Multi-omic analysis along the gut-brain axis points to a functional architecture of autism

James T. Morton

1,2

, Dong-Min Jin

, Robert H. Mills

, Yan Shao

, Gibraan Rahman

6, 7

, Daniel

McDonald

, Kirsten Berding

, Brittany D. Needham

, Mar ́ıa Fernanda Zurita

, Maude David

, Olga V.

Averina

, Alexey S. Kovtun

12, 13

, Antonio Noto

, Michele Mussap

, Mingbang Wang

, Daniel N.

Frank

, Ellen Li

, Wenhao Zhou

, Vassilios Fanos

, Valery N. Danilenko

, Dennis P. Wall

, Pa ́ul

C ́ardenas

, Manuel E. Balde ́on

, Ramnik J. Xavier

23, 24, 25

, Sarkis K. Mazmanian

, Rob Knight

7, 26, 27

Jack A. Gilbert

7, 28

, Sharon M. Donovan

, Trevor D. Lawley

, Bob Carpenter

, Richard Bonneau

1, 3, 29

and Gaspar Taroncher-Oldenburg

Center for Computational Biology, Flatiron Institute, Simons Foundation, New York, NY, USA

The Simons Foundation Autism Research Initiative, Simons Foundation, New York, NY, USA

Center for Genomics and Systems Biology, Department of Biology, New York University, New York, NY,

USA

Precidiag Inc, Watertown, MA, USA

Host-Microbiota Interactions Laboratory, Wellcome Sanger Institute, Hinxton, UK

Bioinformatics and Systems Biology Program, University of California San Diego, San Diego, CA, USA

Department of Pediatrics, School of Medicine, University of California San Diego, San Diego, CA, USA

Division of Nutritional Sciences, University of Illinois, Urbana, IL, USA

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

Microbiology Institute and Health Science College, Universidad San Francisco de Quito, Quito, Ecuador

Departments of Microbiology & Pharmaceutical Sciences, Oregon State University, Corvallis, OR, USA

Vavilov Institute of General Genetics Russian Academy of Sciences, Moscow, Russia

Skolkovo Institute of Science and Technology, Skolkovo, Russia

Department of Biomedical Sciences, School of Medicine, University of Cagliari, Cagliari, Italy

Laboratory Medicine, Department of Surgical Sciences, School of Medicine, University of Cagliari, Italy

Children’s Hospital of Fudan University, National Center for Children’s Health, Shanghai, China

Department of Medicine, University of Colorado Anschutz Medical Campus, Aurora, CO, USA

Department of Medicine, Division of Gastroenterology and Hepatology, Stony Brook University, Stony

Brook, NY, USA

Neonatal Intensive Care Unit and Neonatal Pathology, Department of Surgical Sciences, School of

Medicine, University of Cagliari, Italy

Pediatrics (Systems Medicine), Biomedical Data Science, and Psychiatry and Behavioral Sciences,

Stanford University, Stanford, CA, USA

Institute of Microbiology, COCIBA, Universidad San Francisco de Quito, Quito, Ecuador

Facultad de Ciencias M ́edicas, de la Salud y la Vida, Universidad Internacional del Ecuador, Quito,

Ecuador

Broad Institute of MIT and Harvard, Cambridge, MA, USA

Department of Molecular Biology, Massachusetts General Hospital, Boston, MA, USA

Center for the Study of Inflammatory Bowel Disease, Massachusetts General Hospital, Boston, MA, USA

Department of Computer Science and Engineering, University of California, San Diego, La Jolla,

California, USA

Department of Bioengineering, University of California San Diego, La Jolla, California, USA

Scripps Institution of Oceanography, UC San Diego, La Jolla, CA, USA

Division of Nutritional Sciences, University of Illinois, Urbana, IL, USA

Prescient Design, a Genentech Accelerator, New York, NY, USA

Author Information : These authors contributed equally : James T. Morton, Gaspar Taroncher-Oldenburg

Corresponding authors : Correspondence to James T. Morton or Gaspar Taroncher-Oldenburg

Acknowledgements

We would like to thank Allan Packer, Paul Wang, Natalia Volfovsky, Kelsey Martin and John Spiro for their

critical review of the manuscript. We’d also like to add Kevin Liu, Hannah Sherman and Xue-Jun Kong for

insightful discussions. Y.S. and T.D.L. are supported by the Wellcome Trust (WT098051).

Contributions

J.T.M. and G.T.-O. conceived and designed the study, developed the software, analyzed the data, interpreted

the results and wrote the manuscript; R.B. contributed to study design, data analysis, result interpretation

and manuscript editing; R.H.M. contributed to study design, data analysis and manuscript editing; R.J.X.

and S.K.M. contributed to study design and manuscript editing; G.R. and B.C. contributed to software

development and manuscript editing; D.-M.J. and Y.S. contributed to data analysis and manuscript editing;

K.B., B.D.N., M.F.Z., M.D., O.V.A., A.S.K., A.N., M.M., M.W., D.N.F., E.L., W.Z., V.F., V.N.D., D.P.W.,

M.E.B., R.K., J.G., S.M.D. and T.D.L. provided access to data and contributed to manuscript editing.

Conflict of Interest

R.H.M. is Scientific Director at Precidiag Inc.; T.D.L. is co-founder and Chief Scientific Officer of Microbi-

1

otica; S.K.M. is a co-founder and has equity in Axial Therapeutics; R.B is currently Executive Director of

2

Prescient Design, a Genentech Accelerator; G.T.-O. is a Consultant-in-Residence at the Simons Foundation.

3

CC-BY-NC-ND 4.0 International license

available under a

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made

The copyright holder for this preprint

this version posted March 9, 2022.

;

https://doi.org/10.1101/2022.02.25.482050

doi:

bioRxiv preprint

Abstract

4

Autism is a highly heritable neurodevelopmental disorder characterized by heterogeneous cognitive, behav-

5

ioral and communication impairments. Disruption of the gut-brain axis (GBA) has been implicated in

6

autism, with dozens of cross-sectional microbiome and other omic studies revealing autism-specific profiles

7

along the GBA albeit with little agreement in composition or magnitude. To explore the functional architec-

8

ture of autism, we developed an age and sex-matched Bayesian differential ranking algorithm that identified

9

autism-specific profiles across 10 cross-sectional microbiome datasets and 15 other omic datasets, including

10

dietary patterns, metabolomics, cytokine profiles, and human brain expression profiles. The analysis uncov-

11

ered a highly significant, functional architecture along the GBA that encapsulated the overall heterogeneity

12

of autism phenotypes. This architecture was determined by autism-specific amino acid, carbohydrate and

13

lipid metabolism profiles predominantly encoded by microbial species in the genera

Prevotella

Enterococ-

14

cus

Bifidobacterium

, and

Desulfovibrio

, and was mirrored in brain-associated gene expression profiles and

15

restrictive dietary patterns in individuals with autism. Pro-inflammatory cytokine profiling and virome asso-

16

ciation analysis further supported the existence of an autism-specific architecture associated with particular

17

microbial genera. Re-analysis of a longitudinal intervention study in autism recapitulated the cross-sectional

18

profiles, and showed a strong association between temporal changes in microbiome composition and autism

19

symptoms. Further elucidation of the functional architecture of autism, including of the role the microbiome

20

plays in it, will require deep, multi-omic longitudinal intervention studies on well-defined stratified cohorts

21

to support causal and mechanistic inference.

