of treatment lead-in, animals were assessed for their behavioural phenotype. Treatment groups consisted of: Milk control (n = 11) Kefir -UK4 (n = 9). The order of behavioural tests was as follows; Week 3: Open field test (OF), Marble burying test (MB), Grooming test (Gr); Week 4: 3-Chamber social interaction test (3CT), Elevate plus maze (EPM), Tail suspension test (TST); Week 5: Stress-induced hyperthermia test (SIH), Carmine red intestinal motility test (CaR), FITC-Dextran intestinal permeability test (FTC); Week 6: Fear conditioning (FC); Week 7: Forced swim test (FST); Week 8: Postmortem, the immune system was assessed by flow cytometry, ex vivo intestinal permeability was tested on ileal and colonic tissue using ussing chambers, ileum and colonic serotonergic levels were quantified by high-performance liquid chromatography (HPLC), and the gut microbiome was analysed using metagenomic shotgun sequencing. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) intestinal permeability test, which was conducted as previously described (van de Wouw et al., 2018). Mice were fasted overnight and received an oral gavage of FITC-dextran dissolved in sterile PBS (Sigma, FD4) at 9 a.m. The administered dosage was 600 mg/kg body weight, and the approximate volume administered per mouse was 0.21 mL. Two hours after the FITC-dextran was administered, approximately 50 μl of whole blood was taken by tail-tip. For this procedure, the end of the tail was held with two fingers without restraining the mouse. A 2-4 mm long diagonal incision was made at the end of the tail using a single edge razor blade, and blood was collected in an EDTA-containing capillary. Blood was then transferred to a tube, centrifuged for 15 min at 3.500 g at 4 • C, and plasma was collected and stored at − 80 • C for later analysis. Plasma FITC-dextran concentrations were assessed with a multi-mode plate reader (Victor 3, Perkin Elmer) with an excitation of 490 nm and emission of 520 nm.

Fear conditioning
Fear conditioning was used to assess amygdala-dependent learning memory and was conducted as previously described (Izquierdo et al., 2006). The test consisted of 3 days/phases; 1) Training, 2) Assessment of cued memory, 3) Assessment of contextual memory, each of which was carried on successive days with a 24-hour interval. In phase 1 (training), animals were recorded for 3 min (baseline), followed by 6 toneconditioned stimuli (70 dB, 20 s), followed by a foot shock (0.6 mA, 2 s), with a 1-minute interval. In phase 2 (Assessment of cued memory), mice were placed in a novel context (i.e. black-checkered walls with a solid Plexiglas opaque floor, under which paper was placed containing a 400 μl vanilla solution (79.5% water/19.5% ethanol/1% vanilla-extract solution), and after an initial acclimation period of 2 min, mice received 40 presentations of the tone-conditioned stimuli, each lasting 30 s with a 5-second interval. In phase 3 (Assessment of contextual memory), mice were placed in the context of day 1 and recorded for 5 min, without the presentation of any tone-conditioned stimuli. The fear conditioning apparatus was cleaned with 70% ethanol in-between animals, but not between trials.

Forced swim test
The forced swim test was used to assess depressive-like behaviour and was conducted as previously described (Cryan and Mombereau 2004). Mice were individually placed in a transparent glass cylinder (24×21 cm diameter) containing 15-cm-depth water (23-25 • C) for 6 min. Mice were gently dried after the test and water was renewed after each animal. Experiments were videotaped using a ceiling camera and videos were scored blinded for immobility time in the last 4 min of the test.

Repeated plasma sampling for corticosterone quantification
Plasma from each animal was sampled by tail-tip five minutes before the forced swim test, and repeatedly after the test in 30-min intervals up to 120 min. For the tail-tip, the end of the tail was gently held with two fingers without restraining the mouse. Using a single edge razor blade, a 2-4 mm long diagonal incision was made at the end of the tail.
Approximately 40 μl of whole blood was taken per time point using an EDTA-containing capillary (Fisher Scientific, 749311), deposited in an Eppendorf and centrifuged for 10 min at 3500 g at 4 • C. Plasma was collected and stored at − 80 • C for later corticosterone quantification.

