nihms543590.pdf

The microbiota modulates gut physiology and behavioral

abnormalities associated with autism

Elaine Y. Hsiao

1,2,*

Sara W. McBride

Sophia Hsien

Gil Sharon

Embriette R. Hyde

Tyler McCue

Julian A. Codelli

Janet Chow

Sarah E. Reisman

Joseph F. Petrosino

Paul H. Patterson

1,*,†

, and

Sarkis K. Mazmanian

1,*,†

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

91125, USA

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

CA 91125, USA

Alkek Center for Metagenomics and Microbiome Research, Baylor College of Medicine,

Houston, TX 77030, USA

SUMMARY

Although autism spectrum disorder (ASD) is defined by core behavioral impairments,

gastrointestinal (GI) symptoms are commonly reported. Subsets of ASD individuals display

dysbiosis of the gut microbiota, and some exhibit increased intestinal permeability. Here we

demonstrate GI barrier defects and microbiota alterations in a mouse model displaying features of

ASD, maternal immune activation (MIA). Oral treatment of MIA offspring with the human

commensal

Bacteroides fragilis

corrects gut permeability, alters microbial composition and

ameliorates ASD-related defects in communicative, stereotypic, anxiety-like and sensorimotor

behaviors. MIA offspring display an altered serum metabolomic profile, and

B. fragilis

modulates

levels of several metabolites. Treating naïve mice with a metabolite that is increased by MIA and

restored by

B. fragilis

causes behavioral abnormalities, suggesting that gut bacterial effects on the

host metabolome impact behavior. Taken together, these findings support a gut-microbiome-brain

connection in ASD and identify a potential probiotic therapy for GI and behavioral symptoms of

autism.

INTRODUCTION

Autism spectrum disorder (ASD) is a serious neurodevelopmental condition characterized

by stereotypic behavior and deficits in language and social interaction. The reported

incidence of ASD has rapidly increased to 1 in 88 births in the United States as of 2008

(CDC, 2012), representing a significant medical and social problem. However, therapies for

treating core symptoms of autism are limited. Much research on ASD has focused on

Correspondence to: ehsiao@caltech.edu, php@caltech.edu, sarkis@caltech.edu.

†

Contributed equally

Additional Footnotes

EYH, PHP and SKM designed the study, EYH, SWM, SH, JAC and JC performed the experiments and analyzed the data, ERH, TM,

GS and JP conducted microbiota sequencing and analysis, SER contributed novel reagents, EYH, SWM, GS, JAC, PHP and SKM

wrote the manuscript. All authors discussed the results and commented on the manuscript.

Conflict of interests:

None

Publisher's Disclaimer:

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our

customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of

the resulting proof before it is published in its final citable form. Please note that during the production process errors may be

discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

NIH Public Access

Author Manuscript

Cell

. Author manuscript; available in PMC 2014 December 19.

Published in final edited form as:

Cell

. 2013 December 19; 155(7): 1451–1463. doi:10.1016/j.cell.2013.11.024.

NIH-PA Author Manuscript

genetic, behavioral and neurological aspects of disease, though the contributions of

environmental risk factors (Hallmayer et al., 2011), immune dysregulation (Onore et al.,

2012) and additional peripheral disruptions (Kohane et al., 2012) in the pathogenesis of

ASD have gained significant attention.

Among several comorbidities in ASD, gastrointestinal (GI) distress is of particular interest,

given its reported prevalence and correlation with symptom severity (Buie et al., 2010;

Coury et al., 2012). While some issues remain regarding the standardized diagnosis of GI

symptoms in ASD, abnormalities such as altered GI motility and increased intestinal

permeability have been reported by several laboratories (Boukthir et al., 2010; D’Eufemia et

al., 1996; de Magistris et al., 2010). Moreover, a recent multicenter study of over 14,000

ASD individuals reveals a higher prevalence of inflammatory bowel disease (IBD) and other

GI disorders in ASD patients compared to controls (Kohane et al., 2012). The causes of

autism-associated GI problems remain unclear, but may be linked to gut bacteria as a

number of studies report that ASD individuals exhibit altered composition of the intestinal

microbiota (Adams et al., 2011; Finegold et al., 2010; Finegold et al., 2012; Gondalia et al.,

2012; Kang et al., 2013; Parracho et al., 2005; Williams et al., 2011; Williams et al., 2012).

Though there is as yet no consistency in the specific species of microbes that are altered in

ASD versus controls, three studies employing different methodologies report significantly

elevated levels of

Clostridium

species in ASD individuals (Finegold et al., 2002; Parracho et

al., 2005; Song et al., 2004). Altogether, evidence of GI complications and microbiota

alterations in broadly defined ASD populations raises the intriguing question of whether

such abnormalities can contribute to the clinical manifestations of ASD.

Dysbiosis of the microbiota is implicated in the pathogenesis of several human disorders,

including IBD, obesity and cardiovascular disease (Blumberg and Powrie, 2012).

Commensal bacteria also affect a variety of complex behaviors, including social, emotional

and anxiety-like behaviors, and contribute to brain development and function in mice

(Collins et al., 2012; Cryan and Dinan, 2012) and humans (Tillisch et al., 2013). Long-range

interactions between the gut microbiota and brain underlie the ability of microbe-based

therapies to treat symptoms of multiple sclerosis and depression in mice (Bravo et al., 2011;

Ochoa-Reparaz et al., 2010) and the reported efficacy of probiotics in treating emotional

symptoms of chronic fatigue syndrome and psychological distress in humans (Messaoudi et

al., 2011; Rao et al., 2009).