22

Introduction

23

Autism spectrum disorder (ASD) encompasses a broad range of neurodevelopmental conditions defined by

24

heterogeneous cognitive, behavioral and communication impairments that manifest early in childhood [1]. To

25

date, over a hundred genes have been identified as putatively associated with ASD, with some genotypes now

26

having a standardized clinical diagnosis [2]. However, most of the genetic variants are still associated with

27

heterogeneous phenotypes, making it difficult to identify molecular mechanisms that might be responsible for

28

particular impairments [3]. Some studies have also looked at the presence of abnormalities in different brain

29

regions in children with ASD [4, 5]. However, whether such neuroanatomical features could mechanistically

30

determine autism, and whether environmental factors could induce analogous ASD-like symptoms, remains

31

unresolved [1].

32

In addition to risk factors, one comorbidity that has been linked to ASD with high confidence is the

33

occurrence of gastrointestinal (GI) symptoms, such as constipation, diarrhea, or abdominal bloating, but

34

causal insights remain elusive [6, 7, 8]. Mechanistically, much research has been focused on the interplay

35

between the GI system and processes controlled by the neuroendocrine, neuroimmune, and autonomous

36

nervous systems, all of which converge around the GI tract and together modulate the gut-brain axis (GBA)

37

[9, 10, 11].

38

The GBA facilitates bidirectional communication between the gut and the brain, contributing to brain

39

homeostasis and helping regulate cognitive and emotional functions [9, 12]. Over the past decade, research

40

on the factors modulating the GBA has revealed the central role played by the gut microbiome—the trillions

41

of microbes that colonize the gut—in regulating neuroimmune networks, modifying neural networks, and

42

directly communicating with the brain [13]. Dysregulation of the gut microbiome and the ensuing disruption

43

of the GBA are thought to contribute to the pathogenesis of neurodevelopmental disorders including autism,

44

but the underlying mechanisms and the extent to which the microbiome explains these dynamics is still

45

unknown [14, 15, 16, 17].

46

Several dozen autism gut metagenomics studies have revealed many, albeit inconsistent, variations in

47

microbial diversity in individuals with ASD compared with neurotypical individuals [18, 19, 17]. Similarly,

48

metagenome-based functional reconstructions and metabolic analyses have also shown strong, albeit incon-

49

clusive differences between ASD and neurotypical individuals [20, 21, 22]. Comparative analyses at other

50

omic levels have further shown little agreement across studies [23] raising the question of whether the re-

51

sults obtained so far reflect intrinsic biological differences among cohorts, insufficient statistical power, or

52

experimental biases that preclude meaningful comparisons [24].

53

A wide range of factors could explain the disagreement across studies, including confounding variation

54

due to batch effects, the application of inappropriate statistical methodologies, and the vast phenotypic

55

and genotypic heterogeneity of ASD. Batch effects can be caused by many factors including misspecified

56

experimental designs, technical variability, geographical location, and demographic composition, and several

57

algorithms have been proposed to correct for them, but a lack of standardized statistical methods further

58

complicates interpretation [25, 26, 27, 28]. Microbiome datasets, like other omic datasets, are compositional,

59

and failure to account for the compositional nature of sequencing counts can lead to high false positive and

60

false negative rates when identifying differentially abundant microbes [29, 30, 31]. Microbiome analysis in

61

ASD is further confounded by the phenotypic and genotypic heterogeneity of the disorder, which is known to

62

be critical for stratifying ASD subtypes and constructing reliable diagnostics but is typically not measured

63

or controlled for [32, 33, 1].

64

Understanding functional architecture—the network of interactions among different omic levels that

65

determines individual phenotypes—of complex neurodevelopmental disorders such as autism, requires an

66

accurate and comprehensive characterization of the different omic levels contributing to it [34]. Traditionally

67

focused on the human genomic, metabolic, and cellular components of phenotype determination, mounting

68

evidence of the role the GBA plays in phenotype determination through bidirectional modulatory mechanisms

69

raises the need for considering the metagenomic and metabolic contributions of the microbiome as potential

70

key components of the functional architecture of autism [35, 36].

71

CC-BY-NC-ND 4.0 International license

available under a

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made

The copyright holder for this preprint

this version posted March 9, 2022.

;

https://doi.org/10.1101/2022.02.25.482050

doi:

bioRxiv preprint

To identify autism-specific omic profiles while reducing cohort-specific confounding factors, we have

72

devised a Bayesian differential ranking algorithm to estimate a distribution of microbial differentials, or

73

relative log-fold changes, [31] across multiple potential ASD subtypes implicit in 25 omic datasets (Table 1).

74

A key feature of this approach was to match individual study participants by sex and age within each study

75

to adjust for confounders in childhood development and cohort-specific batch effects. The preponderance of

76

autism among males is well documented and several potentially sex-dependent mechanisms to explain this

77

phenonmenon have been proposed [37]. Furthermore, the development of the microbiome during childhood

78

is a hallmark of microbiome dynamics in the human gut [38, 39, 40, 41, 42]. Our analysis provides insights

79

into the complexity of the interplay among multiple omic levels in ASD, highlights the inherent limitations

80

of cross-sectional studies for understanding the functional architecture of autism, and provides a framework

81

for further studies aimed at better defining the causal relationship between the microbiome and other omic

82

levels and ASD.

83

Results

84

The structure of our analysis consisted of a multi-cohort and multi-omic meta-analysis framework that al-

85

lowed us to combine independent and dependent omic data sets in one integrated analysis [43, 44]. To

86

minimize issues of compositionality and sequencing depth [45], we modeled overdispersion using a negative

87

binomial distribution for modeling sequencing count data [46] (Box 1 “TACKLING METAGENOMIC UN-

88

KNOWNS”). Our differential ranking approach incorporated a case-control matching component consisting

89

of individually pairing ASD children with age- and sex-matched neurotypical control children within each

90

study cohort, allowing us to adjust for confounding variation and batch effects (Supplemental methods). Fi-

91

nally, we cross-referenced the 16S–based microbial differential ranking analysis from eight age-sex matched

92

cohorts against 15 other omic datasets to contextualize the potential functional roles these microbes could

93

play in autism (Figure 1).