Tissue collection
The collection of faecal samples throughout the study was done by single housing mice until 3 pellets were dropped between 10.00 and 12.00 a.m. The order of faecal pellet collection was counterbalanced between groups to minimise the effect of the circadian rhythm. Pellets were snap-frozen on dry ice within 3 min after excretion and subsequently stored at − 80 • C. Faecal pellets from week 8 were used for monoamine quantification using high-performance liquid chromatography (HPLC). The caecum contents collected when animals were sacrificed were used for the gut microbiome analysis.
Animals were sacrificed by decapitation in a random fashion regarding test groups between 9.00 a.m. and 2.00 p.m. Trunk blood was collected in EDTA-containing tubes and 100 μl was put in a separate Eppendorf for flow cytometry. Both tubes were centrifuged for 10 min at 3500 g at 4 • C, after which plasma was collected and stored at − 80 • C for later analysis. The remaining cell pellet of the Eppendorf containing 100 μl blood was stored at 4 • C and subsequently used for flow cytometry.
Mesenteric lymph nodes (MLNs) were dissected, cleaned from fat tissue and stored in RPMI-1640 medium with L-glutamine and sodium bicarbonate (R8758, Sigma), supplemented with 10% FBS (F7524l, Sigma) and 1% Pen/strep (P4333, Sigma) at 4 • C for subsequent flow cytometry. The caecum was weighed, snap-frozen on dry ice and stored at − 80 • C. The length of the colon was measured, and the proximal and distal 2 cm were collected and cut in half. One side was snap-frozen on dry ice and stored at − 80 • C and the other treated with RNAlater (Sigma, R0901). This was done by incubating the tissues for 48 h at 4 • C, after which the RNAlater was removed and tissues were stored at − 80 • C for gene expression analysis. Whole brains were snap-frozen in ice-cold isopentane and stored at − 80 • C. For mucus staining, 1.5 cm of the distal colon was collected containing a faecal pellet to ensure the integrity of the mucus layer. The tissue was subsequently incubated in Carnoy's solution (100% ethanol, chloroform and glacial acetic acid in a ratio of 6:3:1) for 2 h at 4 • C and subsequently stored in 100% ethanol at 4 • C for paraffin embedding.

Intestinal permeability by Ussing chamber
This assay was used to assess ileal and colonic permeability ex vivo and was performed as previously described with minor modifications (Golubeva et al., 2017). A piece of distal ileum (1.5 cm segment taken 2.0 cm proximally from the caecum) and proximal colon (1.0 cm segment) were emptied of their contents, cleaned and put in Krebs buffer (1.2 mM NaH 2 PO 4 , 117 mM NaCl, 4.8 mM KCl, 1.2 mM MgCl 2 , 25 mM NaHCO 3 , 11 mM CaCl 2 and 10 mM glucose). The samples were mounted within 5 min in Ussing chambers with an exposed tissue area of 0.12 cm 2 . No seromuscular stripping was performed. The permeability of the epithelial layer was measured via paracellular flux of 4 kDa FITCdextran (Sigma-Aldrich, FD4). FITC-dextran was added to the mucosal chamber at a final concentration of 2.5 mg/mL. To assess serosal-tomucosal FITC flux across the epithelium, 200 μl was collected from the serosal chamber at baseline and after 60, 90 and 120 min. FITC absorbance was measured at 485 nm excitation/535 nm emission wavelengths through fluorometric analysis using a multi-mode plate reader (Victor 3, Perkin Elmer). FITC flux was then calculated as an increment in fluorescence intensity vs baseline fluorescence in the serosal compartment and presented in ng/mL.

