Microbiota regulate social behaviour via stress response neurons in the brain

Microbiota regulate social behaviour via stress response

neurons in the brain

Wei-Li Wu

1,2,3,

✉

Mark D. Adame

Chia-Wei Liou

2,3

Jacob T. Barlow

Tzu-Ting Lai

Gil

Sharon

Catherine E. Schretter

Brittany D. Needham

Madelyn I. Wang

Weiyi Tang

James Ousey

Yuan-Yuan Lin

Tzu-Hsuan Yao

Reem Abdel-Haq

Keith Beadle

Viviana

Gradinaru

Rustem F. Ismagilov

1,4

Sarkis K. Mazmanian

Division of Biology and Biological Engineering, California Institute of Technology, Pasadena, CA,

USA.

Department of Physiology, College of Medicine, National Cheng Kung University, Tainan, Taiwan.

Institute of Basic Medical Sciences, College of Medicine, National Cheng Kung University,

Tainan, Taiwan.

Division of Chemistry and Chemical Engineering, California Institute of Technology, Pasadena,

CA, USA.

Abstract

Social interactions among animals mediate essential behaviours, including mating, nurturing, and

defence

. The gut microbiota contribute to social activity in mice

, but the gut–brain

connections that regulate this complex behaviour and its underlying neural basis are unclear

Here we show that the microbiome modulates neuronal activity in specific brain regions of male

mice to regulate canonical stress responses and social behaviours. Social deviation in germ-free

and antibiotic-treated mice is associated with elevated levels of the stress hormone corticosterone,

The Author(s), under exclusive licence to Springer Nature Limited 2021

✉

wlwu@ncku.edu.tw .

Author contributions

W.-L.W., M.D.A., C.-W.L., J.T.B., T.-T.L., G.S., C.E.S., M.I.W., W.T., J.O., Y.-Y.L., T.-H.Y. and R.A.-H.

performed the experiments and/or analysed data. J.T.B., G.S., B.D.N. and R.F.I. provided consultations regarding microbiome

analysis. K.B. and V.G. provided novel viral vectors. W.-L.W. and S.K.M. designed the research. S.K.M. supervised the research. W.-

L.W. and S.K.M. integrated the data, interpreted the results, and wrote the manuscript. All authors discussed the results and

commented on the manuscript.

Online content

Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information,

acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code

availability are available at

https://doi.org/10.1038/s41586–021-03669-y

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this paper.

Competing interests

W-L.W., M.D.A., B.D.N., and S.K.M. have filed a provisional patent on this work. All other authors declare no

competing interests.

Additional information

Supplementary information

The online version contains supplementary material available at

https://doi.org/10.1038/s41586–

021-03669-y

Correspondence and requests for materials

should be addressed to W.-L.W.

Peer review information

Nature

thanks Ioana Carcea and the other, anonymous, reviewer(s) for their contribution to the peer review

of this work.

Reprints and permissions information

is available at

http://www.nature.com/reprints

HHS Public Access

Author manuscript

Nature

. Author manuscript; available in PMC 2021 August 07.

Published in final edited form as:

Nature

. 2021 July ; 595(7867): 409–414. doi:10.1038/s41586-021-03669-y.

Author Manuscript

which is primarily produced by activation of the hypothalamus–pituitary–adrenal (HPA) axis.

Adrenalectomy, antagonism of glucocorticoid receptors, or pharmacological inhibition of

corticosterone synthesis effectively corrects social deficits following microbiome depletion.

Genetic ablation of glucocorticoid receptors in specific brain regions or chemogenetic inactivation

of neurons in the paraventricular nucleus of the hypothalamus that produce corticotrophin-

releasing hormone (CRH) reverse social impairments in antibiotic-treated mice. Conversely,

specific activation of CRH-expressing neurons in the paraventricular nucleus induces social

deficits in mice with a normal microbiome. Via microbiome profiling and in vivo selection, we

identify a bacterial species,

Enterococcus faecalis

, that promotes social activity and reduces

corticosterone levels in mice following social stress. These studies suggest that specific gut

bacteria can restrain the activation of the HPA axis, and show that the microbiome can affect social

behaviours through discrete neuronal circuits that mediate stress responses in the brain.

Bidirectional communication between the gut and the brain affects health and disease

Various environmental and/or peripheral factors influence gut–brain interactions, including

the intestinal microbiota. Changes in stress responses, anxiety, locomotion, and social

behaviour have shown that the microbiota contribute to brain development and function and

to behaviour

–

. Specific gut bacterial species contribute to each of these behavioural

domains in mice

. The influence of host–microorganism interactions on complex

behaviours may extend beyond preclinical studies, as the human microbiome is altered in

several neuropsychiatric disorders that are associated with changes in sociability

Sensory processing, internal states, and decision-making are crucial for the control of social

behaviour

. An animal perceives visual, olfactory, pheromonal, auditory, and/or tactile cues

from another animal, which may modulate the internal state of the first animal towards a

decision that will guide a specific response. In addition, past experiences, emotions,

motivation, and physiological inputs shape the internal state and influence the final outcome

of a social response

. Activation or inhibition of specific brain regions or circuits can

determine the type of response towards another animal, such as a decision to mate, nurture,

fight, surrender, flee, or investigate

In addition to outcomes determined by the brain, the regulation of social behaviour may be

influenced by the gastrointestinal tract

. Diet choice affects social aggression

. Most, but

not all

, studies show that rodents lacking gut bacteria display decreased social behaviour

compared to animals with a complex microbiome

. Mouse models of autism spectrum

disorder display changes in gut microbiome composition, and probiotic treatment corrects

certain social outcomes

–

. In humans, gastrointestinal symptoms have been observed

in several psychiatric disorders with a social component

, and individuals with these

disorders also have altered microbiomes compared to healthy individuals

. Despite the

emerging appreciation of the effects of gut bacteria on complex behaviour, the neural

circuits that are influenced by the microbiome to modulate social activity remain unknown.

Gut microbiota affect social activity

Consistent with previous studies

, germ-free (GF) mice showed reduced social activity

towards novel stranger mice, regardless of microbial status (Fig. 1a–c, Extended Data Fig.