94

Age- and sex-matching increases informational content of cross-sectional ASD

95

datasets

96

97

We compared the age- and sex-matched differential ranking analysis to the standard group-averaged dif-

98

ferential ranking analysis across eight out of the ten 16S studies [47, 48, 49, 21, 50, 51, 52, 53]. Age- and

99

sex-matched differential analysis outperformed standard group averaging with respect to

, and its overall

100

performance strictly improved as more studies were added (Figure S1). This performance boost reflected a

101

reduction in model uncertainty with larger cohorts that was indicative of overlapping differentially abundant

102

taxa across studies and of reduced confounding variation.

103

Global differential ranking analysis reveals a distribution of significant ASD-

104

microbiome associations

105

106

A global, age- and sex-matched differential ranking analysis of the eight 16S datasets selected for this study

107

revealed a clear partitioning of microbial differences with respect to ASD and cohort membership (Figure

108

2a, Figure S2). The distribution of the overall case-control differences showed a strong ASD-specific signal

109

driven by 142 microbes more commonly found in ASD children and 32 microbes more commonly found in

110

their control counterparts (Table S1). The variability observed is most likely due to confounding factors such

111

as cohort demographics and geographic location, with the eight cohorts originating from Asia, Europe, South

112

America, and North America. Analogous global differential ranking trends could be observed for the virome,

113

shotgun metagenomics sequencing (SMS), and RNA-seq datasets (Figure S3). To determine whether these

114

highly significant microbiome signals (pvalue

0.0025) could be used to distinguish ASD subjects from their

115

age- and sex-matched control counterparts, we trained random forest classifiers on train/validation/test splits

116

on data derived from 16S—targeted sequencing of the microbial 16S ribosomal RNA gene—and SMS—whole

117

genome sequencing of microbial communities. Despite the strong microbiome effect size, we faced difficulties

118

fitting generalizable classifiers. Our best classifiers had an average cross-validation accuracy of about 75%

119

(Figure 2b), falling within the range of 52%–90% classification accuracy observed in previous studies [50,

120

49, 54]. We suspect that the vast heterogeneity across cohorts hampered classification performance. While

121

cohort size did not impact predictiability (Figure 2c), some cohorts with skewed sex ratios or age ranges did

122

exhibit lower classification performance (Figure. 2d-e). In the Zurita et al. cohort, sex-specific factors could

123

confound classification (four girls and 56 boys) [48], and in the Kang et al. cohort, age-associated microbiome

124

development factors could hamper classification accuracy (all children were 10 years or older). In addition, all

125

subjects in the Kang et al. cohort had known GI symptoms [55], further compromising classifier performance

126

because none of the other studies controlled for this variable. As a result, and analogous to the phenotypic

127

and genotypic heterogeneity observed in ASD, the microbiome composition of ASD children also exhibits

128

high heterogeneity, precluding the identification of a homogeneous universal ASD microbial profile and the

129

construction of generalizable classifiers.

130

Children with ASD exhibit significant individual differences at several omic levels

131

132

Differential ranking analysis of three core omic levels—microbiome (16S and SMS) and human transcriptome

133

(RNAseq)—revealed strong and highly significant differences between ASD subjects and their neurotypical

134

CC-BY-NC-ND 4.0 International license

available under a

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made

The copyright holder for this preprint

this version posted March 9, 2022.

;

https://doi.org/10.1101/2022.02.25.482050

doi:

bioRxiv preprint

counterparts (p-value

0.0025) (Figure 2f; Table S2; Table S3). Two additional omic levels—the metabolome

135

and the virome— didn’t show significance signals (Figure S4, Table S4). Amongst the models that yielded a

136

statistically significant signal, the 16S and SMS datasets had a larger effect size than the RNAseq datasets

137

(Figure 2f). While each omic dataset by itself showed strong associations with ASD, a side-by-side comparison

138

of the 16S and SMS datasets—two datasets that should show high equivalence—revealed a significant lack of

139

overlap between them, highlighting the outstanding challenge of batch effects in microbiome studies (Figure

140

S5). The reasons for this discrepancy could be many, but most likely center around sample size—our

141

study looked at eight 16S datasets versus only three SMS datasets containing 754 individuals and 166

142

individuals, respectively. Another major challenge when estimating species profiles with metagenomics

143

reference libraries is assigning a species identification to a read—there are many reads that do not uniquely

144

map to individual species—and as a result, these multi-mapped reads can give rise to numerous false positive

145

taxa [56, 57, 58, 59]. Based on this, we decided to focus primarily on the 16S datasets to define a global

146

differential ranking profile.

147

Sibling-matching and unrelated sex- and age-matching show significant discrep-

148

ancies

149

150

To determine whether sex- and age-matched differentials could be universally predictive, we compared the

151

16S differentials obtained from the age- and sex-matched cohorts with two sibling-matched cohorts [60] [61].

152

Interestingly, we observed a significant negative correlation between the differentials extracted from the age-

153

and sex-matched cohort and the two sibling-matched cohorts, suggesting that ASD-specific microbes in the

154

sibling-matched studies are enriched in the control group in the age- and sex-matched studies and vice versa

155

(Fig. 2g). Permanova applied to the age-sex matched cohort revealed a strong age-confounder across cohorts

156

(pvalue

0.001), with little confounding variation due to sex (Table S5). In contrast, Permanova applied

157

to these sibling-matched cohorts revealed that household is a major confounder in both cohorts (pvalue

158

0.001), but ruled out age as a confounder and indicated that sex was a confounder only in the David et al.

159

cohort (pvalue

0.001). The observed discrepancies point to different sets of confounders possibly affecting

160

the analysis: in the case of the age- and sex-matched studies, family confounders aren’t typically accounted

161

for, while sibling-matched studies don’t typically adjust for age confounders. In addition, and while cohorts

162

such as the one studied in Maude et al. specifically control for the possibility, siblings often exhibit a higher

163

risk of developing ASD compared to the general population [62].

164

Host cytokine concentrations are correlated with microbial abundances

165

Immune dysregulation, ranging from circulating ‘anti-brain’ antibodies and perturbed cytokine profiles to

166

simply having a family history of immune disorders, has been repeatedly associated with ASD [63, 64].

167

Recently, for example, Zurita et al. showed that concentrations of the inflammatory cytokine transforming

168

growth factor beta (TGF-

) are significantly elevated in ASD children. We reanalyzed this dataset, after age-

169

and sex-matching, and observed that microbial differentials associated with TGF-

and IL-6 concentrations

170

were positively correlated with the global microbial log-fold changes between ASD and control pairs (IL-6 :

171

r=-0.435, p=0; TGF-

: r=0.291, p=0) (Table S6). To validate the integrity of these microbial profiles with

172

respect to the cytokine changes, we calculated the log-ratios of these microbial abundances and showed them

173

to, in turn, be highly correlated with TGF-

and IL-6 concentrations (IL-6 : r=0.50, p=0.0007; TGF-

174

r=0.45, p=0.002) (Figure 3 a-d).