Based on the emerging appreciation of a gut-microbiome-brain connection, we asked

whether modeling behavioral features of ASD in mice also causes GI abnormalities. Several

mouse models of genetic and/or environmental risk factors are used to study ASD. We

utilize the maternal immune activation (MIA) model, which is based on large

epidemiological studies linking maternal infection to increased autism risk in the offspring

(Atladottir et al., 2010; Gorrindo et al., 2012). This is further supported by many studies

linking increased ASD risk to familial autoimmune disease (Atladottir et al., 2009; Comi et

al., 1999) and elevated levels of inflammatory factors in the maternal blood, placenta and

amniotic fluid (Abdallah et al., 2013; Brown et al., 2013; Croen et al., 2008). Modeling MIA

in mice by injecting pregnant dams with the viral mimic poly (I:C) yields offspring that

exhibit core behavioral symptoms of ASD, as well as a common autism neuropathology

(Malkova et al., 2012; Shi et al., 2009). Furthermore, pregnant monkeys exposed to poly

(I:C) yield offspring with cardinal symptoms of ASD(Bauman et al., 2013). Although

several environmental and genetic risk factors for ASD have been investigated in animals,

GI abnormalities have not been reported in preclinical models of ASD. We show herein that

offspring of MIA mice, which display ASD-like behaviors, have defects in intestinal

integrity and alterations in the composition of the commensal microbiota that are analogous

to features reported in human ASD. To explore the potential contribution of GI

Hsiao et al.

Page 2

Cell

. Author manuscript; available in PMC 2014 December 19.

NIH-PA Author Manuscript

complications to core ASD symptoms, we examine whether treatment with the gut

bacterium

Bacteroides fragilis,

demonstrated to correct GI pathology in mouse models of

colitis (Mazmanian et al., 2008) and to protect against neuroinflammation in mouse models

of multiple sclerosis (Ochoa-Reparaz et al., 2010), impacts ASD-related GI and/or

behavioral abnormalities in MIA offspring. Our findings suggest that targeting the

microbiome may represent a novel approach for treating neurodevelopmental disorders such

as autism.

RESULTS

Offspring of Immune-Activated Mothers Exhibit GI Symptoms of Human ASD

Subsets of ASD children are reported to display GI abnormalities, including increased

intestinal permeability or “leaky gut” (D’Eufemia et al., 1996; de Magistris et al., 2010;

Ibrahim et al., 2009). We find that adult MIA offspring, which exhibit cardinal behavioral

and neuropathological symptoms of ASD (Malkova et al., 2012), also have a significant

deficit in intestinal barrier integrity, as reflected by increased translocation of FITC-dextran

across the intestinal epithelium, into the circulation (Figure 1A, left). This MIA-associated

increase in intestinal permeability is similar to that of mice treated with dextran sodium

sulfate (DSS), which induces experimental colitis (Figure 1A, left). Deficits in intestinal

integrity are detectable in 3-week-old MIA offspring (Figure 1A, right), indicating that the

abnormality is established during early life. Consistent with the leaky gut phenotype, colons

from adult MIA offspring contain decreased gene expression of

ZO-1

ZO-2

OCLN

and

CLDN8

, and increased expression of

CLDN15

(Figure 1B). Deficient expression of

ZO-1

also observed in small intestines of adult MIA offspring (Figure S1A), demonstrating a

widespread defect in intestinal barrier integrity. Gut permeability is commonly associated

with an altered immune response (Turner, 2009). Accordingly, colons from adult MIA

offspring display increased levels of interleukin-6 (IL-6) mRNA and protein (Figures 1C

and 1D) and decreased levels of the cytokines/chemokines IL-12p40/p70, IP-10, MIG and

MIP-1

(Figure 1D). Small intestines from MIA offspring also exhibit altered cytokine/

chemokine profiles (Figure S1C). Changes in intestinal cytokines are not accompanied by

overt GI pathology, as assessed by histological examination of gross epithelial morphology

from hematoxylin- and eosin-stained sections (data not shown). Overall, we find that adult

offspring of immune-activated mothers exhibit increased gut permeability and abnormal

intestinal cytokine profiles, features similar to those found in subsets of ASD.

MIA Offspring Display Dysbiosis of the Gut Microbiota

Abnormalities related to the microbiota have been identified in ASD individuals, including

disrupted community composition (Adams et al., 2011; Finegold et al., 2010; Finegold et al.,

2012; Gondalia et al., 2012; Parracho et al., 2005; Williams et al., 2011; Williams et al.,

2012). although it is important to note that a well-defined ASD-associated microbial

signature is lacking thus far. To evaluate whether MIA induces microbiota alterations, we

surveyed the fecal bacterial population by 16S rRNA gene sequencing of samples isolated

from adult MIA or control offspring. Alpha diversity, i.e., species richness and evenness, did

not differ significantly between control and MIA offspring, as measured by several indices

(Figures S2A and S2B). In contrast, unweighted UniFrac analysis, which measures the

degree of phylogenetic similarity between microbial communities, reveals a strong effect of

MIA on the gut microbiota of adult offspring (Figure 2). MIA samples cluster distinctly

from controls by principal coordinate analysis (PCoA) and differ significantly in

composition (Table S3, with ANOSIM R=0.2829, p=0.0030), indicating robust differences

in the membership of gut bacteria between MIA offspring and controls (Figure 2A). The

effect of MIA on altering the gut microbiota is further evident when sequences from the

classes Clostridia and Bacteroidia, which account for approximately 90.1% of total reads,

Hsiao et al.

Page 3

Cell

. Author manuscript; available in PMC 2014 December 19.

NIH-PA Author Manuscript

are exclusively examined by PCoA (R=0.2331, p=0.0070; Figure 2B), but not when

Clostridia and Bacteroidia sequences are specifically excluded from PCoA of all other

bacterial classes (R=0.1051, p=0.0700; Figure 2C). This indicates that changes in the

diversity of Clostridia and Bacteroidia operational taxonomic units (OTUs) are the primary

drivers of gut microbiota differences between MIA offspring and controls.

67 of the 1,474 OTUs detected across any of the samples discriminate between treatment

groups, including those assigned to the bacterial families

Lachnospiraceae

Ruminococcaceae

Erysipelotrichaceae

Alcaligenaceae

Porphyromonadaceae

Prevotellaceae

and

Rikenellaceae

, and unclassified Bacteroidales (Figure 2D and Table S1).

