Plasma and Fecal Metabolite Profiles in Autism Spectrum Disorder

Abstract

Background

Autism Spectrum Disorder (ASD) is a neurodevelopmental condition with hallmark behavioral manifestations including impaired social communication and restricted repetitive behavior. In addition, many affected individuals display metabolic imbalances, immune dysregulation, gastrointestinal (GI) dysfunction, and altered gut microbiome compositions.

Methods

We sought to better understand non-behavioral features of ASD by determining molecular signatures in peripheral tissues through mass spectrometry methods (LC/MS and DMS-MS) with broad panels of identified metabolites. Herein, we present the global metabolome of 231 plasma and 97 fecal samples from a large cohort of children with ASD and typically developing (TD) controls.

Results

Differences in amino acid, lipid, and xenobiotic metabolism discriminate ASD and TD samples. Our results implicate oxidative stress and mitochondrial dysfunction, hormone level elevations, lipid profile changes, and altered levels of phenolic microbial metabolites. We also reveal correlations between specific metabolite profiles and clinical behavior scores. Furthermore, a summary of metabolites modestly associated with GI dysfunction in ASD are provided, and a pilot study of metabolites that can be transferred via fecal microbial transplant into mice were identified.

Conclusions

These findings support a connection between metabolism, GI physiology, and complex behavioral traits, and may advance discovery and development of molecular biomarkers for ASD.

INTRODUCTION

Many diseases are associated with informative metabolic signatures, or biomarkers, that enable diagnoses, predict disease course, and guide treatment strategies. In contrast, autism spectrum disorder (ASD) is diagnosed based on observational evaluation of behavioral symptoms, including reduced social interaction and repetitive/stereotyped behaviors (1). The average age of ASD diagnosis is between 3–4 years old (2), at which time children can receive behavioral therapy, the gold standard treatment. Because earlier diagnosis improves efficacy of behavioral therapies (3, 4), molecular biomarkers represent an attractive approach for identifying ‘at-risk’ populations and may aid development of personalized therapies. This prospect is increasingly important given the rising rate of ASD diagnoses, which currently stands at up to 1 in 59 children in the United States (2), with high variability of worldwide estimates (~1–2%) (5, 6) and no FDA approved drugs for core behavioral symptoms.

Metabolic abnormalities have been reported in ASD (7), though most have measured a small subset of metabolites and many outcomes have not reproduced between cohorts. Mitochondrial dysfunction, which heavily influences systemic metabolism, is estimated to be higher in ASD than controls (5% vs ~0.01%) (8), and genes crucial for mitochondrial function are risk factors for ASD (9). The metabolic abnormalities associated with mitochondrial dysfunction in ASD affect cellular energy, oxidative stress, and neurotransmission in the gut and the brain (9–21). Other metabolic profiles in ASD implicate aromatic and phenolic metabolites, including derivatives of nicotinic, amino acid, and hippurate metabolism (22–32). Various amino acids are detected at differential levels across studies and across sample types, but consistent patterns are difficult to discern (17, 26, 28, 31, 33–36).

Some discrepancies between studies are likely due to differences in sample number, tissue analyzed, and methodology. Other sources of variability include differences in environmental factors, such as diet and gut bacteria, which differ between ASD and TD populations and can influence each other (37–39). Diet is a major source of circulating metabolites, impacting the metabolome directly or indirectly through chemical transformation by the trillions of gut microbes, the microbiome, which has been proposed to modulate complex behaviors (40–42). Such proposed environmental modulators of ASD may integrate with genetic risks to impact behavioral endpoints through the actions of small molecules produced in tissues outside the brain.

Herein, we present a comprehensive comparison of an extensive panel of identified metabolites in human plasma and feces from a large cohort of matched ASD and TD children. We identified differential levels of metabolites ranging from hormones, amino acids, xenobiotics, and lipids, many of which correlate with clinical behavior scores. To our knowledge, this is the first study to concurrently evaluate paired intestinal and systemic metabolomes in a high-powered analysis with a large number of identified metabolites, allowing direct associations between metabolites previously highlighted in ASD samples and discovering new metabolites of interest. These findings support the emerging concept of evaluating non-behavioral features in the diagnosis of ASD.

METHODS AND MATERIALS

Participants:

Samples for this study, aged 3–12 years old, were collected through the UC Davis MIND institute (43, 44). ASD diagnosis was confirmed by trained staff using the Autism Diagnostic Observation Schedule (ADOS), and the Autism Diagnostic Interview-Revised (ADI-R). Subjects were diagnosed prior to 2013 based on DSM IV. TD participants were screened using the Social Communication Questionnaire (SCQ) and scored within the typical range, (below 15). Ninety-seven of the participants, who also provided the stool samples, completed an additional evaluation to determine gastrointestinal (GI) symptoms. GI status was determined using the GI symptom survey (GISS), based upon Rome III Diagnostic Questionnaire for the Pediatric Functional GI Disorders (45) and described in detail elsewhere (44).