175

Four clusters of microbial genera—

Prevotella

Enterococcus

Bifidobacteria

, and

Desulfovibrio

—were pre-

176

dominantly associated with the cytokine differentials. Partial mechanistic insights on some of these cytokine-

177

microbe associations have been previously published. Both

B. longum

and

E. faecalis

have shown anti-

178

inflammatory activities:

B. longum

downregulates IL-6 in fetal human enterocytes in vitro [65] and

179

faecalis

has been observed to upregulate TGF-

in human intestinal cells [66].

P. copri

associations with

180

different cytokines have also been observed in multiple disease contexts [67]. Similarly,

Bifidobacteria

and

181

Prevotella

both co-occurred with phages enriched in ASD or in neurotypical children (Figure S6, Table

182

S7), but while microbes have previously been reported to mediate viral infections [68, 69], the mechanistic

183

underpinnings of these interactions with the host’s immunity remain poorly understood [70, 71, 72].

184

The microbiome metabolic capacity is reflective of the human brain-associated

185

metabolic capacity in ASD

186

187

To determine potential crosstalk between the human brain and the microbiome physiology, we compared

188

the metabolic capacities encoded by the microbial metagenome—combining the individual metabolic capac-

189

ities of thousands of different microbes—and the differentially expressed human genome in the brain, two

190

omic levels representing entirely different biological contexts. We observed that over 100 human metabolic

191

pathways differentially expressed in the brain tissues of ASD individuals had analogous microbial path-

192

ways differentially abundant in the microbiome of children with ASD, suggesting a potential coordination

193

of metabolic pathways across omic levels in ASD (Fig. 3e). Pathways related to amino acid metabolism,

194

carbohydrate metabolism and lipid metabolism were disproportionately represented among the overlapping

195

genes (Table S8).

196

CC-BY-NC-ND 4.0 International license

available under a

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made

The copyright holder for this preprint

this version posted March 9, 2022.

;

https://doi.org/10.1101/2022.02.25.482050

doi:

bioRxiv preprint

The microbiome metabolic capacity reflects restrictive diet patterns in children

197

with ASD

198

Autistic traits in early childhood have been shown to correlate with poor diet quality later in life, however,

199

little is known about how diet quality is directly linked to autistic traits [73]. Here, we re-analyzed the paired

200

microbiome and dietary survey data from Berding et al.. A microbiome-diet co-occurrence analysis revealed

201

startlingly similar amino acid, carbohydrate and lipid metabolism association patterns to those observed

202

in the microbiome-brain metabolic capacity analysis (Table S9). Interestingly, both microbes enriched in

203

ASD and in control subjects co-occurred primarily with amino acid dietary compounds (Figure 3f, Figure

204

S7). Autistic children were less likely to consume foods high in glutamic acid, serine, choline, phenylalanine,

205

leucine, tyrosine, valine and histidine, all compounds involved in neurotransmitter biosynthesis. Even though

206

the metabolomic analysis did not yield statistically significant signals after FDR correction, the metabolites

207

that showed the strongest signal included glutamate and phenylalanine, consistent with the microbiome-

208

diet analysis [74, 75, 76]. Disruptions in the biosynthesis of these neurotransmitter molecules have been

209

implicated in a wide variety of psychiatric disorders, and a recent blood metabolomics study has shown the

210

potential of using branched chain amino acids to define autism subtypes [33]. Due to the incompatibility

211

between the molecular features across datasets, it was not possible to combine any of the metabolomics

212

datasets to boost the statistical power, which remains a major limitation of metabolomics technologies at

213

present (see Methods).

214

Differential microbial rankings show disease-specific correlations

215

One major challenge in determining microbiome-disease associations is identifying correlations specific to a

216

particular condition and not generally present across diseases [77, 78]. To determine how specific to ASD our

217

global differential microbiome profile was, we cross-referenced it against differential ranking results obtained

218

from an Inflammatory Bowel Disorder (IBD) dataset [79] and a Type 1 Diabetes (T1D) dataset [80]. IBD

219

shares some comorbidities with ASD [81, 82], while no direct correlation between ASD and T1D has been

220

reported to date. The analysis revealed a notable overlap between microbes enriched in ASD and IBD, and

221

this overlap was stronger than both the overlap between IBD and T1D and between ASD and T1D (Figure

222

S8). Whether this ASD-IBD overlap is suggestive of a common microbial profile or is confounded by the

223

restrictive dietary nature of these two clinical conditions is currently unclear. Higher resolution and properly

224

designed clinical studies will have to be performed to get to a mechanistic understanding of the potential

225

microbiome connection between these two conditions.

226

ASD microbiome profiles weaken after fecal matter transplant consistent with

227

reported behavioral improvement

228

While the preceding cross-sectional analyses showed significant associations among several omic levels (vi-

229

rome, microbiome, immunome) or diet and ASD, insights into causality are still limited. By contrast,

230

longitudinal intervention studies provide an opportunity to obtain stronger insights into causality. To test

231

this, we re-analyzed data from a two-year, open label fecal matter transplant (FMT) study with 18 children

232

with ASD [83]. In this study, the children were subjected to a two-week antibiotic treatment and a bowel

233

cleanse followed by two days of high dose FMT treatment and eight weeks of daily maintenance FMT doses.

234

Based on one of the most common evaluation scales for ASD, the Childhood Autism Rating Scale (CARS),

235

significant improvements were achieved after the ten week course of treatment. Two months later the initial

236

gains were largely maintained, and a two-year follow-up showed signs of further improvement in most of the

237

patients. The results are consistent with a potential role of the microbiome in improving autism symptoms,

238

but how the underlying changes in microbiome composition related to those seen in other studies remained

239

unknown.

240

Here, we re-analyzed the original raw data in the context of the ASD profiles revealed by our cross-

241

sectional differential ranking analysis (Table S10). All microbes associated with ASD in the 18 children prior

242

to the FMT treatment had been identified as ASD-associated microbes in our age- and sex-matched cross-

243

sectional analysis, recapitulating 74% of the cross-sectional profile. Immediately following FMT treatment,

244

the abundances of the ASD-associated microbes decreased in all 8 children (Figure 4). The two-year follow-

245

up analysis revealed that all the ASD-associated microbes, mostly

Enterococcaceae

, continued to be depleted.

246

Consistent with the findings of Kang et al., we also observed

Desulfovibrio sp.

, and

P. copri

increase over

247

the two year period, while

Bifidobacteria sp.

could be found both among depleted and enriched species and

248

other

Prevotella sp.

were depleted, pointing to a potentially wide functional diversity within these genera

249

not noted in the original study.

250

Discussion

251

The functional architecture of ASD, and in particular the potential role the microbiome plays in modulating

252

the GBA in the context of autism, remains poorly understood due to disagreements among existing micro-

253

biome and other omic studies. Our Bayesian model highlighted a distribution of highly significant microbial

254

differentials obtained from individual age- and sex-pairings between children with ASD and neurotypicals,

255

and parallel analyses at the immunome, human transcriptome, and dietome levels revealed strong associ-

256

ations among omic levels. The virome and the direct metabolome signals, while present, were markedly

257

weaker than the other omic signals. The inferred ASD-specific metabolic profiles from the microbiome and

258

the human transcriptome, on the other hand, showed a high and significant degree of overlap in microbial

259

and human pathways expressed in the gut and in the brain, respectively. The metabolic connection implied

260

by this overlap, which included differentially enriched carbohydrate and amino acid metabolic pathways in

261

CC-BY-NC-ND 4.0 International license

available under a

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made

The copyright holder for this preprint

this version posted March 9, 2022.