Wu et al.

Page 2

Nature

. Author manuscript; available in PMC 2021 August 07.

Author Manuscript

1a–c). Notably, non-social activity was indistinguishable between GF and specific-pathogen-

free (SPF) mice, which harbour complex microbiota (Extended Data Fig. 1f, g). Adult SPF

mice treated orally with a broad-spectrum antibiotic cocktail (ABX) to deplete their

microbiota also showed decreased social activity (Fig. 1a, d, Extended Data Fig. 1d). Social

activity was not altered by presentation of a novel ABX mouse (Extended Data Fig. 1e).

Non-social activity was not changed by antibiotic treatment (Extended Data Fig. 1f).

Culture-dependent and culture-independent analysis showed that the microbiota were fully

depleted in GF and ABX mice (Extended Data Fig. 2). Neither administration of antibiotics

directly into the brain nor chronic systemic injection of antibiotics altered social behaviour

(Extended Data Fig. 1h–j), which suggests that antibiotics are not neurotoxic. Depletion of

the microbiome reduced social activity regardless of animal sex, social isolation time, and

age (Extended Data Fig. 1k–m) but did not cause altered behaviour in the three-chamber

social test (Extended Data Fig. 1n–q).

Anxiety-like behaviour was lower in GF mice than in their SPF counterparts, as previously

reported

, although there was no change in this behaviour in ABX mice (Extended Data Fig.

3a–h). Overall locomotor activity was not affected in the GF or ABX groups (Extended Data

Fig. 3a, e, i). GF status did not alter water intake (Extended Data Fig. 3j), and the heightened

olfactory investigatory behaviour seen in GF mice exposed to soiled bedding from SPF mice

is probably due to novelty (Extended Data Fig. 3k). Habituation and dishabituation to an

equally novel social odour were similar in GF and SPF mice (Extended Data Fig. 3l).

Responses to microbial scents may influence olfactory communication

. As volatile odours

from GF rats in oestrus induce similar sexual responses in male rats to odours from control

rats

, moderate effects on olfaction in GF mice are unlikely to explain the stark contrast in

social behaviour compared to SPF mice. Collectively, these data show that the absence of

gut microbiota impairs social activity in mice, and that social deviations are not likely to be a

consequence of olfactory dysfunction, anxiety, or altered locomotion.

Microbiota affect brain region activity

We investigated brain activation patterns in GF and ABX mice. Expression of c-Fos, a

measure of neuronal activation, was increased after a social encounter in brain regions

associated with stress responses, including the paraventricular nucleus of the hypothalamus

(PVN), bed nucleus of stria terminalis (BNST) and hippocampal dentate gyrus (DG) (Fig.

1e–l). We observed a trend towards increased c-Fos staining in the basolateral amygdala

(BLA) of GF and ABX mice (Extended Data Fig. 1r, s). GF mice only showed increased

brain activation in response to social interaction (Extended Data Fig. 1t). Several immediate-

early genes in the hippocampus and hypothalamus of GF mice were elevated following

exposure to stranger animals (Extended Data Fig. 4a–d). The effects of gut bacteria on the

social behaviour studied are not likely to be mediated through vasopressin, whereas an

oxytocin-dependent pathway cannot be completely ruled out

(Extended Data Fig. 5a–e,

Supplementary Information 1). We conclude that depletion of the microbiota activates or de-

represses activity in distinct brain regions in response to a social interaction.

Wu et al.

Page 3

Nature

. Author manuscript; available in PMC 2021 August 07.

Author Manuscript

Gut bacteria regulate corticosterone

Gut bacteria modulate the HPA stress response in mice

and affect corticosterone production

by the HPA axis

. Accordingly, serum corticosterone levels were more robustly

increased after a transient social encounter in GF and ABX mice than in SPF controls (Fig.

1m–o). Corticosterone concentrations trended upward immediately after a social encounter

and lasted at least one hour in ABX mice (Extended Data Fig. 1u). Corticosterone levels

were increased in GF mice, but not ABX mice, exposed to a novel cage (Extended Data Fig.

1v). Changes in corticosterone levels were not subject to circadian rhythms (Extended Data

Fig. 1w, x). Transplantation of gut bacteria from SPF donors into GF mice corrected social

activity and lowered corticosterone (Extended Data Fig. 1y, z), showing that the effects of

microbiota alterations are reversible. The increases in corticosterone levels in GF and ABX

mice following social interaction were not due to differences in PVN neuronal density

(Extended Data Fig. 5f–k, Supplementary Information 1). We therefore investigated the

hypothesis that the HPA axis mediates the observed behavioural outcomes in GF and ABX

mice.

We inhibited corticosterone production using both pharmacological and surgical approaches

(Fig. 2a). In GF mice, injection of the corticosterone synthesis blocker metyrapone (MET)

reduced serum corticosterone levels and significantly increased social activity compared to

injection of the vehicle control, carboxymethylcellulose (CMC) (Fig. 2b, c). The adrenal

gland is a major source of corticosterone, and adrenalectomy (ADX) blocked corticosterone

increases in ABX mice compared to sham controls (ABX-sham versus ABX-ADX) (Fig.

2d). Notably, ABX-ADX mice showed normal social activity (vehicle-sham versus ABX-

ADX; Fig. 2e, CMC

1st

and CMC

2nd

), whereas ABX-sham mice showed reduced social

activity compared to controls (vehicle-sham versus ABX-sham; Fig. 2e, CMC

1st

and

CMC

2nd

). Administration of RU-486 to antagonize glucocorticoid receptors led to similar

social activity in vehicle-sham and ABX-sham mice, whereas MET treatment to inhibit

corticosterone synthesis increased social behaviour in ABX-sham mice compared to vehicle-

sham mice (Fig. 2e). ABX-ADX mice showed increased social activity following treatment

with either RU-486 or MET (Fig. 2e). By contrast, acute injection of corticosterone impaired

social behaviour in all groups (Extended Data Fig. 6a). We observed minimal effects on non-

social activity following surgery or drug treatment (Extended Data Fig. 6b–d).