Plasma corticosterone and cytokine assessment
Corticosterone quantification of plasma samples (20 μl) obtained in the forced swim test was performed using a corticosterone ELISA (Enzo Life Sciences, ADI-901-097) according to the manufacturer's guidelines. Plasma adrenaline and noradrenaline were quantified using an ELISA (Abnova, KA1877) according to the manufacturer's instructions with one minor modification, where 15 μl plasma was used instead of 300 μl.
Absorbance was read using a Biotek Synergy H1 plate reader equipped with Gen5 software (Biotek, Winooski, VT, USA). Cytokine levels from plasma samples collected during euthanasia were quantified using the V-PLEX Proinflammatory Panel 1 Mouse Kit (MSD, K15048D). Cytokine quantification was done according to the manufacturer's guidelines with one modification, where 20 μl plasma sample was added onto the plate and incubated overnight (15 h) at 4 • C, after which the rest of the protocol was carried out as suggested by the guidelines. Values under the fit curve range and detection range were excluded.
Pictures were taken using full-bright microscopy (Olympus BX51 brightfield Microscope) and analysed using. The mucus thickness (μm) was subsequently measured using ImageJ2 software. As mucus thickness measurements of the same animal were quite variable between different pictures, we chose to take 81 measurements per animal. Specifically, 3 pictures were made of each of 9 different sections, after which 3 measurements were taken of each picture. These 81 measurements were non-normally distributed, so the median was taken of all measurements, which represented the colonic mucus thickness of that animal.

High-performance liquid chromatography
5-hydroxytryptamine (5-HT) and 5-hydroxyindoleacetic acid (5-HIAA) concentrations were determined using HPLC based on the methodology previously described (Clarke et al., 2013). Briefly, mobile phase consisted of HPLC-grade 0.1 M citric acid, 0.1 M sodium dihydrogen phosphate monohydrate, 0.01 mM EDTA disodium salt (Alkem/ Reagecon), 5.6 mM octane-1-sulphonic acid (Sigma Aldrich), and 9% (v/v) methanol (Alkem/Reagecon). The pH of the mobile phase was adjusted to 2.8 using 4 N sodium hydroxide (Alkem/Reagecon). Homogenization buffer consisted of mobile phase with the addition of 20 ng/20 µl of the internal standard, N-methyl 5-HT (Sigma Aldrich). Briefly, tissue samples were sonicated (Sonopuls HD 2070) for 4 s in 500 µl cold homogenization buffer during which they were kept chilled. Tissue homogenates were then centrifuged at 14,000g for 20 min at 4 • C. The supernatant was collected and the pellet was discarded. The supernatant was then briefly vortexed and 30 µl of supernatant was spiked into 270 µl of mobile phase. 20 µl of the 1:10 dilution was injected into the HPLC system (Shimadzu, Japan) which was comprised of a SCL 10-Avp system controller, LC-10AS pump, SIL-10A autoinjector, CTO-10A oven, LECD 6A electrochemical detector, and Class VP-5 software. The chromatographic conditions were flow rate of 0.9 mL/min using a Kinetex 2.6u C18 100A × 4.6 mm column (Phenomenex), oven temperature of 30 • C, and detector settings of +0.8 V. 5-HT and 5-HIAA external standards (Sigma Aldrich, H7752 and H8876, respectively) were injected at regular intervals during sample analyses. Monoamines in unknown samples were determined by their retention times compared to external standards. Peak heights of the analyte: internal standard ratio were measured and compared with external standards, results were expressed as µg of neurotransmitter per gram of tissue.

Gut microbiome analysis
Metagenomic DNA was extracted from cecal contents using the QiaAmp PowerFecal Pro DNA Kit as per the manufacturer's instructions. Whole-metagenome shotgun libraries were prepared using the Nextera XT kit in accordance with the Nextera XT DNA Library Preparation Guide from Illumina and sequenced on the Illumina NextSeq 500 with a NextSeq 500/550 High Output Reagent Kit v2.5 (300 cycles). All sequencing was performed at the Teagasc Moorepark sequencing facility in accordance with standard Illumina protocols.
Mouse caecal metagenomic data (Milk control and kefir-UK4 groups) from a previous study (van de Wouw et al., 2020b), using this kefir were processed using the same method for adapter removal, quality trimming, host read removal, and taxonomic and functional profiling as used in the current study. Differentially abundant features from the previous study were identified by LEfSe and both datasets were processed together using Maaslin2 to identify patterns across both, while controlling for study (i.e., this study versus the previous one).

Statistical analysis
All behavioural and physiological data were assessed for normality using the Shapiro-Wilk test and Levene's test for equality of variances. The effect of kefir was determined by an unpaired Student's t-test when data were normally distributed and Mann-Whitney U test when data were non-normally distributed. Body weight and fear conditioning data were assessed using repeated-measures ANOVA. Statistical significance for social preference and recognition in the 3-chamber sociability test was assessed using Wilcoxon signed ranks test. Parametric data are depicted as bar graphs with points as individual data points and expressed as mean ± SEM. Non-parametric data is depicted as a box with whiskers plot. Statistical analysis was performed using SPSS software version 24 (IBM Corp). A p-value <0.05 was deemed significant.