Additional methods in supplement.

RESULTS

Plasma and Fecal Metabolomes Differ between ASD and TD

Plasma samples from 130 ASD and 101 TD children were analyzed along with fecal samples collected from a subset of these same ASD (n=57) and TD individuals (n=40) (Figures S1A–S1G). Samples and metrics of behavioral and GI scores were obtained from the UC Davis MIND institute (43, 44). The ASD group was stratified into subsets of children with GI symptoms (ASD+GI) or without GI symptoms (ASD-GI), to explore potential effects of comorbid intestinal dysfunction in ASD (40 out of 130 ASD samples were +GI). This stratification was based on symptoms associated with ASD including diarrhea, constipation, and irritable bowel syndrome (IBS)-like symptoms.

Samples were analyzed by the global metabolite panel (plasma and fecal samples) and the complex lipid panel (plasma samples) by Metabolon, Inc. (Durham, NC), which identified a total panel of 1,611 plasma and 814 fecal metabolites (Tables S1–S6). Overall, we discovered that 194 metabolites were differentially abundant between the ASD and TD groups in plasma and 19 metabolites were differentially abundant in feces (Figures 1A,1B, Tables S1 and S4). Using a quantitative assay for a targeted panel of metabolites, we observed a high correlation between relative abundance and precise concentration (r²~0.97) for all tested metabolites except N-acetylserine (r²=0.63) (Figure S1H). Overall, these data expand on previous evidence that the metabolomic profile of ASD and TD populations display differences not only in the gut compartment, but also in circulation, which may affect the levels of metabolites throughout the body, including the brain (46, 47).

Figure 1. Plasma and Fecal Metabolomes Differ between ASD and TD groups.

(A-B) The number of significantly elevated and decreased metabolites (p-value<0.05, q-value<0.1) in ASD samples compared to the TD control group by ANOVA contrasts in plasma and feces, respectively. Samples are stratified by all samples or samples without or with GI symptoms (−GI, +GI). (C-D) Pathway analysis results of human plasma and fecal comparisons (all samples), indicating which metabolomic pathways are the most altered between groups, with enrichment value plotted and p-value to the right of each bar. Metabolites within each pathway could be observed at either higher or lower levels, as this plot only indicates changes. (E-F) Top 30 most distinguishing metabolites between each group in plasma and feces by random forest analysis, with mean decrease accuracy along the x-axis, which is determined by randomly permuting a variable, running the observed values through the trees, and then reassessing the prediction accuracy. If a variable is important to the classification, the prediction accuracy will drop after such a permutation. Metabolites known to be produced by (asterisks) or influenced by (triangles) the gut microbiota are denoted. The super pathway to which each metabolite belongs to is indicated by color of sphere and defined in the legend. (G-H) Scaled intensity values indicating relative levels of the most distinguishing molecules between ASD and TD (all samples) in plasma. Asterisks indicate significance (p-value<0.05, q-value<0.1) in ANOVA contrasts performed on total metabolomics dataset. Data are represented as mean ± SEM. (I-J) Scaled intensity values indicating relative levels of the most distinguishing molecules between ASD and TD (all samples) in feces. Asterisks indicate significance (p-value<0.05, q-value<0.1) in ANOVA contrasts performed on total metabolomics dataset. Data are represented as mean ± SEM. (K) Top correlated plasma metabolites that covary with margaroylcarnitine and indolelactate. (L) Top correlated fecal metabolites that covary with nicotinamide and 9-HOTrE. LPC, lysophosphatidylcholine; CE, cholesterol ester; FFA, free fatty acid; androst., androstane; hydroxypreg, hydroxypregnenalone; PFOS, perfluorooctanesulfonic acid; hydroxy-CMPF, hydroxy-3-carboxy-4-methyl-5-propyl-2- furanpropionate; DHEA-s, dehydroepiandrosterone sulfate; 9-HOTrE, 9S-hydroxy-10E,12Z,15Z-octadecatrienoic acid; AMP, adenosine monophosphate.