The vagus nerve is a key pathway for gut–brain communication

. However, social

behaviour and corticosterone levels were not changed by subdiaphragmatic vagotomy (SDV)

in ABX mice (ABX-sham versus ABX-SDV) (Fig. 2f, Extended Data Fig. 6e–g), which

suggests that the vagus nerve does not participate in social impairment resulting from

microbiota depletion.

Next, we deleted the gene encoding the glucocorticoid receptor (

Nr3c1

) in specific brain

regions by stereotaxic injection of AAV-hSyn-Cre-GFP bilaterally into

Nr3c1

f/f

mice (Fig.

2g, Extended Data Fig. 7a, d, g). In ABX-treated

Nr3c1

f/f

mice, knockout of

Nr3c1

in the

DG (

Nr3c1

ΔDG

) or BNST (

Nr3c1

ΔBNST

) resulted in an increase in social activity compared

to ABX-treated wild-type mice injected with AAV-hSyn-Cre-GFP (control) (Fig. 2h, j).

Global blockade of glucocorticoid receptor signalling by RU-486 enhanced social activity

Wu et al.

Page 4

Nature

. Author manuscript; available in PMC 2021 August 07.

Author Manuscript

only in ABX control mice, but not ABX

Nr3c1

ΔDG

Nr3c1

ΔBNST

mice (Fig. 2h, j).

Accordingly, serum corticosterone levels (Fig. 2i, k) and c-Fos expression (Extended Data

Fig. 7c, f, j) in the PVN and BNST were reduced in ABX

Nr3c1

ΔDG

and

Nr3c1

ΔBNST

mice

after social interaction. Increased social behaviour and decreased corticosterone were

observed in

Nr3c1

ΔDG

and

Nr3c1

ΔBNST

mice only after antibiotic treatment, but not in mice

with an intact microbiome (Extended Data Fig. 7k, m, p), with the exception of decreased

social activity in SPF

Nr3c1

ΔBNST

mice (Extended Data Fig. 7n).

Bilateral stereotaxic delivery of AAV-hSyn-Cre-GFP into the hypothalamic region

(including the PVN) of

Nr3c1

f/f

mice (

Nr3c1

ΔHYPO

) decreased social behaviour compared

to non-ablated control mice following ABX treatment, with or without administration of

RU-486 (Fig. 2l). Serum corticosterone (Fig. 2m) and c-Fos expression in the DG (Extended

Data Fig. 7i, j) were significantly increased in ABX-treated

Nr3c1

ΔHYPO

mice, supporting

the notion that the glucocorticoid receptor in the hypothalamus serves as a negative regulator

of the HPA axis. Non-social activity was not affected in

Nr3c1

ΔDG

Nr3c1

ΔHYPO

mice, but

was reduced in ABX-treated

Nr3c1

ΔBNST

mice (Extended Data Fig. 7b, e, h, l, o, r).

Decreased social behaviour and increased corticosterone in

Nr3c1

ΔHYPO

mice were

observed only following antibiotic treatment (Fig. 2l, m, Extended Data Fig. 7q, s). These

results indicate that glucocorticoid receptors have distinct functions in different brain

regions, and that they affect the levels of corticosterone and social behaviours that are

regulated by the microbiota.

CRH neurons modulate social behaviour

Although global changes in stress-related genes were not observed in the DG, Ammon’s

horn, or hypothalamus (Extended Data Fig. 4e–l), the

Crh

gene was selectively upregulated

in the hypothalamus of ABX mice after social interaction (but not following novel cage

exposure) (Fig. 3a). We used a chemogenetic approach with designer receptors exclusively

activated by designer drugs (DREADDs) to deliver a mutated human M4 muscarinic

receptor, hM4Di

, via AAV to inactivate CRH-expressing neurons. Specifically, AAV-

hSyn-DIO-hM4Di-mCherry (hM4Di) was bilaterally injected into the PVN of

Crh-ires-Cre

mice, with AAV-hSyn-DIO-mCherry (mCherry) used as a control (Fig. 3b, Extended Data

Fig. 8a). Treatment with clozapine

-oxide (CNO), a designer drug for DREADDs,

markedly increased social behaviour in ABX mice in an acute fashion (Fig. 3c), and

decreased corticosterone in ABX hM4Di mice, but not in ABX mCherry mice (Fig. 3d).

Inactivation of CRH neurons in the PVN also decreased c-Fos staining after social

interaction (Fig. 3e–g), but did not alter c-Fos expression in the BNST or DG (Extended

Data Fig. 8c–f). We were unable to determine whether c-Fos staining was specific to CRH

neurons. Delivery of AAV-hSyn-DIO-hM4Di-mCherry bilaterally into the BNST, another

region with high expression of CRH, did not alter social behaviour, corticosterone levels, or

brain c-Fos expression (Extended Data Fig. 8g–m). Non-social activity was unchanged

following inactivation of CRH neurons in the PVN or BNST (Extended Data Fig. 8b, i).

Unilateral injection of the retrograde neural tracer cholera toxin B subunit (CTB) into the

PVN of SPF and GF mice (Extended Data Fig. 9a) showed a trend towards a reduction in

labelled neuronal projections from the PVN in the BNST, lateral septum (LS), and medial

Wu et al.

Page 5

Nature

. Author manuscript; available in PMC 2021 August 07.

Author Manuscript

amygdala (MeA) in the ipsilateral hemisphere of SPF mice compared with GF mice

(Extended Data Fig. 9b, c). We next co-injected CTB into the PVN and Fluorogold into the

BNST of ABX

Crh-ires-Cre;Ai14D

reporter mice (Extended Data Fig. 9d). The PVN

receives similar projections from the BNST, LS, and MeA in both ABX and control mice

(Extended Data Fig. 9e, g). Labelling alternative projections showed that the PVN and

BNST are bidirectionally connected to each other and receive neuronal projections from the

LS and MeA (Extended Data Fig. 9f, g). Notably, retrograde labelling from the PVN and

BNST was similar in both ABX and vehicle-treated mice (Extended Data Fig. 9e–g).