Kefir is well-tolerated in BTBR mice
No differences were found in body weight, locomotor activity or the weight of various adipose deposits between test and control mice throughout the study, indicating that kefir was well-tolerated in BTBR mice (sFig. 1).

Kefir reduces repetitive behaviour in the marble burying test
Previous work from our lab has shown that BTBR mice have an enhanced engagement in repetitive behaviours in the marble burying test compared to C57BL/6 mice (Golubeva et al., 2017). Mice receiving kefir showed a trend towards decreased marbles buried (t(17) = 2.104, p = 0.051) ( Fig. 2A). This was confirmed by measuring the time mice spent engaging in burying behaviour throughout this test (U = 18, p = 0.033) (Fig. 2B).

Kefir does not affect other measures of anxiety-and depressive-like behaviour
No differences were found in any of the other tests addressing anxiety-like behaviour, which included the assessment of self-grooming behaviour (sFig. 2A), the open field test (sFig. 2B), the elevated plus maze (sFig. 2C) and the stress-induced hyperthermia test (sFig. 2D). In addition, no changes were observed in any of the tests assessing depressive-like behaviour, like the tail suspension test (sFig. 2E) and forced swim test (sFig. 2F).

Kefir does not affect fear-dependent learning and memory
Kefir administration does not influence fear-dependent learning, nor cued and contextual memory in the fear conditioning test (sFig. 3A-D).

Kefir does not affect intestinal permeability
Considering previous work has shown that BTBR mice have decreased MUC2 mRNA expression in the colon, which is the primary constituent of the mucus layer, we wondered if kefir could influence the Fig. 2. Kefir decreased repetitive behaviour in the marble burying test. Animals were assessed for repetitive behaviour using the marble burying test. Both the number of marbles buried was quantified (A), as well as the time animals performed burying behaviour (B). The amount of marbles buried was normally distributed and analysed using an unpaired Student's t-test. The burying time was non-normally distributed and analysed using the Mann-Whitney U test. Significant differences are depicted as: *p < 0.05. All data are expressed as mean ± SEM or box-plot (n = 8-11). Dots on each graph represent individual animals. Fig. 3. All groups show social preference, while only mice receiving kefir show significant differences in social recognition. Social preference (A, B) and social recognition (C, D) were assessed using the 3-chamber social interaction test. All data were non-normally distributed and analysed using the Wilcoxon signed ranks test, except for social preference data for control mice, which were analysed using a paired Student's t-test. Significant differences are depicted as: *p < 0.05 and **p < 0.01. Data are expressed as either mean ± SEM (A, B) (n = 9-11). Dots on each graph represent individual animals. colonic mucus layer (Golubeva et al., 2017). However, kefir administration did not affect colonic mucus thickness (sFig . 4A and B). Similarly, it was previously reported that BTBR mice have increased epithelial permeability (Golubeva et al., 2017), which also remained unaffected by kefir treatment as measured in vivo (sFig. 4C), and ex vivo (sFig. 4D and E).

Kefir does not change gut serotonin levels and gastrointestinal motility
BTBR mice have previously been shown to have decreased 5-HT levels in the colon and ileum, as well as decreased intestinal motility compared to C57BL/6 mice (Golubeva et al., 2017). We hypothesised that kefir administration might ameliorate these deficits, but we did not observe any kefir-induced difference in ileal 5HIAA, 5-HT levels, as well as in the 5HIAA/5-HT ratio (sFig. 5A-C). In addition, no differences were present in colonic 5-HT levels and the 5HIAA/5-HT ratio (sFig. 5E and F), even though a trend was observed towards decreased 5HIAA levels (U = 21, p = 0.062) (sFig. 5D). Similarly, we did not observe any differences in gastrointestinal motility as determined by carmine red administration (sFig. 5G), average faecal pellet weight (sFig. 5H), faecal water content (sFig. 5I) and colon length (sFig. 5J). Interestingly, kefir UK4 did increase cecum weights (U = 17, p = 0.017) (sFig. 5K). Notably, previous work from our lab has shown that BTBR mice have no altered cecum weight compared to C57BL/6 mice (data not published).