;

https://doi.org/10.1101/2022.02.25.482050

doi:

bioRxiv preprint

ASD, is a remarkable observation given the fundamental difference between the gut and brain physiologies,

262

which would a priori suggest a reduced overlap in metabolic capacities. The microbiome-diet co-occurrence

263

analysis also highlighted a reduced intake of amino acids and carbohydrates linked to specific microbiome

264

profiles in ASD children. These metabolic and dietary imbalances, particularly regarding glutamate levels,

265

were further apparent, albeit weakly, in the serum, fecal and urine metabolomes we analyzed. This multi-

266

scale overlap we observed along the GBA points to the existence of a functional architecture of ASD driven

267

by the metabolic potential at the genomic and metagenomic levels.

268

While the differential distributions we determined were highly significant, the global analysis did not

269

provide reliable ASD classifiers or uncover universal microbial ‘smoking guns’ linked to autism. However,

270

several microorganisms consistently detected across omic levels pointed to potentially interesting functional

271

connections. For example, our analysis suggested that

B. longum

exhibited a down-regulation of IL-6, which

272

has been observed across a number of in vitro and cohort studies[84, 85]. The diet co-occurrence analysis also

273

showed a strong association between

P. copri

and carbohydrate depletion in ASD. The population dynamics

274

P. copri

have been reported to be driven primarily by carbohydrates in the diet [67]. Multiple other

275

microbes, including several

Bifidobacteria

Enterococcus

and

Desulfovibrio

species, stood out in the immune

276

and viral analyses. In the FMT study, the relative proportions of several

Prevotella

Bifidobacteria

Desul-

277

fovibrio

and

Enterococcus

species also showed strong associations with ASD symptoms, further suggesting

278

a causal role for these microorganisms in shaping autism symptoms.

279

Despite our inability to determine actual metabolomic profiles at this point (see Methods), our metabolite

280

analysis based on microbiome- and brain-derived metabolite inferences as well as the diet-derived metabolite

281

data reveals a picture of a unifying and distinct ASD functional architecture. With the brain, the immunome

282

and diet as major effectors, the multi-factorial complexity of ASD is reduced to a multi-scale set of inter-

283

actions centered around human and bacterial metabolism that in turn determines phenotypic, genomic and

284

metagenomic attributes via multiple feedback loops (Figure 5). The association of specific genotypes with

285

ASD has been clearly established [2]; the pivotal role of the immune system in mediating the communication

286

between the gut microbiome and the human brain as well as other peripheral systems is also firmly estab-

287

lished [86]; further, the central role of the microbiome in mediating diet-derived nutrient mobilization has

288

been extensively documented [87]; and several hard-wired feedback loops among these effectors such as the

289

hypothalamus-mediated regulation of appetite and diet, have also been described [10].

290

A major limitation of our meta-analysis is the lack of consistent behavioral, genotypic or electronic

291

healthcare record data that would have allowed subtyping of the ASD subjects in light of environmental

292

confounders. Furthermore, we cannot definitely recommend age- and sex-matching over sibling matching

293

based on our analysis. Age remains a major confounding factor in early childhood microbiome development

294

and controlling for this is key for understanding microbial fluctuations [88]. On the other hand, sibling-

295

matching may help control environmental factors, but mostly rules out the ability to age-match subjects,

296

thus potentially introducing the age confounder [89]. And while our approach revealed strong associations

297

among the microbiome, other omic levels, and ASD, the vast heterogeneity in behavioral patterns and in

298

genotypes is a major obstacle in constructing diagnostics and treatments for ASD symptoms [90, 32].

299

Our analysis has further exposed the limitations of cross-sectional cohort studies and the need for lon-

300

gitudinal intervention studies to further our understanding of the functional architecture of ASD. Building

301

realistic causal models of autism needs to take into account the multi-factorial complexity underlying differ-

302

ent ASD subtypes, which will require a concerted effort to simultaneously analyze several omic levels and at

303

clinically relevant time scales. For instance, understanding the engraftment dynamics of FMT and its func-

304

tional implications on the recipients’ gut microbiomes requires frequent initial sampling of the microbiome,

305

immunome and metabolome, but tracing any behavioral changes over time requires less frequent sampling

306

over periods of up to several years in combination with reliable behavioral, medical and dietary surveys

307

[91, 92]. Collecting and integrating such multi-scale omic datasets presents unique logistical and analytical

308

challenges.

309

Managing data acquisition and access will require coordinating multiple sites and potentially centralizing

310

some aspects of sample processing. Recent initiatives such as The Environmental Determinants of Diabetes in

311

the Young (TEDDY) study, an international long-term, multi-center initiative to link specific environmental

312

triggers to particular Type 1 diabetes–associated genotypes, provide a blueprint for similar approaches in

313

ASD [93]. A key component of such an initiative would be the establishment of standardized sampling

314

and processing protocols that would minimize technical confounders, one of the top confounders at most

315

omic levels. For instance, our analysis showed major batch effects when comparing 16S and SMS datasets

316

across cohorts (r=-0.023, p=0.48, Figure S5b) as well as within a cohort (r=0.17, p=1e-5, Figure S5b). And

317

while there are extensive efforts underway to calibrate microbiome datasets [94], other omic levels such as

318

the metabolome [26] present even more fundamental technical issues that make it imperative to develop

319

concerted strategies to be able to include them in an integrated analysis.

320

In addition to the considerable variations in statistical properties across datasets, interactions among

321

omic levels are mostly underdetermined, making the construction of informative models a major challenge.

322

Determining the necessary biologically relevant and unbiased assumptions is a non-trivial process and can

323

inadvertently lead to model mis-specifications resulting in misleading conclusions. As pointed out recently

324

for genetic, environmental and microbiome models in ASD [95, 96], addressing these issues will be critical

325

to inferring causal mechanisms from population-scale studies. In addition, and given the vast heterogeneity

326

of ASD, designing cohort studies that minimize confounding factor effects will be key to furthering our un-

327

derstanding of autism. For example, while our analysis could not identify ASD subtypes implicated in GI

328

symptoms, we have determined stronger associations between gut microbes, host immunity, brain expression

329

and dietary patterns than previously reported, highlighting the potential for boosting the statistical power

330

and biological insight with comprehensive omic analyses. We conclude that multi-omic longitudinal inter-

331

vention studies on appropriately stratified cohorts, in combination with comprehensive patient metadata,

332

provide an optimal approach to advance our understanding of the etiology of autism to the next level.