Therefore, decreased social behaviour in GF and ABX mice is probably due to changes in

neural activity rather than neural circuitry, with the PVN being the major region, although

we cannot rule out other circuits.

Stress pathways promote social deficits

To test whether activation of CRH neurons in the PVN is sufficient to induce social

impairment under more ‘natural’ conditions (rather than in GF or ABX mice), we bilaterally

injected AAV-hSyn-DIO-hM3Dq-mCherry (hM3Dq) into the PVN of SPF

Crh-ires-Cre

mice, using mice injected with AAV-hSyn-DIO-mCherry (mCherry) as controls (Fig. 3h,

Extended Data Fig. 8n). hM3Dq is a mutated human M3 muscarinic receptor that induces

neuronal firing when activated

. Injection of CNO increased corticosterone in SPF hM3Dq

mice, but not in SPF mCherry mice (Fig. 3i). Chemogenetic activation of CRH neurons in

the PVN following intraperitoneal injection of CNO significantly decreased social behaviour

in SPF hM3Dq mice (Supplementary Video 1), but not in SPF mCherry mice (Fig. 3j) or

SPF hM3Dq mice injected with vehicle (Supplementary Video 2). Activation of CRH

neurons in the PVN also increased non-social activity in SPF hM3Dq mice (Extended Data

Fig. 8o, Supplementary Video 1), which was also observed in mice under stress

conditions

Administration of synthetic corticotropin-releasing factor (CRF) directly into the PVN of

SPF mice reduced social activity towards a novel mouse (Fig. 3k–m, Extended Data Fig.

8p). Notably, CRF levels affected non-social activity: low levels of CRF increased non-

social activity and high levels had no effect (Extended Data Fig. 8q). Injection of

corticosterone or the glucocorticoid receptor agonist dexamethasone into the DG and BNST

of SPF mice also decreased social activity (Fig. 3n, o, p, Extended Data Fig. 8r, s).

Therefore, activation of CRH

neurons in the PVN and of glucocorticoid receptor-

expressing neurons in the DG and BNST induce social alterations, revealing a neural

pathway that regulates social behaviour. We speculate that this circuit may also mediate

social activity in response to non-microbial cues.

Enterococcus faecalis

restores social behaviour

We next sought to identify gut bacterial species that affect social activity in mice. Treatment

of SPF mice with different combinations of antibiotics (ampicillin (A), vancomycin (V),

neomycin (N), and metronidazole (M)) showed that a microorganism(s) that is exclusively

sensitive to neomycin appeared to be responsible for modulating social activity and c-Fos

expression in the PVN (Fig. 4a, b, Extended Data Fig. 10a). After social interaction,

Wu et al.

Page 6

Nature

. Author manuscript; available in PMC 2021 August 07.

Author Manuscript

corticosterone levels were lower in AVM-treated mice than in AVNM-treated mice (Fig. 4c).

Notably, transplantation of microbiota from antibiotic-treated donor mice into untreated GF

recipients transferred the associated social activity phenotypes (Fig. 4d) and decreased

serum corticosterone levels (Fig. 4e), suggesting that specific (neomycin-sensitive) bacterial

species mediate social behaviour.

Standard bacterial 16S rRNA gene sequencing did not profile the microbiome because of

low biomass in faecal samples following antibiotic depletion (Extended Data Fig. 2e–k).

Using a recently developed quantitative microbiome sequencing framework

, we identified

a taxon in the

Enterococcus

genus that was present in AVM-treated and AVM-microbiota

recipient mice, and absent in mice that received faecal samples from AVNM donors (Fig. 4f,

g, Extended Data Fig. 10b–d). The predominant bacterial species from the AVM microbiota

E. faecalis

(

E.f

.) (Fig. 4g, Extended Data Fig. 10d).

Adult SPF mice were treated for three weeks with ABX as above, and then switched to

regular water and colonized with

E. faecalis

(ABX +

E.f

.) or treated with sodium

bicarbonate control (ABX + control) by gavage (Fig. 4h). Matched vehicle mice (no

antibiotics) were gavaged with bicarbonate. Mice were first behaviour tested before gavage

(first trial) to confirm that antibiotics reduced social activity in this paradigm (Extended Data

Fig. 10e). Notably, after three weeks of colonization with

E. faecalis

(Extended Data Fig.

10f), ABX +

E.f

. mice showed an increase in social activity compared to their first trial (Fig.

4i) or ABX + control mice (Extended Data Fig. 10g). Corticosterone levels were reduced in

ABX +

E.f

. mice compared to ABX + control mice, to levels similar to vehicle-treated mice

(Fig. 4j). These data reveal that

E. faecalis

increases social activity in mice with ABX-

depleted microbiota, but they do not exclude the possibility that other bacteria have similar

or synergistic effects. GF mice mono-colonized with

E. faecalis

(GF +

E.f

.) also showed an

increase in social activity and decreased c-Fos expression in the brain compared to GF mice

(Extended Data Fig. 10h–k), with corticosterone levels unaffected (Extended Data Fig. 10l),

potentially as a result of developmental issues in GF mice or the lack of other

microorganisms. We conclude that specific members of the gut microbiota, such as

faecalis

, can affect social behaviour in mice.

Discussion

We have provided evidence that a complex microbiome suppresses an overactive stress

response during encounters with a novel mouse by dampening the HPA-axis-mediated

production of corticosterone; we confirmed this result by restoring sociability upon removal

of the adrenal gland, antagonism of the glucocorticoid receptor, or pharmacological

inhibition of corticosterone synthesis in mice devoid of gut bacteria. Antibiotic depletion of

the microbiota in adult mice, used here as research tool and not suggested clinically,

recapitulated many of the findings with GF mice. It is tempting to speculate that symbiotic

bacteria have evolved properties that promote social behaviours among animals under stress

to disseminate microorganisms within a population

, create social groups among animals to

preserve microbial communities

, and/or influence animal mating to expand host–

microbiome symbiosis across generations

Wu et al.