Kefir increases Lachnospiraceae bacterium A2 and decreases Clostridiaceae abundance
It has previously been shown that BTBR mice have a reduced alpha diversity (Golubeva et al., 2017). Kefir administration did not significantly impact alpha or beta diversity of the caecal microbiome (Fig. 6A   Fig. 4. Kefir increases stress-induced corticosterone responses yet ameliorates blood LY6C hi monocyte changes. Plasma corticosterone levels were quantified before, and after forced swim test stress (A). Adrenalin and noradrenalin levels were quantified from plasma obtained when animals were sacrificed. For flow cytometry, cells were selected based on FSC-height and SSC-height. LY6C hi monocytes were subsequently selected bas on CD11b and LY6C receptor expression, after which CCR2 receptor expression was quantified by assessing median fluorescent intensity (MFI). The histogram plot depicts CCR2 staining intensity of the representative samples of each group 120 min post-stress (D). Circulating LY6C hi monocytes (CD11b+, LY6C hi ) were quantified at baseline, 30 min and 120 min post forced swim test stress (E). The relative decrease in LY6C hi monocyte levels were subsequently calculated (F). And CCR2 expression (median fluorescent intensity -MFI) was quantified at baseline, 30 min and 120 min post-stress. All data were normally distributed. Stress effects were investigated using a one-way ANOVA following Dunnett post hoc correction. Kefir effects were analysed using an unpaired Student's t-test. Significant differences are depicted as: ## p < 0.01 and ### p < 0.001 for stress effects; *p < 0.05 and **p < 0.01 for kefir effects. All data are expressed as mean ± SEM (n = 9-11). Dots on each graph represent individual animals. and B). However, kefir supplementation did increase the relative abundance of Lachnospiraceae bacterium A2 (LDA = 4.60, p = 0.048), and reduced the relative abundance of the genus Clostridium and family Clostridiaceae (LDA = 4.25, p = 0.025; LDA = 4.25, p = 0.025, respectively) (Fig. 6C). Lachnospiraceae bacterium A2 was not present in the administered kefir (van de Wouw et al., 2020b), and has not been identified as a component of the kefir microbiota by others.
We investigated differentially abundant pathways associated with the bacterial taxon that was differentially abundant following kefir supplementation. The increased relative abundance of L. bacterium A2 was accompanied by an increase in 28 pathways encoded by this species (Fig. 7D). Furthermore, L. bacterium A2 contained genes encoding for 9 GBMs, one of which was SAM (sTable 1).

Fig. 5.
Kefir increases MLN Treg cell, but not circulating Treg cell levels. Cells were selected based on FSC-height and SSC-height. T helper cells were subsequently selected based on CD4 receptor expressions and Treg cells finally were selected on CD25 receptor expression and FoxP3 transcription factor expression (A). T helper cells (CD4+) were quantified in mesenteric lymph nodes (MLNs) (B). In addition, Treg cell (CD4+, CD25+, FoxP3+) levels were quantified in MLNs (C). The origin (thymus vs periphery) of Treg cells were subsequently assessed by quantifying the expression of the transcription factor Helios (D). The same analyses were done for blood (E-G). All data were normally distributed and analysed using an unpaired Student's t-test. Significant differences are depicted as: **p < 0.01. All data are expressed as mean ± SEM (n = 9-11). Dots on each graph represent individual animals. Fig. 6. Kefir increases Lachnospiraceae bacterium A2 and decreases Clostridiaceae abundance without altering alpha and beta diversity. Alpha diversity investigated using inverse Simpson, phylogenetic diversity, Shannon and Simpson indices (A). Beta diversity was calculated using Bray-Curtis, Chao, Jaccard and Unifrac (B). Differentially abundant taxa were identified using LEfSe (C). Dots on each graph in A & B represent individual animals (n = 9-11). LDA: Linear Discriminant Analysis.