Of these 67 discriminatory OTUs (relative abundance: 13.3±1.65% control, 15.93±0.62%

MIA), 19 are more abundant in the control samples and 48 are more abundant in MIA

samples. Consistent with the PCoA results (Figures 2A–C), the majority of OTUs that

discriminate MIA offspring from controls are assigned to the classes Bacteroidia (45/67

OTUs or67.2%; 12.02±1.62% control, 13.48±0.75% MIA) and Clostridia (14/67 OTUs

or20.9%; 1.00±0.25% control, 1.58±0.34% MIA). Interestingly,

Porphyromonadaceae

Prevotellaceae

, unclassified Bacteriodales (36/45 discriminatory Bacteroidial OTUs or 80%;

4.46±0.66% control, 11.58±0.86% MIA), and

Lachnospiriceae

(8/14 discriminatory

Clostridial OTUs or 57%; 0.28±0.06% control, 1.13±0.26% MIA) were more abundant in

MIA offspring. Conversely,

Ruminococcaceae

(2 OTUs),

Erysipelotrichaceae

(2 OTUs),

and the beta Proteobacteria family

Alcaligenaceae

(2 OTUs) were more abundant in control

offspring (Figure 2D and Table S1; 0.95±0.31% control, 0.05±0.01% MIA). This suggests

that specific

Lachnospiraceae

, along with other Bacteroidial species, may play a role in

MIA pathogenesis, while other taxa may be protective. Importantly, there is no significant

difference in the overall relative abundance of Clostridia (13.63 ± 2.54% vs 14.44 ± 2.84%;

p=0.8340) and Bacteroidia (76.25 ± 3.22% vs 76.22 ± 3.46%; p=0.9943) between MIA

offspring and controls (Figure 2E, left), indicating that alterations in the membership of

Clostridial and Bacteroidial OTUs drive major changes in the gut microbiota between

experimental groups.

Overall, we find that MIA leads to dysbiosis of the gut microbiota, driven primarily by

alterations in specific OTUs of the bacterial classes Clostridia and Bacteroidia. Changes in

OTUs classified as

Lachnospiraceae

and

Ruminococcaceae

of the order Clostridiales

parallel select reports of increased

Clostridium

species in the feces of subjects with ASD

(Finegold et al., 2012). Altogether, modeling MIA in mice induces not only behavioral and

neuropathological features of ASD, but also microbiome changes as described in subsets of

ASD individuals.

Bacteroides fragilis

Improves Gut Barrier Integrity in MIA Offspring

Gut microbes play an important role in the development, maintenance and repair of the

intestinal epithelium (Turner, 2009). To determine whether targeting the gut microbiota

could impact MIA-associated GI abnormalities, we treated mice with the human commensal

B. fragilis

at weaning, and tested for GI abnormalities at 8 weeks of age.

B. fragilis

has

previously been shown to ameliorate experimental colitis (Mazmanian et al., 2008; Round

and Mazmanian, 2010). Remarkably,

B. fragilis

treatment corrects intestinal permeability in

MIA offspring (Figure 3A). In addition,

B. fragilis

treatment ameliorates MIA-associated

changes in expression of

CLDNs

8 and 15 (Figure 3B). Similar changes are observed in

protein levels of CLDNs8 and 15 in the colon, with restoration by

B. fragilis

treatment

(Figures 3C and 3D). No effects of

B. fragilis

on tight junction expression are observed in

the small intestine (Figure S1B), consistent with the fact that

Bacteroides

species are

predominantly found in the colon. Finally, the presence of GI defects prior to probiotic

Hsiao et al.

Page 4

Cell

. Author manuscript; available in PMC 2014 December 19.

NIH-PA Author Manuscript

administration (Figure 1A, right) suggests that

B. fragilis

may treat, rather than prevent, this

ASD-related pathology in MIA offspring.

B. fragilis

treatment also restores MIA-associated increases in colon IL-6 mRNA and

protein levels (Figures 3E and 3F). Levels of other cytokines are altered in both colons and

small intestines of MIA offspring (Figures 1D and S1C), but these are not affected by

fragilis

treatment, revealing specificity for IL-6. We further find that recombinant IL-6

treatment can modulate colon levels of both CLDN 8 and 15

in vivo

and in

in vitro

colon

organ cultures (data not shown), suggesting that

B. fragilis

-mediated restoration of colonic

IL-6 levels could underlie its effects on gut permeability. Collectively, these findings

demonstrate that

B. fragilis

treatment of MIA offspring improves defects in GI barrier

integrity, and corrects alterations in tight junction and cytokine expression.

B. fragilis

Treatment Restores Specific Microbiota Changes in MIA

Offspring

The finding that

B. fragilis

ameliorates GI defects in MIA offspring prompted us to examine

its effects on the intestinal microbiota. No significant differences are observed following

fragilis

treatment of MIA offspring by PCoA (ANOSIM R=0.0060 p=0.4470; Table 3), in

microbiota richness (PD: p=0.2980, Observed Species: p=0.5440) and evenness (Gini:

p=0.6110, Simpson Evenness: p=0.5600; Figures 4A, S2A and S2B), or in relative

abundance at the class level (Figure S2C). However, evaluation of key OTUs that

discriminate adult MIA offspring from controls reveals that

B. fragilis

treatment

significantly alters levels of 35 OTUs (Table S2). Specifically, MIA offspring treated with

B. fragilis

display significant restoration in the relative abundance of 6 out of the 67 OTUs

found to discriminate MIA from control offspring (Figure 4B and Table S2), which are

taxonomically assigned as unclassified Bacteroidia and Clostridia of the family

Lachnospiraceae

(Figure 4B and Table S2). Notably, these alterations occur in the absence

of persistent colonization of

B. fragilis

, which remains undetectable in fecal and cecal

samples isolated from treated MIA offspring (Figures S2D and S2E). Phylogenetic

reconstruction of the OTUs that are altered by MIA and restored by

B. fragilis

treatment

reveals that the Bacteroidia OTUs cluster together into a monophyletic group and the

Lachnospiraceae

OTUs cluster into 2monophyletic groups (Figure 4C). This result suggests

that, although treatment of MIA offspring with

B. fragilis

may not lead to persistent

colonization, this probiotic corrects the relative abundance of specific groups of related

microbes of the

Lachnospiraceae

family as well as unclassified Bacteriodales. Altogether,

we demonstrate that treatment of MIA offspring with

B. fragilis

ameliorates MIA-associated

dysbiosis of the commensal microbiota.