Page 7

Nature

. Author manuscript; available in PMC 2021 August 07.

Author Manuscript

The gut microbiome is altered in several neuropsychiatric conditions that involve social

deficits, such as autism spectrum disorder

, and findings in rodents and humans have

implicated changes in gut bacteria as a contributing factor to brain morphology, activity,

transcriptional patterns, neurogenesis, expression of neurotransmitters, and many complex

behaviours

. The discovery of a specific neuronal pathway that responds to signals from the

gut may enable interventions to modulate social behaviours through safe, natural, and non-

invasive approaches. Future research will aim to uncover the microbial molecules that are

responsible for modulating social activity and to identify host receptors and cell types that

receive microbial signals and translate them into specific behavioural outputs.

Methods

Mice

Wild-type C57BL/6J (00064),

Nr3c1

f/f

(021021; B6.Cg-

Nr3c1

tm1.1Jda

/J),

Crh-ires-Cre

(012704; B6(Cg)-

Crh

tm(cre)Zjh

/J), and Ai14D (007914; B6.Cg-

t(ROSA)26Sor

tm14(CAG-tdTomato)Hze

/J) mice were obtained through Jackson Laboratory.

C57BL/6J germ-free (GF) mice were bred in the Gnotobiotic Animal Facility at Caltech. In

addition, wild-type C57BL/6JNarl mice used in SDV, brain cannulation, chronic antibiotic

intraperitoneal injection, partial colonization, and anxiety-like behaviour experiments were

obtained through National Laboratory Animal Center, Taiwan. All experiments were

performed with male mice except for Extended Data Fig. 1k. All mice were group housed

(2–5 mice per cage) unless specified with a 13-h light/11-h dark cycle (lights on at 06:00) at

21–23 °C and 45–55% relative humidity within a range of 30–70% in ventilated cages

(Super Mouse 750, Lab Products). Unless specified, SPF mice were fed 5053 PicoLab

Rodent Diet (LadDiet) and the SPF breeders were fed a mix of half 5053 and half 5058

PicoLab Rodent Diet. SPF C57BL/6J mice obtained from Jackson Laboratory were yielded

from multiple litters (at least 6–8 litters) and randomly assigned to subject groups (SPF,

vehicle, or ABX) or novel mice (SPF or ABX) to eliminate the maternal effect. GF mice

used in this study were yielded from multiple litters (at least 6 litters) and randomly assigned

to testing groups to eliminate the maternal effect. No statistical methods were used to

predetermine sample size. The investigators were not blinded to allocation during

experiments and outcome assessment.

For experimental mice, SPF, GF, vehicle and ABX mice were transferred and maintained in

sterilized cages and fed autoclaved 5010 PicoLab Rodent Diet (LadDiet) and provided with

autoclaved drinking water. GF mice were housed outside the GF isolator for no more than a

week. GF mice used in this study were tested for behaviour at 11–15 weeks of age. The SPF

mice used for comparison with GF mice were all age-matched. All GF mice completed the

behaviour test once and were processed with tissue collections 30–60 min after behaviour

testing. Faecal samples from GF mice were freshly collected at the endpoint of each

experiment. The faeces were plated onto Brucella agar plates with 5% sheep’s blood

(B0150; Teknova) and cultured aerobically and anaerobically to screen for contamination in

the sterile housing. All SPF mice were handled with the same procedures as GF mice. All

experiments were performed with approval of the Caltech and NCKU Institutional Animal

Care and Use Committee (IACUC).

Wu et al.

Page 8

Nature

. Author manuscript; available in PMC 2021 August 07.

Author Manuscript

Antibiotic cocktail (ABX) treatment

Gut microbiota were depleted in adult mice (8–12 weeks) by treatment with a cocktail of

antibiotics for 3–4 weeks. The recipe included drinking water with ampicillin (1 g/l),

vancomycin (0.5 g/l), neomycin (1 g/l), and metronidazole (0.5 g/l), and was sweetened with

1% w/v of sucrose and filtered with a 0.22-μm filter. To avoid confounding effects from

chronic stress induced by oral gavage, antibiotics were administered in the drinking water ad

libitum. Untreated (vehicle control) mice received 1% w/v of sucrose in filtered drinking

water. The length of ABX treatment was adapted based on the recovery of body weight after

ABX treatment. All antibiotics were United States Pharmacopeia (USP) grade, or at

minimum cell culture grade. Drinking water was prepared and changed weekly. All vehicle-

treated mice went through the same handling procedures as ABX mice. Faecal samples were

freshly collected at the endpoint of each experiment and plated onto Brucella agar plates

with 5% sheep’s blood (B0150; Teknova) and cultured aerobically and anaerobically to test

whether gut bacteria were successfully depleted following ABX treatment.

Corticosterone measurement

Whole blood was collected by either cardiac puncture or retro-orbital bleeding and placed

into Micro tube Z gel (41.1378.005; Sarstedt). Serum was separated by centrifuge at 10,000

for five minutes and stored at −80 °C until use. Corticosterone levels were detected using the

Corticosterone EIA Kit (K014-H5; Arbor Assays) according to the manufacturer’s protocol.

Owing to the effects of the circadian rhythm on corticosterone levels, we collected blood

samples only between 13:00 and 17:00 on each day of the experiment.

Adrenalectomy

Mice were anaesthetized with 5% isoflurane in an induction box followed by maintenance

on a nose cone with 1–2% isoflurane during surgery. A single dose of sustained-release

buprenorphine (1 mg/kg) was subcutaneously injected before surgery. The surgical fields

were covered with a sterile drape and sterile gloves were worn. A 1.5-cm dorsal midline

incision was made with its midpoint centred over the 13th rib. All of the underlying muscle

on either side of the spinal column was incised. The adrenal glands are located in the fat pad

that covers the cranial portion of the kidney. The adrenal glands were isolated and removed

using a small curved forcep and a micro scissor. Bupivacaine (1 mg/kg) was applied

subcutaneously along the incision just before wound closure. The peritoneal opening was

closed with a 5–0 subcutaneous suture (Vicryl) and 7-mm wound clips (Roboz) were applied

to close the dermis. For postoperative care, mice were monitored for signs of pain or distress

at least three times a day for three days, and were supplied with drinking water containing

30 mg/kg ibuprofen for seven days with 0.9% sodium chloride ad libitum. The clips were

removed two weeks after surgery under isoflurane anaesthesia. Sham-operated control mice

underwent the same procedure as described above without removal of the adrenal glands.