333

CC-BY-NC-ND 4.0 International license

available under a

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made

The copyright holder for this preprint

this version posted March 9, 2022.

;

https://doi.org/10.1101/2022.02.25.482050

doi:

bioRxiv preprint

References

334

[1] Catherine Lord, Traolach S Brugha, Tony Charman, James Cusack, Guillaume Dumas, Thomas Fra-

335

zier, Emily J H Jones, Rebecca M Jones, Andrew Pickles, Matthew W State, Julie Lounds Taylor,

336

and Jeremy Veenstra-VanderWeele. Autism spectrum disorder.

Nat Rev Dis Primers

, 6(1):5, January

337

2020.

338

[2] F Kyle Satterstrom, Jack A Kosmicki, Jiebiao Wang, Michael S Breen, Silvia De Rubeis, Joon-Yong

339

An, Minshi Peng, Ryan Collins, Jakob Grove, Lambertus Klei, Christine Stevens, Jennifer Reichert,

340

Maureen S Mulhern, Mykyta Artomov, Sherif Gerges, Brooke Sheppard, Xinyi Xu, Aparna Bhaduri,

341

Utku Norman, Harrison Brand, Grace Schwartz, Rachel Nguyen, Elizabeth E Guerrero, Caroline

342

Dias, Autism Sequencing Consortium, iPSYCH-Broad Consortium, Catalina Betancur, Edwin H Cook,

343

Louise Gallagher, Michael Gill, James S Sutcliffe, Audrey Thurm, Michael E Zwick, Anders D Børglum,

344

Matthew W State, A Ercument Cicek, Michael E Talkowski, David J Cutler, Bernie Devlin, Stephan J

345

Sanders, Kathryn Roeder, Mark J Daly, and Joseph D Buxbaum. Large-Scale exome sequencing study

346

implicates both developmental and functional changes in the neurobiology of autism.

Cell

, 180(3):568–

347

584.e23, February 2020.

348

[3] Lilia M Iakoucheva, Alysson R Muotri, and Jonathan Sebat. Getting to the cores of autism.

Cell

349

178(6):1287–1298, September 2019.

350

[4] Cynthia Mills Schumann, Julia Hamstra, Beth L Goodlin-Jones, Linda J Lotspeich, Hower Kwon,

351

Michael H Buonocore, Cathy R Lammers, Allan L Reiss, and David G Amaral. The amygdala is

352

enlarged in children but not adolescents with autism; the hippocampus is enlarged at all ages.

Journal

353

of neuroscience

, 24(28):6392–6401, 2004.

354

[5] Cynthia Mills Schumann and David G Amaral. Stereological analysis of amygdala neuron number in

355

autism.

Journal of Neuroscience

, 26(29):7674–7679, 2006.

356

[6] Radu Lefter, Alin Ciobica, Daniel Timofte, Carol Stanciu, and Anca Trifan. A descriptive review

357

on the prevalence of gastrointestinal disturbances and their multiple associations in autism spectrum

358

disorder.

Medicina

, 56(1):11, December 2019.

359

[7] Amirhossein Modabbernia, Eva Velthorst, and Abraham Reichenberg. Environmental risk factors for

360

autism: an evidence-based review of systematic reviews and meta-analyses.

Mol. Autism

, 8:13, March

361

2017.

362

[8] Elaine Y Hsiao. Gastrointestinal issues in autism spectrum disorder.

Harv. Rev. Psychiatry

, 22(2):104,

363

2014.

364

[9] Emeran A Mayer. Gut feelings: the emerging biology of gut–brain communication.

Nat. Rev. Neurosci.

365

12(8):453–466, July 2011.

366

[10] Livia H Morais, Henry L Schreiber, 4th, and Sarkis K Mazmanian. The gut microbiota-brain axis in

367

behaviour and brain disorders.

Nat. Rev. Microbiol.

, 19(4):241–255, April 2021.

368

[11] John F Cryan, Kenneth J O’Riordan, Kiran Sandhu, Veronica Peterson, and Timothy G Dinan. The

369

gut microbiome in neurological disorders.

Lancet Neurol.

, 19(2):179–194, February 2020.

370

[12] Amar Sarkar, Siobh ́an Harty, Katerina V-A Johnson, Andrew H Moeller, Rachel N Carmody, Soili M

371

Lehto, Susan E Erdman, Robin I M Dunbar, and Philip W J Burnet. The role of the microbiome in

372

the neurobiology of social behaviour.

Biol. Rev. Camb. Philos. Soc.

, 95(5):1131–1166, October 2020.

373

[13] Amanda Jacobson, Daping Yang, Madeleine Vella, and Isaac M Chiu. The intestinal neuro-immune

374

axis: crosstalk between neurons, immune cells, and microbes.

Mucosal Immunol.

, 14(3):555–565,

375

February 2021.

376

[14] Sue Grenham, Gerard Clarke, John F Cryan, and Timothy G Dinan. Brain–Gut–Microbe communi-

377

cation in health and disease.

Front. Physiol.

, 0, 2011.

378

[15] The Microbiome-Gut-Brain axis in health and disease.

Gastroenterol. Clin. North Am.

, 46(1):77–89,

379

March 2017.

380

[16] Yoko M Ambrosini, Dana Borcherding, Anumantha Kanthasamy, Hyun Jung Kim, Auriel A Willette,

381

Albert Jergens, Karin Allenspach, and Jonathan P Mochel. The Gut-Brain axis in neurodegenerative

382

diseases and relevance of the canine model: A review.

Front. Aging Neurosci.

, 0, 2019.

383

[17] Mingyu Xu, Xuefeng Xu, Jijun Li, and Fei Li. Association between gut microbiota and autism spectrum

384

disorder: A systematic review and Meta-Analysis.

Front. Psychiatry

, 10:473, July 2019.

385

[18] Pedro Andreo-Mart ́ınez, Mar ́ıa Rubio-Aparicio, Julio S ́anchez-Meca, Alejandro Veas, and

386

Agust ́ın Ernesto Mart ́ınez-Gonz ́alez. A meta-analysis of gut microbiota in children with autism.

387

Autism Dev. Disord.

, May 2021.

388

[19] Navya Bezawada, Tze Hui Phang, Georgina L Hold, and Richard Hansen. Autism spectrum disorder

389

and the gut microbiota in children: A systematic review.

Ann. Nutr. Metab.

, 76(1):16–29, January

390

2020.

391

CC-BY-NC-ND 4.0 International license

available under a

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made

The copyright holder for this preprint

this version posted March 9, 2022.

;

https://doi.org/10.1101/2022.02.25.482050

doi:

bioRxiv preprint

[20] Narueporn Likhitweerawong, Chanisa Thonusin, Nonglak Boonchooduang, Orawan Louthrenoo,

392

Intawat Nookaew, Nipon Chattipakorn, and Siriporn C Chattipakorn. Profiles of urine and blood

393

metabolomics in autism spectrum disorders.

Metab. Brain Dis.

, August 2021.