B. fragilis

Treatment Corrects ASD-Related Behavioral Abnormalities

Studies suggest that GI issues in children with ASD can contribute to the development,

persistence, and/or severity of symptoms (Buie et al., 2010; Coury et al., 2012). To explore

the potential impact of GI dysfunction on core ASD behavioral abnormalities, we tested

whether

B. fragilis

treatment impacts autism-related behaviors. We replicated previous

findings that adult MIA offspring display cardinal behavioral features of ASD (Malkova et

al., 2012). Open field exploration involves mapping an animal’s movement in an open arena

to measure locomotion and anxiety (Bourin et al., 2007). MIA offspring display decreased

entries and time spent in the center of the arena, which is indicative of anxiety-like behavior

(Figure 5A; compare saline (S) to poly (I:C) (P)). The pre-pulse inhibition (PPI) task

measures the ability of an animal to inhibit its startle in response to an acoustic tone

(“pulse”) when it is preceded by a lower-intensity stimulus (“pre-pulse”). Deficiencies in

PPI are a measure of impaired sensorimotor gating, and are observed in several

Hsiao et al.

Page 5

Cell

. Author manuscript; available in PMC 2014 December 19.

NIH-PA Author Manuscript

neurodevelopmental disorders, including autism (Perry et al., 2007). MIA offspring exhibit

decreased PPI in response to 5 or 15 db pre-pulses (Figure 5B). The marble burying test

measures the propensity of mice to engage repetitively in a natural digging behavior that is

not confounded by anxiety (Thomas et al., 2009). MIA offspring display increased

stereotyped marble burying compared to controls (Figure 5C). Ultrasonic vocalizations are

used to measure communication by mice, wherein calls of varying types and motifs are

produced indifferent social paradigms (Grimsley et al., 2011). MIA offspring exhibit ASD-

related deficits in communication, as indicated by reduced number and duration of ultrasonic

vocalizations produced in response to a social encounter (Figure 5D). Finally, the three-

chamber social test is used to measure ASD-related impairments in social interaction

(Silverman et al., 2010). MIA offspring exhibit deficits in both sociability, or preference to

interact with a novel mouse over a novel object, and social preference (social novelty), or

preference to interact with an unfamiliar versus a familiar mouse (Figure 5E and 5F).

Altogether, these behavioral assays measure the cardinal diagnostic symptoms of ASD, in

addition to ASD-associated anxiety and deficient sensorimotor gating, and confirm that the

MIA model reflects behavioral features of autism.

Remarkably, oral treatment with

B. fragilis

ameliorates many of these ASD-related

behaviors.

B. fragilis

-treated MIA offspring do not exhibit anxiety-like behavior in the open

field (Figure 5A; compare poly(I:C) (P) to poly(I:C)+

B. fragilis

(P+BF)), as shown by

restoration in the number of center entries and duration of time spent in the center of the

arena.

B. fragilis

improves sensorimotor gating in MIA offspring, as indicated by increased

combined PPI in response to 5 and 15 db pre-pulses (Figure 5B), with no significant effect

on the intensity of startle to the acoustic stimulus (data not shown).

B. fragilis

-treated mice

also exhibit decreased levels of stereotyped marble burying and restored communicative

behavior(Figure 5C and 5D). Interestingly,

B. fragilis

raises the duration per call by MIA

offspring to levels exceeding those observed in saline controls (Figure 5D), suggesting that

despite normalization of the propensity to communicate (number of calls), there is a

qualitative difference in the types of calls generated with enrichment of longer syllables.

Although

B. fragilis

-treated MIA offspring exhibit improved communicative, repetitive,

anxiety-like and sensorimotor behavior, they retain deficits in sociability and social

preference (Figure 5E). Interestingly, this parallels the inability to improve social behavior

by administration of risperidone to ASD individuals (Canitano and Scandurra, 2008) and to

CNTNAP2 knockout mice, a genetic mouse model for ASD (Penagarikano et al., 2011).

These data suggest that there may be differences in the circuitry or circuit plasticity

governing social behavior as compared to the other behaviors, and that

B. fragilis

treatment

may modulate specific circuits during amelioration of ASD-related behavioral defects.

Interestingly, behavioral improvement in response to

B. fragilis

treatment of MIA offspring

is not associated with changes in systemic immunity (Figure S3A–D) and is not dependent

on polysaccharide A (PSA), a molecule previously identified to confer immunomodulatory

effects by

B. fragilis

(Figure S3E) (Mazmanian et al., 2008; Ochoa-Reparaz et al., 2010;

Round and Mazmanian, 2010). Furthermore, amelioration of behavior is not specific to

fragilis

, as similar treatment with

Bacteroides thetaiotaomicron

also significantly improves

anxiety-like, repetitive and communicative behavior in MIA offspring (Figure S4). This is

consistent with our finding that

B. fragilis

treatment improves ASD-related behavior in the

absence of evident colonization of

B. fragilis

in the GI tract (Figure S2D and S2E), and thus,

may be functioning through persistent shifts in the composition of resident microbiota

(Figure 4). There is, however, some degree of specificity to bacterial treatment, as

administration of

Enterococcus faecalis

has no effect on anxiety-like and repetitive behavior

in MIA offspring (data not shown).

Hsiao et al.

Page 6

Cell

. Author manuscript; available in PMC 2014 December 19.