Lachnospiraceae bacterium A2 correlates with repetitive behaviour and MLN Treg cells following kefir administration
Correlations between gut microbial taxa and behaviour and immunity parameters significantly changed by kefir administration revealed 12 significant correlations (Fig. 8A). Importantly, an increased L. bacterium A2 abundance correlated with reduced marble burying time and increased Treg cells in MLNs (r s = -0.60, p = 0.007; r s = 0.49, p = 0.027 respectively) (Fig. 8B), which subsequently correlated with marble burying time and Treg cells, where 34 significant correlations were observed (Fig. 8C). Importantly, 27 of these pathways (highlighted bold in Fig. 8C) were also increased due to the increase in relative L. bacterium A2 abundance (Fig. 7D). Finally, there was a significant correlation between the GBM S-Adenosylmethionine synthesis abundance in L. bacterium A2 with marble burying time (r s = -0.67, p = 0.002).

Key predicted metabolic pathways involved in the effects of kefir UK4 on Treg cells in MLNs and repetitive behaviour
Considering that kefir UK4 reduced repetitive behaviour and increased Treg cells in MLNs in both this study, as well as the previous study (van de Wouw et al., 2020b), we combined both datasets to identify which of the previously identified 27 functional pathways were most likely involved in exerting the effects of kefir. As such, functional pathways were correlated with repetitive behaviour (number of marbles buried) and Treg cells in MLNs using data from both studies (sTable 2). There were 6 functional pathways that associated with Treg cells in MLNs following a correction for multiple comparisons, where 3 were related to glycolysis pathways (PWY66-400: glycolysis VI, r s = − 0.3135, q = 0.027; GLYCOLYSIS: glycolysis I, r s = − 0.2417, q = 0.036; PWY-5484: glycolysis II, r s = − 0.2734, q = 0.036), while the rest were associated with L-histidine degradation (HISDEG-PWY, r s = 0.1319, q = 0.036), L-valine biosynthesis (VALSYN-PWY, r s = 0.2099, q = 0.022), and pyruvate fermentation to isobutanol (PWY-7111, r s = 0.2099, q = 0.022). Notably, both L-valine biosynthesis and pyruvate fermentation to isobutanol pathways were present in L. bacterium A2 and correlated with marble burying, indicating that these might be key pathways through which kefir UK4 exerts its effects on Treg cells in MLNs and repetitive behaviour.