394

[21] Jiang Zhu, Xueying Hua, Ting Yang, Min Guo, Qiu Li, Lu Xiao, Ling Li, Jie Chen, and Tingyu Li.

395

Alterations in gut vitamin and amino acid metabolism are associated with symptoms and neurodevel-

396

opment in children with autism spectrum disorder.

J. Autism Dev. Disord.

, July 2021.

397

[22] Xin-Jie Xu, Xiao-E Cai, Fan-Chao Meng, Tian-Jia Song, Xiao-Xi Wang, Yi-Zhen Wei, Fu-Jun Zhai,

398

Bo Long, Jun Wang, Xin You, and Rong Zhang. Comparison of the metabolic profiles in the plasma and

399

urine samples between autistic and typically developing boys: A preliminary study.

Front. Psychiatry

400

12:657105, June 2021.

401

[23] Atiqah Azhari, Farouq Azizan, and Gianluca Esposito. A systematic review of gut-immune-brain

402

mechanisms in autism spectrum disorder.

Dev. Psychobiol.

, 61(5):752–771, July 2019.

403

[24] Patrick D Schloss. Identifying and overcoming threats to reproducibility, replicability, robustness, and

404

generalizability in microbiome research.

MBio

, 9(3), June 2018.

405

[25] Yiwen Wang and Kim-Anh LˆeCao. Managing batch effects in microbiome data.

Brief. Bioinform.

406

21(6):1954–1970, December 2020.

407

[26] Wei Han and Liang Li. Evaluating and minimizing batch effects in metabolomics.

Mass Spectrom.

408

Rev.

, November 2020.

409

[27] Hoa Thi Nhu Tran, Kok Siong Ang, Marion Chevrier, Xiaomeng Zhang, Nicole Yee Shin Lee, Michelle

410

Goh, and Jinmiao Chen. A benchmark of batch-effect correction methods for single-cell RNA sequenc-

411

ing data.

Genome Biol.

, 21(1):12, January 2020.

412

[28] Jelena

Cuklina, Patrick G A Pedrioli, and Ruedi Aebersold. Review of batch effects prevention,

413

diagnostics, and correction approaches.

Methods Mol. Biol.

, 2051:373–387, 2020.

414

[29] Gregory B Gloor, Jia Rong Wu, Vera Pawlowsky-Glahn, and Juan Jos ́e Egozcue. It’s all relative:

415

analyzing microbiome data as compositions.

Ann. Epidemiol.

, 26(5):322–329, May 2016.

416

[30] Doris Vandeputte, Gunter Kathagen, Kevin D’hoe, Sara Vieira-Silva, Mireia Valles-Colomer, Jo ̃ao

417

Sabino, Jun Wang, Raul Y Tito, Lindsey De Commer, Youssef Darzi, S ́everine Vermeire, Gwen Falony,

418

and Jeroen Raes. Quantitative microbiome profiling links gut community variation to microbial load.

419

Nature

, 551(7681):507–511, November 2017.

420

[31] James T Morton, Clarisse Marotz, Alex Washburne, Justin Silverman, Livia S Zaramela, Anna Edlund,

421

Karsten Zengler, and Rob Knight. Establishing microbial composition measurement standards with

422

reference frames.

Nat. Commun.

, 10(1):2719, June 2019.

423

[32] Yuan Luo, Alal Eran, Nathan Palmer, Paul Avillach, Ami Levy-Moonshine, Peter Szolovits, and Isaac S

424

Kohane. A multidimensional precision medicine approach identifies an autism subtype characterized

425

by dyslipidemia.

Nat. Med.

, 26(9):1375–1379, September 2020.

426

[33] Alan M Smith, Joseph J King, Paul R West, Michael A Ludwig, Elizabeth L R Donley, Robert E

427

Burrier, and David G Amaral. Amino acid dysregulation metabotypes: Potential biomarkers for

428

diagnosis and individualized treatment for subtypes of autism spectrum disorder.

Biol. Psychiatry

429

85(4):345–354, February 2019.

430

[34] Patrick F Sullivan and Daniel H Geschwind. Defining the genetic, genomic, cellular, and diagnostic

431

architectures of psychiatric disorders.

Cell

, 177(1):162–183, 2019.

432

[35] Ran Blekhman, Julia K Goodrich, Katherine Huang, Qi Sun, Robert Bukowski, Jordana T Bell,

433

Timothy D Spector, Alon Keinan, Ruth E Ley, Dirk Gevers, et al. Host genetic variation impacts

434

microbiome composition across human body sites.

Genome biology

, 16(1):1–12, 2015.

435

[36] Braden T Tierney, Yingxuan Tan, Aleksandar D Kostic, and Chirag J Patel. Gene-level metagenomic

436

architectures across diseases yield high-resolution microbiome diagnostic indicators.

Nature communi-

437

cations

, 12(1):1–12, 2021.

438

[37] Sarah L Ferri, Ted Abel, and Edward S Brodkin. Sex differences in autism spectrum disorder: a review.

439

Curr. Psychiatry Rep.

, 20(2):9, March 2018.

440

[38] Maria Gloria Dominguez-Bello, Filipa Godoy-Vitorino, Rob Knight, and Martin J Blaser. Role of the

441

microbiome in human development.

Gut

, 68(6):1108–1114, June 2019.

442

[39] Marie-Claire Arrieta, Leah T Stiemsma, Nelly Amenyogbe, Eric M Brown, and Brett Finlay. The

443

intestinal microbiome in early life: health and disease.

Front. Immunol.

, 5:427, September 2014.

444

[40] Gwen Falony, Marie Joossens, Sara Vieira-Silva, Jun Wang, Youssef Darzi, Karoline Faust, Alexan-

445

der Kurilshikov, Marc Jan Bonder, Mireia Valles-Colomer, Doris Vandeputte, Raul Y Tito, Samuel

446

Chaffron, Leen Rymenans, Chlo ̈e Verspecht, Lise De Sutter, Gipsi Lima-Mendez, Kevin D’hoe, Karl

447

Jonckheere, Daniel Homola, Roberto Garcia, Ettje F Tigchelaar, Linda Eeckhaudt, Jingyuan Fu, Lies-

448

bet Henckaerts, Alexandra Zhernakova, Cisca Wijmenga, and Jeroen Raes. Population-level analysis

449

of gut microbiome variation.

Science

, 352(6285):560–564, April 2016.

450

CC-BY-NC-ND 4.0 International license

available under a

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made

The copyright holder for this preprint

this version posted March 9, 2022.

;

https://doi.org/10.1101/2022.02.25.482050

doi:

bioRxiv preprint

[41] Tanya Yatsunenko, Federico E Rey, Mark J Manary, Indi Trehan, Maria Gloria Dominguez-Bello,

451

Monica Contreras, Magda Magris, Glida Hidalgo, Robert N Baldassano, Andrey P Anokhin, Andrew C

452

Heath, Barbara Warner, Jens Reeder, Justin Kuczynski, J Gregory Caporaso, Catherine A Lozupone,

453

Christian Lauber, Jose Carlos Clemente, Dan Knights, Rob Knight, and Jeffrey I Gordon. Human gut

454

microbiome viewed across age and geography.