NIH-PA Author Manuscript

The Serum Metabolome is Modulated by MIA and

B. fragilis

Treatment

Metabolomic studies have shown that gut microbial products are found in many extra-

intestinal tissues, and molecules derived from the microbiota may influence metabolic,

immunologic and behavioral phenotypes in mice and humans (Blumberg and Powrie, 2012;

Nicholson et al., 2012). Given that MIA offspring display increased gut permeability, tight

junction defects and dysbiosis, we hypothesized that gut bacteria may affect the metabolome

of mice. We utilized gas chromatography/liquid chromatography with mass spectrometry

(GC/LC-MS)-based metabolomic profiling to identify MIA-associated changes in serum

metabolites. 322 metabolites were detected in sera from adult mice (Table S5). Interestingly,

MIA leads to statistically significant alterations in 8% of all serum metabolites detected

(Table S4). Furthermore, postnatal

B. fragilis

treatment has a significant effect on the serum

metabolome, altering 34% of all metabolites detected (Table S5 and Figure S5).

B. fragilis

Treatment Corrects Levels of MIA-Induced Serum Metabolites

Consistent with the notion that increased intestinal permeability leads to leakage of gut-

derived metabolites into the bloodstream, we hypothesized that

B. fragilis

-mediated

improvement of intestinal barrier integrity would restore serum levels of certain metabolites.

We therefore focused on serum metabolites that are significantly altered by MIA treatment

and restored to control levels by

B. fragilis

treatment. The most dramatically affected

metabolite is 4-ethylphenylsulfate (4EPS), displaying a striking 46-fold increase in serum

levels of MIA offspring that is completely restored by

B. fragilis

treatment (Figure 6A). This

metabolite is of particular interest because of the reported production of 4EPS by GI

microbes and proposed role for 4EPS in communication by mice (Lafaye et al., 2004).

Moreover, we find that compared to conventionally-colonized (SPF; specific pathogen free)

mice, germ-free (GF) mice display nearly undetectable levels of serum 4EPS, indicating that

serum 4EPS is derived from, or modulated by, the commensal microbiota (Figure 6B).

Interestingly, 4EPS is suggested to be a uremic toxin, as is

-cresol (4-methylphenol), a

chemically related metabolite reported to be a possible urinary biomarker for autism (Altieri

et al., 2011; Persico and Napolioni, 2013). MIA offspring also exhibit elevated levels of

serum

-cresol, although the increase does not reach statistical significance (Table S5). The

fact that 4EPS shares close structural similarity to the toxic sulfated form of

cresol (4-

methylphenylsulfate; 4MPS) is intriguing as the two metabolites may exhibit functional

overlap (Figure S6A).

In addition to 4EPS, MIA offspring display significantly increased levels of serum

indolepyruvate, a key molecule of the tryptophan metabolism pathway, which is restored to

control levels by

B. fragilis

treatment (Figure 6A). Indolepyruvate is generated by

tryptophan catabolism and, like 4EPS, is believed to be produced by gut microbes (Smith

and Macfarlane, 1997) (Figure S6B). Moreover, the elevation in serum indolepyruvate

observed in MIA offspring is reminiscent of reported increases inindolyl-3-acryloylglycine

(IAG) in human ASD (Bull et al., 2003), which is involved in GI homeostasis and produced

by bacterial metabolism (Keszthelyi et al., 2009). MIA offspring also exhibit increased

levels of serum serotonin (0.05<

<0.10; Tables S3 and S4), reflecting another alternation in

tryptophan metabolism, analogous to the well-established hyperserotonemia endophenotype

of autism. MIA also leads to altered serum glycolate, imidazole propionate and N-

acetylserine levels (Figure 6A), which are corrected by

B. fragilis

treatment. How changes

in these metabolites may be relevant to ASD or GI dysfunction is currently unknown, but

may be an exciting area for future study. These findings demonstrate that specific

metabolites are altered in MIA offspring and normalized by

B. fragilis

treatment, with at

least two molecules (4EPS and indolepyruvate) having potential relevance to ASD.

Hsiao et al.

Page 7

Cell

. Author manuscript; available in PMC 2014 December 19.

NIH-PA Author Manuscript

A Serum Metabolite Induces ASD-Related Behavior

MIA-dependent increases of specific metabolites, and their restoration by

B. fragilis

, suggest

that small molecules may play a role in ASD-related behaviors. To test this hypothesis, we

examined whether increasing serum 4EPS is sufficient to cause any ASD-related behavioral

abnormalities in naïve mice. Mice were treated with 4EPS potassium salt (Figures S7A–C)

or vehicle, daily from 3 weeks of age (when MIA offspring display gut permeability) to 6

weeks of age (when behavior testing begins). Remarkably, systemic administration of the

single metabolite, 4EPS, to naïve wild-type mice is sufficient to induce anxiety-like

behavior similar to that observed in MIA offspring (Figure 6C). Relative to vehicle-treated

controls, mice exposed to 4EPS travel comparable distances in the open field but spend less

time in the center arena (Figure 6C). Also, in the PPI test, 4EPS-treated mice exhibit

increased intensity of startle in response to the unconditioned primary stimulus, but no

significant alterations in PPI (Figure 6D), representing anxiety-associated potentiation of the

startle reflex (Bourin et al., 2007). Conversely, there are no significant differences between

4EPS-treated versus saline-treated mice in marble burying or USV behavior (Figures S7D

and S7E), suggesting that elevating serum 4EPS levels specifically promotes anxiety-like

behavior. While not a core diagnostic criterion, anxiety is a common co-morbidity that may

contribute to cardinal ASD symptoms. Furthermore, it is possible that complex behaviors

may be modulated by combinations of metabolites. In summary, these data reveal that

elevated systemic levels of a metabolite regulated by gut microbes causes an ASD-related

behavior, suggesting that molecular connections between the gut and the brain maybe

associated with autism.

DISCUSSION

ASD is a complex disorder with poorly defined etiologies and no effective or targeted cure.