Drug administration

To systemically inhibit corticosterone synthesis, mice were intraperitoneally injected with 50

mg/kg metyrapone (3292; R&D) and returned to their home cage. To block glucocorticoid

receptors, mice were intraperitoneally injected with 40 mg/kg RU-486 (also known as

Wu et al.

Page 9

Nature

. Author manuscript; available in PMC 2021 August 07.

Author Manuscript

mifepristone; M8046; Sigma-Aldrich) and returned to their home cage. Metyrapone and

RU-486 were freshly dissolved in 0.5% carboxymethylcellulose sodium (CMC; C9481;

Sigma) on the day of the reciprocal social interaction test. Injection with 0.5% CMC was

used as a baseline control. Behaviour testing was performed 40 min after injection.

For acute corticosterone exposure, mice were intraperitoneally injected on two consecutive

days with 10 mg/kg corticosterone 2-hydroxypropyl-

-cyclodextrin complex (C174; Sigma-

Aldrich) and returned to their home cage. The corticosterone 2-hydroxypropyl-

cyclodextrin complex was used to facilitate solubility of the steroid. The reciprocal social

interaction test was performed 40 min after the second corticosterone injection.

For DREADD-based chemogenetic activation or inactivation, mice expressing hM3Dq,

hM3Di or mCherry were intraperitoneally injected with 2–3 mg/kg CNO (Enzo Life

Sciences) and returned to their home cage. The reciprocal social interaction test was

performed 32–40 min after the second CNO injection.

Viral vectors

AAV-hSyn-DIO-hM4Di-mCherry (44362-AAV5); AAV-hSyn-DIO-hM3Dq-mCherry

(44361-AAV5); AAV-hSyn-DIO-mCherry (50459-AAV5); AAV-hSyn-EGFP (50465-AAV5)

viruses were purchased through Addgene. AAV-hSyn-Cre-GFP (AAV5) was produced at the

Vector core at the University of North Carolina at Chapel Hill.

Stereotaxic surgery

Adult (7–8 weeks old) mice were deeply anaesthetized with 5% isoflurane in oxygen and

kept at 1–2% isoflurane during surgery. In addition, 5 mg/kg ketoprofen was subcutaneously

given once before surgery. Surgery was performed with a stereotaxic frame (Model 1900;

David Kopf Instruments). The bregma was located by using a centring scope (Model 1915;

David Kopf Instruments) and aligned using a stereotaxic alignment indicator (Model 1905;

David Kopf Instruments). The skull was exposed, and holes were produced using a

stereotaxic drill (Model 1911; David Kopf Instruments) with a no. 79 micro drill bit (Drill

Bit City). Viruses were injected into the brain in both hemispheres using a pulled glass

capillary (1B120F-4; World Precision Instruments) with a nanolitre injector (Nanoliter2010;

World Precision Instruments) at a flow rate of 23 nl/minute controlled by a micro controller

(Micro4; World Precision Instruments). The glass capillaries were left in place for five

minutes to prevent backflow. Stereotaxic injection coordinates (in millimetres) were based

on the Paxinos and Franklin atlas

: PVN (AP: –0.80, ML: ±0.25; DV: −4.75), BNST (AP:

0.26, ML: ±0.90; DV: −4.30), DG (AP: −1.50, ML: ±0.40; DV: −2.25). The diagrams of

brain injection were based on the Paxinos and Franklin atlas

(Extended Data Figs. 7a, d, g,

8a, g, n). Volumes delivered were 207 nl for most regions, and 299 nl for the DG.

Bupivacaine (1 mg/kg) was applied subcutaneously along the incision before wound closure.

The incision on the scalp was closed with tissue adhesive (Gluture topical adhesive; Abbott

Laboratories). For postoperative care, mice were monitored daily for signs of pain or distress

for at least three days and were supplied for seven days with drinking water containing 30

mg/kg ibuprofen ad libitum. Behavioural experiments were performed at least three weeks

Wu et al.

Page 10

Nature

. Author manuscript; available in PMC 2021 August 07.

Author Manuscript

after virus injection. All surgically manipulated animals underwent histological examination

after death to ensure viruses were correctly injected.

Guide canula implantation surgery

Mice were anaesthetized with 5% isoflurane in oxygen in a Plexiglas cage. After

anaesthesia, mice were placed in a digital stereotaxic device (Stoelting) delivering isoflurane

to maintain anaesthesia throughout the surgery and injected subcutaneously with 5 mg/kg

ketoprofen. The custom guide cannula was implanted in the following regions at the

following coordinates based on the Paxinos and Franklin atlas

: lateral ventricle (LV), AP:

−0.1 mm; ML: 1.0 mm; DV: −2.0 mm; PVN, AP: −0.8, ML: ±0.25, DV: −4.75; DG, AP:

−1.5, ML: ±0.4, DV: −2.25; BNST: AP: 0.26, ML: .., DV: −4.3. The diagrams of brain

cannulation were based on the Paxinos and Franklin atlas

(Extended Data Fig. 8p, r, s).

Two to four small stainless-steel screws were installed into the skull to anchor the acrylic

maintaining the guide cannula. Then, a dummy stainless-steel plug was implanted in the

cannula to prevent clogging with blood or cerebrospinal fluid at the cannula opening. For

postoperative care, mice were monitored daily for signs of pain or distress for at least three

days and were supplied for seven days with drinking water containing 30 mg/kg ibuprofen

ad libitum.