Discussion
In the present study, we demonstrate that kefir supplementation was well-tolerated and improved the repetitive behaviour of BTBR mice, a mouse model of ASD (Golubeva et al., 2017). In addition, kefir prolonged stress-induced corticosterone responses and ameliorated the immune response to an acute stressor. Finally, kefir increased the level of anti-inflammatory Treg cells in MLNs of BTBR mice.
Many of the behavioural and physiological findings are consistent with our previous data in normal C57BL/6 mice (van de Wouw et al., 2020b). For instance, these previous studies showed that kefir UK4 decreased repetitive behaviour and increased Treg cells in MLNs, whereas other measures of anxiety and depressive-like behaviour, gastrointestinal motility and gut serotonin levels remained unaffected in control mice. Some divergence did emerge between the effects of kefir UK4 in BTBR mice versus that shown in C57BL/6 mice, with kefir UK4 failing to increase fear-dependent contextual learning or peripheral Treg cell levels in the former strain. These latter differences might be explained by the way kefir affects these different mouse strains as, for instance, fear-dependent learning is decreased in BTBR mice (Stapley et al., 2013). Overall, this indicates that the physiological and behavioural effects of the fermented milk drink kefir are reproducible.
We show that specific behavioural deficits in BTBR mice, such as an increased engagement in repetitive behaviour (Golubeva et al., 2017), were alleviated. Furthermore, mice receiving kefir showed social recognition, which was absent in mice receiving milk. Nonetheless, no significant difference was found in the social recognition index between the two groups, indicating that more research is warranted to validate if kefir can ameliorate social impairments associated with ASD. Previous studies have also indicated that the gut microbiota provides a potential therapeutic target for ASD (Hsiao et al., 2013;Kumar and Sharma, 2016). For instance, repeated intraperitoneal administration of the gut microbial-derived metabolite butyrate has also been shown to attenuate social deficits and repetitive behaviour in BTBR mice (Kratsman et al., 2016). These observed changes were associated with increased GABAergic signalling in the prefrontal cortex as indicated by increased GABA receptor subunit gamma-1 gene expression. In addition, subsequent administration of a reverse agonist of the GABA A receptor reversed the effects of butyrate on social deficits, indicating that butyrate ameliorates social deficits in BTBR mice by enhancing GABAergic signalling (Kratsman et al., 2016). Similarly, the consumption of a fermented milk product containing probiotic bacteria changed the activity of the prefrontal cortex in healthy women (Tillisch et al., 2013). We have previously shown that this kefir increases the capacity of the gut microbiota to produce GABA (van de Wouw et al., 2020b), indicating that kefir might have increased GABA signalling in this study.
The BTBR mouse model also displays a persistent immune dysregulation, similar to what has been reported in individuals with ASD (Careaga et al., 2015). For instance, elevated T cell activation in circulation (Ashwood et al., 2011a), decreased anti-inflammatory Treg cells , and increased IFNγ expression and Th1/2 ratio in the cortex have been reported (Li et al., 2009). Similarly, BTBR mice show increased T helper cell activation, decreased levels of antiinflammatory Treg cells and changes in transcription factor expression associated with T helper cell subsets in the CNS (Zhang et al., 2013;Bakheet et al., 2017). Our data demonstrated that kefir administration increases the prevalence of Treg cells in MLNs, even though Treg cell and IL-10 levels remained unaffected in peripheral circulation.
Our data also showed that kefir prolonged stress-induced increases in plasma corticosterone 60 min post-stress. Even though acute stress, as well as increases in corticosterone, have previously been associated with innate immune cell trafficking and activation (Böbel et al., 2018;Boehme et al., 2020;Dhabhar et al., 2012;van de Wouw et al., 2020a), kefir ameliorated acute stress-induced LY6C hi monocyte decreases, which was associated with a trend towards a reduced stress-induced expression of the trafficking receptor CCR2 (Prinz and Priller 2010). It is therefore interesting to note that stress-induced changes in glucocorticoids have been implicated in both the activation of the immune system and suppression of ongoing stress responses (Sapolsky et al., 2000;van de Wouw et al., 2019). One might postulate that the initial stress-induced increase in corticosterone facilitates immune cell trafficking, after which corticosterone could contribute to the recovery of stress-induced immune cell trafficking. Alternatively, the increased levels of peripheral LY6C hi monocytes may also reflect an increased mobilization of myeloid cells from the bone marrow or spleen (Swirski et al., 2009;Wohleb et al., 2013). Nonetheless, the finding that kefir ameliorated stress-induced decreases in LY6C hi monocyte and CCR2 Correlations were subsequently performed between functional pathways and behaviour and immune measurements (C), after which significant pathways were marked bold if they were significantly enriched in mice receiving kefir. Dots on panels A and B represent individual animals (n = 9-11) while dots on panel C represent statistically significant correlations. receptor expression may indicate decreased LY6C hi monocyte trafficking into the CNS (Prinz and Priller 2010). Interestingly, ASD is associated with elevated levels of circulating cytokines associated with the innate immune system, such as IL-1β, IL-6, IL-8 and IL-12 (Ashwood et al., 2011b). In addition, increased levels of IL-1β, IL-6, IL-8 and TNFα have been reported in the cortex of individuals with ASD (Li et