Here we demonstrate that in addition to displaying cardinal behavioral and

neuropathological symptoms of ASD, offspring of immune-activated mothers exhibit

defective GI integrity, dysbiosis of the commensal microbiota and alterations in serum

metabolites that are similar to endophenotypes observed in ASD individuals. Collectively,

these findings reveal MIA as a model with face validity for co-morbid GI symptoms and

microbiome profiles found in ASD. We find that oral treatment with the human commensal

B. fragilis

corrects intestinal permeability defects, as well as altered levels of tight junction

proteins and cytokines. The ability of

B. fragilis

to selectively ameliorate MIA-associated

increases in colon IL-6 is interesting as this cytokineis required for the development of

behavioral deficits in MIA offspring (Smith et al., 2007). Many cytokines including IL-6

regulate tight junction expression and intestinal barrier integrity, and further, a variety of

enteric microbes are known to regulate intestinal tight junction and cytokine levels (Turner,

2009). Our study suggests that

B. fragilis

is able to ameliorate leaky gut by directly targeting

tight junction expression, cytokine production and/or microbiome composition. Intriguingly,

recent analysis in humans shows that among the

Bacteroidaceae

family, only a single

phylotype most closely related to

B. fragilis

is selectively depleted in ASD children

compared to matched controls, and most dramatically in those subjects with more severe GI

issues (Dae-Wook Kang and Rosa Krajmalnik-Brown, personal communication). Thus, the

correlation between

B. fragilis

and improved intestinal health is present in both mice and

humans.

Consistent with the role of GI microbes in regulating intestinal permeability and metabolic

homeostasis (Nicholson et al., 2012; Wikoff et al., 2009), we show that

B. fragilis

treatment

corrects MIA-associated changes in specific serum metabolites that appear to have a gut

origin, suggesting

B. fragilis

may prevent leakage of harmful molecules from the GI lumen.

In a proof-of-concept test of the this hypothesis, we reveal that the microbially-modulated

Hsiao et al.

Page 8

Cell

. Author manuscript; available in PMC 2014 December 19.

NIH-PA Author Manuscript

metabolite 4EPS, which is elevated in the circulation by MIA and restored by

B. fragilis

treatment, is sufficient to induce anxiety-like behavior in naïve mice. These data indicate

that metabolomic changes contribute to the onset and/or persistence of autism-related

behavioral abnormalities. Notably, we show that commensal microbes are required for the

production of serum 4EPS in mice. Several species of

Clostridium

are believed to be

producers of the precursor 4-ethylphenol (Nicholson et al., 2012), consistent with our

findings that levels of the

Lachnospiraceae

family of Clostridia and serum 4EPS are

elevated in MIA offspring, and both are corrected by

B. fragilis

treatment. Moreover, the

structural similarity of 4EPS to

cresol, which also derives from

Clostridium

species

(Persico and Napolioni, 2013), suggests they may be produced through similar biosynthetic

pathways (see Figure S6A). Although not all autism-like behaviors are affected by 4EPS

alone, our results warrant the examination of several other serum metabolites, perhaps in

combination, for their potential to impact the spectrum of autism-related behaviors.

Remarkably,

B. fragilis

treatment ameliorates abnormal communicative, stereotyped,

sensorimotor and anxiety-like behaviors in MIA offspring, supporting emerging evidence

for a gut-brain link in ASD. A role for commensal bacteria in modulating behavioris

supported by studies revealing differences between GF and SPF mice in anxiety-like (Heijtz

et al., 2011), locomotor and social behavior (Desbonnet et al., 2013). Treatment of SPF

animals with commensal microbes can ameliorate depressive (Bravo et al., 2011) and

anxiety-like behavior in mice (Bercik et al., 2011), and probiotic treatment has been

beneficial in treating psychological distress and chronic fatigue symptoms in humans

(Messaoudi et al., 2011; Rao et al., 2009). Our findings provide a novel mechanism by

which a human commensal bacterium can improve ASD-related GI deficits and behavioral

abnormalities in mice, possibly explaining the rapid increase in ASD prevalence by

identifying the microbiome as a critical environmental contributor to disease. We propose

the transformative concept that autism is, at least in part, a disease involving the gut that

impacts the immune, metabolic and nervous systems, and that microbiome-mediated

therapies may be a safe and effective treatment for ASD.

EXPERIMENTAL PROCEDURES

See supplemental information for additional details and references.

Animals and MIA

Pregnant C57BL/6N mice (Charles River; Wilmington, MA) were injected i.p. on E12.5

with saline or 20 mg/kg poly(I:C) according to methods described in ref. (Smith et al.,

2007). All animal experiments were approved by the Caltech IACUC.

B. fragilis

treatment

Mice were selected at random for treatment with

Bacteroides fragilis

(ATCC 9343) or

vehicle, every other day for 6 days at weaning. 1×10

CFU of freshly grown

B. fragilis

, or

vehicle, in 1.5% sodium bicarbonate was administered in sugar-free applesauce over

standard food pellets. The same procedure was used for mutant

B. fragilis

PSA and

thetaiotaomicron

Intestinal permeability assay

Mice were fasted for 4 hours before gavage with 0.6 g/kg 4 kDa FITC-dextran (Sigma

Aldrich). 4 hours later, serum samples were read for fluorescence intensity at 521 nm using

an xFluor4 spectrometer (Tecan). Mice were fed 3% dextran sulfate sodium salt (DSS; MP

Biomedicals) in drinking water for 7 days to induce colitis.

Hsiao et al.

Page 9

Cell

. Author manuscript; available in PMC 2014 December 19.

NIH-PA Author Manuscript

Intestinal qRT-PCR, Western blots, and cytokine profiles

Gut tissue was flushed with HBSS and either a) homogenized in Trizol for RNA isolation

and reverse transcription according to ref. (Hsiao and Patterson, 2011) or b) homogenized in

Tissue Extraction Reagent I (Invitrogen) containing protease inhibitors (Roche) for protein

assays. For cytokine profiling, mouse 20-plex cytokine arrays (Invitrogen) were run on the

Luminex FLEXMAP 3D platform by the Clinical Immunobiology Correlative Studies

Laboratory at the City of Hope (Duarte, CA). Western blots were conducted according to

standard methods and probed with rabbit anti-claudin 8 or rabbit anti-claudin 15 (Invitrogen)

at 1:100 dilution.

Microbial DNA extraction, 16S rRNA gene amplification and pyrosequencing

Bacterial DNA was extracted from mouse fecal pellets using the MoBioPowerSoil Kit

following protocols benchmarked by the NIH Human Microbiome Project. The V3-V5

regions of the 16S rRNA gene were PCR amplified using individually barcoded universal

primers containing linker sequences for sequencing using a multiplexed 454-Titanium

pyrosequencer.