The customized cannulization set includes a guide cannula, injector, dummy, and cap

(Extended Data Fig. 8t). The guide cannula was implanted 0.5 mm above the PVN, DG, and

BNST. The tip of the guide cannula track can be visualized in each implanted mouse. The

injector was designed to reach 0.5 mm below the tip of guide cannula. As the injector was

made of 33G fine needle, the needle tracks produced by the injector were not visible in brain

slices. The cannulization set for the PVN, DG, and BNST was customized (RWD Life

Science) with the following specification: PVN (guide cannula: double_OD 0.41 mm-27G/

C.C0.5/B7.8/M3.5/C = 4.25 mm, dummy cannula: double_OD 0.20 mm-27G/C. C0.5/Mates

with M3.5/G = 0.5 mm, Injector: double_OD0.21 mm-33G/C. C0.5/Mates with M3.5/C =

4.25 mm/G = 0.5 mm), DG (guide cannula: double_OD 0.41 mm-27G/C.C0.8/B7.8/M3.5/C

= 1.75 mm, dummy cannula: double_OD 0.20 mm-27G/C.C0.8/Mates with M3.5/G = 0.5

mm, Injector: double_OD0.21 mm-33G/C.C0.8/Mates with M3.5/C = 1.75 mm/G = 0.5

mm), BNST (guide cannula: double_OD 0.41 mm-27G/C.C1.8/ B7.8/M3.5/C = 3.8 mm,

dummy cannula: double_OD 0.20 mm-27G/C. C1.8/Mates with M3.5/G = 0.5 mm, Injector:

double_OD0.21 mm-33G/C. C1.8/Mates with M3.5/C = 3.8 mm/G = 0.5 mm).

Intracerebroventricular (ICV) antibiotics injection

Ampicillin (1 mg/l) and metronidazole (0.5 mg/l) were dissolved in artificial cerebrospinal

fluid (ACSF: 7.46 g/l NaCl, 0.19 g/l KCL, 0.14 g/l CaCl

, 0.19g/l MgCl

, 1.76 g/l NaHCO

0.18 g/l NaH

, 0.61 g/l glucose in ddH

O) and adjusted to pH 7.4. The dissolved

antibiotics were filtered through a 0.22-μm filter. Mice were infused with antibiotics during

social behaviour tests with an infusion rate of 7 nl/s for 3 min.

CRF infusion in PVN

CRF was dissolved in saline (high-dose: 210 μM; low-dose: 42 μM) and injected into each

hemisphere of PVN during anaesthesia with an infusion rate of 4.5 nl/s for a total infusion

Wu et al.

Page 11

Nature

. Author manuscript; available in PMC 2021 August 07.

Author Manuscript

volume of 450 nl. Once complete, the injection syringe remained in tissue for 2 min to

prevent backflow. After 2 min, the injection syringe was placed in the other hemisphere of

PVN and the same volume of CRF was infused at the same flow rate. The mice were then

placed in a novel cage to recover from anaesthesia until social behaviour testing.

Corticosterone and dexamethasone infusion in DG and BNST

Corticosterone (65 μM) or dexamethasone (20 mM) was dissolved in DMSO, diluted in

saline, and injected bilaterally into DG and BNST of anaesthetized mice at an infusion rate

of 4.5 nl/s up to 450 nl. The injection syringe remained in tissue for 2 min to prevent

backflow. After 2 min, the injection syringe was placed in the other hemisphere of PVN and

the same volume of corticosterone or dexamethasone was infused at the same flow rate. The

mice were then placed in a novel cage to recover from anaesthesia until social behaviour

testing.

Subdiaphragmatic vagotomy (SDV)

Mice were habituated to a liquid diet (Research Diets; AIN-76A) for two days and were

fasted overnight before surgery. The abdominal surgery site was shaved and wiped with the

skin disinfectant chlorhexidine three times before incision. Mice were then anaesthetized

using 1–5% isoflurane and injected subcutaneously with 5 mg/kg ketoprofen. An incision

was made along the abdominal midline and the muscle and skin were separated using blunt

scissors. The liver was then gently moved using a sterile cotton swab to expose the stomach

and oesophagus. The ventral and dorsal trunks of the vagus nerve were resected using sharp

forceps. In the sham operation, the stomach and oesophagus were exposed but the vagal

trunks were not resected. The organs were then placed back to their anatomical position. The

incision along the muscular layer was closed with absorbable suture and treated with topical

lidocaine (0.25%) and the skin incision was closed with non-absorbable suture and treated

with

-butyl cyanoacrylate adhesive (3M Vetbond) to prevent infection. After surgery, the

mice were placed into a clean cage placed on a heat pad. During the recovery time, mice

were given a hydrating, nutritious gel pack (DietGel 76A 2oz, Clear H

O) for two days

before receiving a normal chow diet. Mice were supplied with drinking water containing

ibuprofen (20 mg/dl) ad libitum.

After behaviour testing, SDV was validated by intraperitoneal injection of cholecystokinin

(CCK-8) following fasting-induced food consumption. Each mouse was fasted for 20 h,

placed into a single cage, and injected i.p. with CCK-8 (8μg/kg, Sigma-Aldrich). After 2 h

ad libitum feeding post-injection, food intake was recorded. The anorexia signal by CCK-8

is transmitted through the vagus nerve to acutely decrease satiety.

Hippocampal dentate gyrus (DG) and Ammon’s horn microdissection

Microdissection of the DG and Ammon’s horn was performed as described

. In brief,

brains were sampled from deeply anaesthetized mice and placed in cold PBS for five

minutes. The midbrain, brainstem, and cerebellum were removed so that only the cerebrum

remained. The cerebrum was sagittally sliced along the midline of the brain. The cerebral

hemisphere was placed in ice-cold PBS in a Petri dish and the thalamus and hypothalamus

were removed under a dissection microscope. Once the medial side of the hippocampus was

Wu et al.

Page 12

Nature

. Author manuscript; available in PMC 2021 August 07.

Author Manuscript

exposed, the DG could be readily visualized. A 27-gauge needle was inserted into the edge

of DG and moved along the septo-temporal axis of the hippocampus to retrieve the DG. The

rest of the hippocampus (Ammon’s horn) was collected with sharp forceps. The tissues were

pooled from both hemispheres and placed in RNAlater (Qiagen) for storage. The samples

were stored at −80 °C until RNA extraction. The diagrams of brain microdissection were

based on the Paxinos and Franklin atlas

(Extended Data Fig. 4e).