Nature

, 486(7402):222–227, May 2012.

455

[42] Yan Shao, Samuel C Forster, Evdokia Tsaliki, Kevin Vervier, Angela Strang, Nandi Simpson, Nitin

456

Kumar, Mark D Stares, Alison Rodger, Peter Brocklehurst, Nigel Field, and Trevor D Lawley. Stunted

457

microbiota and opportunistic pathogen colonization in caesarean-section birth.

Nature

, 574(7776):117–

458

121, October 2019.

459

[43] Adib Shafi, Tin Nguyen, Azam Peyvandipour, Hung Nguyen, and Sorin Draghici. A Multi-Cohort

460

and Multi-Omics Meta-Analysis framework to identify Network-Based gene signatures.

Front. Genet.

461

10:159, March 2019.

462

[44] Alexander Kaever, Manuel Landesfeind, Kirstin Feussner, Burkhard Morgenstern, Ivo Feussner, and

463

Peter Meinicke. Meta-analysis of pathway enrichment: combining independent and dependent omics

464

data sets.

PLoS One

, 9(2):e89297, February 2014.

465

[45] Gregory B Gloor, Jean M Macklaim, Vera Pawlowsky-Glahn, and Juan J Egozcue. Microbiome datasets

466

are compositional: And this is not optional.

Front. Microbiol.

, 8:2224, November 2017.

467

[46] Michael I Love, Wolfgang Huber, and Simon Anders. Moderated estimation of fold change and disper-

468

sion for RNA-seq data with DESeq2.

Genome Biol.

, 15(12):550, 2014.

469

[47] Kirsten Berding and Sharon M Donovan. Dietary patterns impact temporal dynamics of fecal micro-

470

biota composition in children with autism spectrum disorder.

Front Nutr

, 6:193, 2019.

471

[48] Mar ́ıa Fernanda Zurita, Pa ́ul A C ́ardenas, Mar ́ıa Elena Sandoval, Mar ́ıa Caridad Pe ̃na, Marco For-

472

nasini, Nancy Flores, Marcia H Monaco, Kirsten Berding, Sharon M Donovan, Thomas Kuntz, Jack A

473

Gilbert, and Manuel E Balde ́on. Analysis of gut microbiome, nutrition and immune status in autism

474

spectrum disorder: a case-control study in ecuador.

Gut Microbes

, 11(3):453–464, May 2020.

475

[49] Zhou Dan, Xuhua Mao, Qisha Liu, Mengchen Guo, Yaoyao Zhuang, Zhi Liu, Kun Chen, Junyu Chen,

476

Rui Xu, Junming Tang, Lianhong Qin, Bing Gu, Kangjian Liu, Chuan Su, Faming Zhang, Yankai Xia,

477

Zhibin Hu, and Xingyin Liu. Altered gut microbial profile is associated with abnormal metabolism

478

activity of autism spectrum disorder.

Gut Microbes

, 11(5):1246–1267, September 2020.

479

[50] Jennifer Fouquier, Nancy Moreno Huizar, Jody Donnelly, Cody Glickman, Dae-Wook Kang, Juan Mal-

480

donado, Rachel A Jones, Kimberly Johnson, James B Adams, Rosa Krajmalnik-Brown, and Catherine

481

Lozupone. The gut microbiome in autism: Study-Site effects and longitudinal analysis of behavior

482

change.

mSystems

, 6(2), April 2021.

483

[51] Rong Zou, Fenfen Xu, Yuezhu Wang, Mengmeng Duan, Min Guo, Qiang Zhang, Hongyang Zhao, and

484

Huajun Zheng. Changes in the gut microbiota of children with autism spectrum disorder.

Autism

485

Res.

, 13(9):1614–1625, September 2020.

486

[52] Dae-Wook Kang, James B Adams, Ann C Gregory, Thomas Borody, Lauren Chittick, Alessio Fasano,

487

Alexander Khoruts, Elizabeth Geis, Juan Maldonado, Sharon McDonough-Means, Elena L Pollard,

488

Simon Roux, Michael J Sadowsky, Karen Schwarzberg Lipson, Matthew B Sullivan, J Gregory Ca-

489

poraso, and Rosa Krajmalnik-Brown. Microbiota transfer therapy alters gut ecosystem and improves

490

gastrointestinal and autism symptoms: an open-label study.

Microbiome

, 5(1):10, January 2017.

491

[53] Prjna687773. https://www.ncbi.nlm.nih.gov/bioproject/PRJNA687773.

492

[54] Chloe X Yap, Anjali K Henders, Gail A Alvares, David L A Wood, Lutz Krause, Gene W Tyson,

493

Restuadi Restuadi, Leanne Wallace, Tiana McLaren, Narelle K Hansell, Dominique Cleary, Rachel

494

Grove, Claire Hafekost, Alexis Harun, Helen Holdsworth, Rachel Jellett, Feroza Khan, Lauren P

495

Lawson, Jodie Leslie, Mira Levis Frenk, Anne Masi, Nisha E Mathew, Melanie Muniandy, Michaela

496

Nothard, Jessica L Miller, Lorelle Nunn, Gerald Holtmann, Lachlan T Strike, Greig I de Zubicaray,

497

Paul M Thompson, Katie L McMahon, Margaret J Wright, Peter M Visscher, Paul A Dawson, Cheryl

498

Dissanayake, Valsamma Eapen, Helen S Heussler, Allan F McRae, Andrew J O Whitehouse, Naomi R

499

Wray, and Jacob Gratten. Autism-related dietary preferences mediate autism-gut microbiome associ-

500

ations.

Cell

, November 2021.

501

[55] Dae-Wook Kang, Zehra Esra Ilhan, Nancy G Isern, David W Hoyt, Daniel P Howsmon, Michael Shaffer,

502

Catherine A Lozupone, Juergen Hahn, James B Adams, and Rosa Krajmalnik-Brown. Differences in

503

fecal microbial metabolites and microbiota of children with autism spectrum disorders.

Anaerobe

504

49:121–131, February 2018.

505

[56] A Gonzalez, Y V ́azquez-Baeza, J B Pettengill, A Ottesen, D McDonald, and R Knight. Avoiding

506

pandemic fears in the subway and conquering the platypus.

mSystems

, 1(3), May 2016.

507

[57] Simon H Ye, Katherine J Siddle, Daniel J Park, and Pardis C Sabeti. Benchmarking metagenomics

508

tools for taxonomic classification.

Cell

, 178(4):779–794, August 2019.

509

CC-BY-NC-ND 4.0 International license

available under a

(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made

The copyright holder for this preprint

this version posted March 9, 2022.

;

https://doi.org/10.1101/2022.02.25.482050

doi:

bioRxiv preprint