16S rRNA gene sequence analysis

16S data was processed and its diversity was analyzed using QIIME 1.6 software package

(Caporaso et al., 2010b). OTUs were assigned taxonomic classification using the basic

BLAST classifier (Altschul et al, 1990). For tree-based alpha- and beta diversity analyses,

representative sequences for each OTU were aligned using PyNAST (Caporaso et al.,

2010a) and a phylogenetic tree was constructed based on this alignment using Fast Tree

(Price et al., 2009). Beta diversity was assessed from unweighted UniFrac, using the

Analysis of Similarity (ANOSIM; Fierer et al 2010), Permutational multivariate analysis of

variance (PERMANOVA;(Anderson, 2008)), Permutational analysis of multivariate

dispersions (PERMDISP;(Anderson et al., 2006)), and Moran’s I.

Identification of differences in specific OTUs

Key OTUs were identified using: (1) Metastats comparison (White et al., 2009) and (2)

Random Forests algorithm, under QIIME (Knights et al., 2011) or coupled with Boruta

feature selection, in the Genbore emicrobiome tool set (Riehle et al., 2012), and (3) Galaxy

platform-based LDA Effect Size analysis (LEfSe; (Segata et al., 2011)). Key OTUs were

than aligned using the SINA aligner (

http://www.arb-silva.de/aligner/

; (Pruesse et al., 2012),

compared to the SILVA reference database release 111 (Quast et al., 2013) using Arb

(Ludwig et al., 2004) and visualized using Fig Tree (

http://tree.bio.ed.ac.uk/software/

figtree/

). Heat maps of key OTUs were generated by extracting their relative abundance

from the OTU table and clustering data by correlation using Cluster 3.0 (de Hoon et al.,

2004). Abundance data was visualized using Java TreeView (Saldanha, 2004).

Behavioral testing

MIA and control offspring were behaviorally tested as in refs. (Hsiao et al., 2012) and

(Malkova et al., 2012). Mice were tested beginning at 6 weeks of age for pre-pulse

inhibition, open field exploration, marble burying, social interaction and adult ultrasonic

vocalizations

Metabolomics screening

Serum samples were extracted and analyzed on GC/MS, LC/MS and LC/MS/MS platforms

by Metabolon, Inc. For LC/MS and LC/MS/MS, samples were run on a Waters ACQUITY

UPLC and Thermo-Finnigan LTQ mass spectrometer. For GC/MS, samples were analyzed

on a Thermo-Finnigan Trace DSQ fast-scanning single-quadrupole mass spectrometer using

Hsiao et al.

Page 10

Cell

. Author manuscript; available in PMC 2014 December 19.

NIH-PA Author Manuscript

electron impact ionization. Chemical entities were identified by comparison to metabolomic

library entries of purified standards.

4EPS sufficiency experiments

Wildtype mice were injected i.p. with saline or 30 mg/kg 4EPS potassium salt daily from 3

to 6 weeks of age. A dose-response curve was generated by measuring serum 4EPS levels at

various times after i.p. injection of 30 mg/kg 4EPS (Figure S7C). Mice were behaviorally

tested as described above from 6 to 9 weeks of age.

Statistical Analysis

Statistical analysis was performed using Prism software (Graphpad). Data are plotted in the

figures as mean ± SEM. Differences between two treatment groups were assessed using

two-tailed, unpaired Student

test with Welch’s correction. Differences among three or

more groups were assessed using one-way ANOVA with Bonferroni post hoc test. Two-way

repeated measures ANOVA with Bonferroni post hoc test was used for analysis of PPI and

CD4+ T-cell stimulation data. Two-way ANOVA with contrasts was used for analysis of the

metabolite data. Significant differences are indicated in the figures by *

<0.05, **

p<0.01,

***

<0.001. Notable near-significant differences (0.5<

<0.1) are indicated in the figures.

Notable non-significant (and non-near significant) differences are indicated in the figures by

“n.s.”.

Supplementary Material

Refer to Web version on PubMed Central for supplementary material.

Acknowledgments

We acknowledge Reyna Sauza, Jaime Rodriguez and Taren Thron for caring for the animals, Dr. Michael

Fischbach (UCSF) for advising on pathways of 4EPS and indolepyruvate synthesis, Dr. NadimAjami (Baylor) for

providing helpful comments on the manuscript, Greg Donaldson (Caltech) for conducting experiments on microbial

viability, Dr. Kym Faull (UCLA) for conducting pilot GC/MS experiments, Dr. Alessio Fasano (Massachusetts

General) for conducting pilot microbiota sequencing experiments and Dr. Jerrold Turner (U Chicago) for providing

histological analysis of intestinal sections. This work was supported by a Caltech Innovation Fellowship (to EYH),

Autism Speaks Weatherstone Fellowship (to EYH), NIH/NRSA Pre-doctoral Fellowship (to EYH), Human

Frontiers Science Program Fellowship (to GS), DOD Graduate Fellowship (to JAC), NSF Graduate Research

Fellowship (to JAC), Autism Speaks Trailblazer Award (to PHP and SKM), Caltech Innovation grant (to PHP and

SKM), Congressionally Directed Medical Research Award (to PHP and SKM), Weston Havens Award (to PHP and

SKM), Callie D. McGrath Charitable Foundation awards (to PHP) and NIMH grant MH100556 (to PHP and SKM).

References

Abdallah MW, Larsen N, Grove J, Norgaard-Pedersen B, Thorsen P, Mortensen EL, Hougaard DM.

Amniotic fluid inflammatory cytokines: Potential markers of immunologic dysfunction in autism

spectrum disorders. The world journal of biological psychiatry : the official journal of the World

Federation of Societies of Biological Psychiatry. 2013; 14:528–538. [PubMed: 22175527]

Adams JB, Johansen LJ, Powell LD, Quig D, Rubin RA. Gastrointestinal flora and gastrointestinal

status in children with autism--comparisons to typical children and correlation with autism severity.