Brain sampling for IEG expression

GF mice were transferred out of isolators and temporarily co-housed with stranger GF mice

from multiple cages. The brains of GF mice were immediately collected and dissected into

hippocampus, hypothalamus, midbrain, and brainstem regions. Brains of SPF mice were

sampled and handled following the same procedure. The primers for IEGs were adapted

from a previous publication.

RNA extraction and quantitative real-time polymerase chain reaction (qRT–PCR)

RNA extraction of brain samples was based on the manufacturer’s protocol (Rneasy Mini

Kit; Qiagen). RNA concentration and quality was measured by NanoDrop (Thermo

Scientific). RNA (1 μg) from each sample was reverse transcribed using the iScript cDNA

synthesis kit (Bio-Rad).

Gene expression in brain subregions was measured using Power SYBR Green PCR master

mix (ThermoFisher Scientific) and analysed using ABI Prism 7900HT system (Life

Technologies). Gene expression was normalized to

Actb

Gapdh

mRNA. Data are

presented as fold-change in gene expression in each group relative to the control group. The

primer sequences were adapted from the Harvard PrimerBank database

Fluorogold labelling

To label neurons in the PVN in a retrograde manner, mice were given a single intraperitoneal

injection of 100 μl Fluorogold (2% w/v; Fluorochrome). The mice were perfused six days

after injection

. Brains were sampled and stained using the standard protocols described

above.

Retrograde tracing

Stereotaxic injection was performed as described above. SPF, GF, vehicle-treated and ABX-

treated mice were stereotaxically injected into the PVN with 46 μl CTB-488 (0.5% w/v;

C22841; ThermoFisher). Vehicle-treated and ABX-treated mice were stereotaxically

injected into the BNST with 46 μl Fluorogold (2% w/v; Fluorochrome). Stereotaxic injection

coordinates (in millimetres) and diagrams (Extended Data Fig. 9a, d) were based on the

Paxinos and Franklin atlas

: PVN (AP: −0.80, ML: ±0.25; DV: −4.75), BNST (AP: 0.26,

ML: ±0.90; DV: −4.30). Mice were perfused one week after injection. GF mice were

maintained under ABX treatment as described above to limit microbial contamination. All

surgically injected animals underwent histological examination to ensure the tracers were

correctly injected.

Wu et al.

Page 13

Nature

. Author manuscript; available in PMC 2021 August 07.

Author Manuscript

Brain sample collection for c-Fos staining

All brain samples for c-Fos expression were collected one hour after the reciprocal social

interaction test. Mice were anaesthetized by intraperitoneal injection with a mixture of 100

mg/kg ketamine and 10 mg/kg xylazine. The mice were then perfused via the cardiovascular

system with PBS followed by 4% paraformaldehyde (Electron Microscopy Sciences). Brains

were removed and post-fixed in 4% paraformaldehyde for 3–5 days at 4 °C. The brains were

kept in PBS with 0.02% sodium azide at 4 °C until sectioning.

Brain sectioning and immunohistochemistry

The brains were embedded in 4% UltraPure low melting point agarose (ThermoFisher) and

were coronally sectioned by vibratome (VT1000S; Leica) at a thickness of 50 μm, with the

exception of the brain sections for Fig. 1, which were sectioned at a thickness of 100 μm.

Brain sections of 50 μm were collected and stained every 0.15 mm. Brain sections of 100 μm

were collected sequentially and stained entirely. The brain sections were stored free-floating

in PBS with 0.02% sodium azide at 4 °C until staining.

Free-floating sections were incubated with primary antibody in blocking solution (10%

horse serum, 0.1% triton X-100, and 0.02% sodium azide in PBS) overnight at room

temperature. The next day, sections were incubated with fluorescence-conjugated secondary

antibody for 1.5–2 h at room temperature. Between each step and after secondary antibody

staining, sections were thoroughly washed with PBS or PBS with 0.1% triton-X-100 at least

three times for 5 min each. The stained free-floating sections were then mounted onto

Superfrost Plus microscope slides (Fisher Scientific) in PBS. Excess PBS from adhered

sections was carefully removed. Slides were dried at room temperature for 2–5 min.

ProLong Diamond anti-fade mountant with DAPI (150–200 μl; ThermoFisher Scientific)

was applied to the slides before placing the coverslip. The slides were immersed in mountant

overnight before imaging.

Primary antibodies and their dilutions were: goat anti-c-Fos (1:250; SC-52; Santa Cruz),

mouse anti-NeuN (1:1,000; MAB377; Millpore Sigma), rabbit anti-oxytocin (1:10,000;

20068; Immunostar), rabbit anti-vasopressin (1:2,000; 20069; Immunostar), and rabbit anti-

fluorescent gold (1:1,000; AB153-I; EMD Millpore Sigma). The fluorescence-conjugated

secondary antibodies were donkey anti-goat (1:1,000), donkey anti-rabbit (1:1,000), and

donkey anti-mouse (1:1,000) (ThermoFisher Scientific).

Microscopic imaging and image analysis

Imaging was performed using a Zeiss LSM 800 inverted confocal laser scanning microscope

(Carl Zeiss) with Zen software (Carl Zeiss). For Fig. 1e, i, and the DG in Extended Data

Figs. 7, 8, confocal images were obtained by

-stacks covering the entire

-axis range of the

sections. The interval for each focal plane was 2 μm. The images were then projected in the

visualization plane with maximum intensity voxels by maximum intensity projection using

Zen software. In addition, the DG and PVN for Fig. 1e, i were imaged in tiled images

covering the entire brain area using Zen software. For other images shown in this study,

confocal images were captured in single plane with the highest intensity of DAPI. A 20×

objective lens was used for all images, except those in Fig. 1, which were taken using a 10×

Wu et al.

Page 14

Nature

. Author manuscript; available in PMC 2021 August 07.

Author Manuscript