al., 2009b), and increased levels of CCL2, IL6, TGF-β1 and IGFBP-1 have been demonstrated in the cerebellum, middle frontal gyrus and anterior cingulate gyrus (Vargas et al., 2005). Furthermore, microglia activation and increased microglia density have been reported in the dorsolateral prefrontal cortex of ASD individuals (Morgan et al., 2010). Similarly, macrophages derived from BTBR mice secrete more inflammatory cytokines, which correlates with their ASD-associated behavioural phenotype (Onore et al., 2013). Overall, these data indicate that kefir may modulate peripheral immunoregulation, which might have relevance to neuroimmune activation (Reader et al., 2015;Kim et al., 2018). Future research should address whether these kefir-induced changes in acute stress-induced LY6C hi monocyte decreases have relevance to neuroinflammation. It is therefore interesting to note that kefir has been shown to reduce microglial activation in spontaneously hypertensive rats (de Almeida Silva et al., 2020).
In line with previous work (van de Wouw et al., 2020b), kefir did not impact gut 5-HT levels and gastrointestinal motility, even though kefir did increase cecum weight. Interestingly, germ-free mice, as well as mice receiving the prebiotic GOS/inulin or antibiotics have been shown to have increased caecum sizes (Barrat et al., 2008;Grover and Kashyap, 2014;Kang et al., 2017b), indicating that an altered cecum weight is neither good nor bad per se. Finally, no differences in ileal or colonic permeability were observed. In line with these findings is an absence of change in colonic mucosal layer thickness, which represents an additional barrier against the infiltration of gut microbes and their products into the host circulation (Johansson et al., 2011).
Analysis of the caecal microbiome revealed no changes in alpha and beta diversity following kefir administration, which is in line with a previous study from our lab in C57BL/6j mice (van de Wouw et al., 2020b). Kefir did increase relative L. bacterium A2 abundance and reduced the relative abundance of the Clostridium genus and Clostridiaceae, the latter being a pattern that has been reported previously in mice (Marquina et al., 2002). Interestingly, the relative abundance of the Lachnospiraceae family is reduced in BTBR mice (Golubeva et al., 2017), and in humans with ASD in some (Cao et al., 2021), but not all (Zhang et al., 2021), reports. Notably, although gut microbial changes in ASD patients are well-documented, specific microbial signature for the ASD autistic gut has been elusive (Fattorusso et al., 2019;Andreo-Martínez et al., 2020). L. bacterium A2 and its functional potential also correlated with repetitive behaviour and Treg cells in MLNs and, although correlation does not necessarilly imply causation, it will be intriguing to further investigate a potential role for this taxon in kefirinduced effects on behaviour and immunity in this mouse strain. Importantly, L. bacterium A2 is not a member of the kefir microbiota and was detected in only 1 mouse in the previous study with C57BL/6 mice, with no increase in L. bacterium A2 being observed in that instance (data not shown), despite a decrease in repetitive behaviour and MLN Treg cells being observed in that study (van de Wouw et al., 2020b). Differences in host genetics or baseline microbiome may contribute to differences across the two murine models.
It was notable from a microbiome perspective that kefir increased the relative abundance of the GBM SAM, which was also previously reported in C57BL/6j mice (van de Wouw et al., 2020b). Decreased SAM levels have been reported in individuals with ASD (James et al., 2004), while both increased and decreased methionine levels have been reported (Zheng et al., 2017). It is interesting to note that oral administration of SAM ameliorates repetitive behaviour and gene expression in the prefrontal cortex in a different valproic acid-induced ASD model in male mice Weinstein-Fudim et al., 2019). Further analysis of the predicted function of the gut microbiome and its association with repetitive behaviour and Treg cells in MLNs using this study and a previously published kefir study revealed that L-valine biosynthesis and pyruvate fermentation to isobutanol pathways may mediate these kefir UK4-induced effects. Notably, some other studies have reported elevated levels of valine and pyruvate in individuals with ASD (Legido et al., 2013;Zou et al., 2020).
Finally, given that ASD is a neurodevelopmental disorder (Lyall et al., 2017) and the supplementation of kefir in the present study started in adulthood, it may be that starting kefir administration at an earlier stage of life could lead to an even greater amelioration of some of the other physiological or behavioural deficits present in BTBR mice, which remained unaffected in this study. Furthermore, different kefirs have been demonstrated to impact the gut microbiome, host immunity, behaviour and metabolism in distinct ways (Bourrie et al., 2021;van de Wouw et al., 2020b). As such, the observed effects of the kefir strain in this study might not translate to other kefir beverages or, indeed, beverages may exist that have even more profound impacts.

Conclusions
These data demonstrate that the traditionally fermented milk drink kefir reduces specific behavioural impairments, as well as modulate peripheral immunoregulation, in a mouse model of ASD. These results reinforce the notion of kefir as a potential intervention to positively modulate the microbiota-gut-brain axis and show that kefir supplementation might prove a viable strategy in improving specific symptoms in ASD. Moreover, it further confirms the potential of microbiota targeted nutritional interventions that may support mental health . Finally, studies on the validation of kefir as a dietary intervention to improve ASD symptomatology in a clinical setting are now warranted.