BMC Gastroenterol. 2011; 11:22. [PubMed: 21410934]

Altieri L, Neri C, Sacco R, Curatolo P, Benvenuto A, Muratori F, Santocchi E, Bravaccio C, Lenti C,

Saccani M, et al. Urinary p-cresol is elevated in small children with severe autism spectrum

disorder. Biomarkers. 2011; 16:252–260. [PubMed: 21329489]

Atladottir HO, Pedersen MG, Thorsen P, Mortensen PB, Deleuran B, Eaton WW, Parner ET.

Association of family history of autoimmune diseases and autism spectrum disorders. Pediatrics.

2009; 124:687–694. [PubMed: 19581261]

Hsiao et al.

Page 11

Cell

. Author manuscript; available in PMC 2014 December 19.

NIH-PA Author Manuscript

Atladottir HO, Thorsen P, Ostergaard L, Schendel DE, Lemcke S, Abdallah M, Parner ET. Maternal

infection requiring hospitalization during pregnancy and autism spectrum disorders. J Autism Dev

Disord. 2010; 40:1423–1430. [PubMed: 20414802]

Bauman MD, Iosif AM, Smith SE, Bregere C, Amaral DG, Patterson PH. Activation of the Maternal

Immune System During Pregnancy Alters Behavioral Development of Rhesus Monkey Offspring.

Biological psychiatry. 2013

Bercik P, Park AJ, Sinclair D, Khoshdel A, Lu J, Huang X, Deng Y, Blennerhassett PA, Fahnestock

M, Moine D, et al. The anxiolytic effect of Bifidobacterium longum NCC3001 involves vagal

pathways for gut-brain communication. Neurogastroenterol Motil. 2011; 23:1132–1139. [PubMed:

21988661]

Blumberg R, Powrie F. Microbiota, disease, and back to health: a metastable journey. Sci Transl Med.

2012; 4:137rv137.

Boukthir S, Matoussi N, Belhadj A, Mammou S, Dlala SB, Helayem M, Rocchiccioli F, Bouzaidi S,

Abdennebi M. Abnormal intestinal permeability in children with autism. Tunis Med. 2010; 88:685–

686. [PubMed: 20812190]

Bourin M, Petit-Demouliere B, Dhonnchadha BN, Hascoet M. Animal models of anxiety in mice.

Fundamental & clinical pharmacology. 2007; 21:567–574. [PubMed: 18034657]

Bravo JA, Forsythe P, Chew MV, Escaravage E, Savignac HM, Dinan TG, Bienenstock J, Cryan JF.

Ingestion of Lactobacillus strain regulates emotional behavior and central GABA receptor

expression in a mouse via the vagus nerve. Proc Natl Acad Sci U S A. 2011; 108:16050–16055.

[PubMed: 21876150]

Brown AS, Sourander A, Hinkka-Yli-Salomaki S, McKeague IW, Sundvall J, Surcel HM. Elevated

maternal C-reactive protein and autism in a national birth cohort. Mol Psychiatry. 2013

Buie T, Campbell DB, Fuchs GJ 3rd, Furuta GT, Levy J, Vandewater J, Whitaker AH, Atkins D,

Bauman ML, Beaudet AL, et al. Evaluation, diagnosis, and treatment of gastrointestinal disorders

in individuals with ASDs: a consensus report. Pediatrics. 2010; 125(Suppl 1):S1–18. [PubMed:

20048083]

Canitano R, Scandurra V. Risperidone in the treatment of behavioral disorders associated with autism

in children and adolescents. Neuropsychiatr Dis Treat. 2008; 4:723–730. [PubMed: 19043516]

CDC. Prevalence of autism spectrum disorders - autism and developmental disabilities monitoring

network, 14 sites, United States, 2008. MMWR Surveill Summ. 2012; 61:1–19.

Collins SM, Surette M, Bercik P. The interplay between the intestinal microbiota and the brain. Nat

Rev Microbiol. 2012; 10:735–742. [PubMed: 23000955]

Comi AM, Zimmerman AW, Frye VH, Law PA, Peeden JN. Familial clustering of autoimmune

disorders and evaluation of medical risk factors in autism. Journal of child neurology. 1999;

14:388–394. [PubMed: 10385847]

Coury DL, Ashwood P, Fasano A, Fuchs G, Geraghty M, Kaul A, Mawe G, Patterson P, Jones NE.

Gastrointestinal conditions in children with autism spectrum disorder: developing a research

agenda. Pediatrics. 2012; 130(Suppl 2):S160–168. [PubMed: 23118247]

Croen LA, Braunschweig D, Haapanen L, Yoshida CK, Fireman B, Grether JK, Kharrazi M, Hansen

RL, Ashwood P, Van de Water J. Maternal mid-pregnancy autoantibodies to fetal brain protein:

the early markers for autism study. Biol Psychiatry. 2008; 64:583–588. [PubMed: 18571628]

Cryan JF, Dinan TG. Mind-altering microorganisms: the impact of the gut microbiota on brain and

behaviour. Nat Rev Neurosci. 2012; 13:701–712. [PubMed: 22968153]

D’Eufemia P, Celli M, Finocchiaro R, Pacifico L, Viozzi L, Zaccagnini M, Cardi E, Giardini O.

Abnormal intestinal permeability in children with autism. Acta Paediatr. 1996; 85:1076–1079.

[PubMed: 8888921]

de Magistris L, Familiari V, Pascotto A, Sapone A, Frolli A, Iardino P, Carteni M, De Rosa M,

Francavilla R, Riegler G, et al. Alterations of the intestinal barrier in patients with autism spectrum

disorders and in their first-degree relatives. J Pediatr Gastroenterol Nutr. 2010; 51:418–424.

[PubMed: 20683204]

Desbonnet L, Clarke G, Shanahan F, Dinan TG, Cryan JF. Microbiota is essential for social

development in the mouse. Molecular psychiatry. 2013

Hsiao et al.

Page 12

Cell

. Author manuscript; available in PMC 2014 December 19.

NIH-PA Author Manuscript