We then used Random Forest machine learning analysis to determine if metabolite profiles can unbiasedly predict whether the sample came from an ASD and TD donor. Overall, the modest predictive accuracy of this machine learning approach was 69% for plasma and 67% for feces. To test whether focusing on the most discriminating metabolites would improve the prediction, we repeated the Random Forest analysis using the top 30 metabolites and found that the predictive accuracy for plasma improved slightly to 70% when using all ASD samples, and to 74% when using only ASD-GI samples. For fecal samples, the predictive accuracy improved to 75% using all ASD and to 67% using only ASD-GI. The top 30 metabolites, calculated by measuring the mean decrease in accuracy of the machine learning algorithm, are useful to describe the strongest drivers of overall metabolic differences between ASD and TD populations. These most discriminatory metabolites were primarily from the lipid, amino acid, xenobiotic, and cofactor/vitamin super pathways (Figures 1E–1F). Several of these metabolites have been previously linked to ASD, such as steroids, bile acids, acyl-carnitines, and nicotinamide metabolites (48–51). Further, multiple molecules known to be produced or manipulated by the gut microbiota also featured prominently, including 4-ethylphenyl sulfate (4EPS), which is elevated in an ASD mouse model, and indolelactate, a microbe-derived tryptophan metabolite (Figure 1E) (52, 53). The two most discriminatory molecules in plasma (Figures 1G–1H) and feces (Figures 1I–1J) are depicted, which are detected in almost every sample except for 9HoTre, which has a slightly lower percent fill in ASD compared to TD samples (84% vs 91%, respectively). Metabolites correlating the strongest with these discriminatory metabolites are closely related on a structural and metabolic level (Figures 1K–1L).

Metabolite Levels Correlate with Clinical Behavioral Scores

Using clinical metadata for ASD individuals, we correlated the levels of individual metabolites to the verbal, social, and nonverbal scores of standard diagnostic measures: the Autism Diagnostic Interview, Revised (ADIR), a parent questionnaire, and the cumulative Autism Diagnostic Observation Schedule severity score (ADOS-SS), conducted by trained health professionals (1) (Tables S1 and S4). To contextualize the biological relevance of different metabolomes between ASD and TD, individual metabolites were integrated into biochemical pathways for pathway enrichment analysis, revealing the degree of change within each. Here, we identified changes mostly in lipid, xenobiotic, and nucleotide pathways associated with diverse physiological processes (Figures S2C). We found that verbal and social scores primarily correlate with lipid metabolism pathways and that nonverbal scores have the fewest correlations. Verbal scores show modest, negative correlations with levels of individual metabolites including various plasma glycero-lipids (i.e. diacylglycerols, glycerophophorylcholine, monoacylglycerols), and both plasma and fecal levels of bile acids and sphingolipids. However, these correlations are not particularly striking on an individual level (Figure S2D, Table S1 and S4). Social scores positively correlate with total free fatty acids levels in plasma, most significantly with a C18 chain length (Figure S2E, Table S1). After FDR correction, no individual metabolites significantly correlated with nonverbal score.

The ADOS-SS correlated with diverse metabolite pathways, including amino acids and food/plant component pathways (S2C), which may be partly due to the diverse array of symptoms integrated into the ADOS-SS score. On an individual metabolite level, we observe significant positive correlations and trends between the ADOS-SS and metabolites in pathways of oxidative stress (cysteine, methionine, SAM and glutathione pathways) (Figure 2A, Table S1). Some of these molecules were found in higher levels in the ASD feces and at lower levels in the ASD plasma, such as hypotaurine and to a lesser degree, taurine (Tables S1 and S4, Figures 2B, 2C, S3A and S3B), which might indicate altered fecal production, excretion or differential uptake into the plasma potentially through varied intestinal permeability. Taurine plays many roles throughout the host, and has previously been measured at altered levels in ASD, although with little consensus (19, 20, 26–28, 35, 54–56). Hypotaurine and taurine deficiency has been shown to lead to defects in cell differentiation in the brain (56) and their dysregulation could alter neuronal signaling (57).

Figure 2. Metabolite Levels Correlate with ADOS-SS.

(A) Correlation of ADOS-SS with metabolites from the cysteine, methionine, and glutathione pathways. Significant metabolites corresponding to the linear regression in the graph are listed along with spearman coefficients and p-values. Refer to color legend at bottom. (B-C) Scaled intensity values indicating relative levels of hypotaurine in feces (B) and plasma (C) (all samples). Data are represented as mean ± SEM. Asterisks indicate significance in ANOVA contrasts performed on total metabolomics dataset with an FDR cutoff of q<0.1. (p-values: **p<0.01, ***p<0.001). (D) Correlations of gamma-glutamyl amino acids with ADOS-SS, with spearman coefficients and p-values to the right. Refer to color legend at bottom. (E) The top 5 most positively correlated plasma metabolites with ADOS-SS, with spearman coefficients and p-values to the right. Refer to color legend at bottom.

Dysregulated amino acid degradation, homeostasis, and import into the brain have been implicated as a cause of neuronal stress in ASD, and supporting metabolomic data has shown perturbations of various amino acid pathways, such as glutamate, methionine, glutathione, and gamma-glutamyl metabolites (12, 19, 26, 28, 31, 35, 51, 58, 59), some of which show associations with the ADOS-SS (Table S1). Oxidative stress-related glutathione pathway precursors, gamma-glutamyl amino acids, can influence levels of neurotransmitters such as gamma-aminobutyric acid (GABA) (57, 60), which is widely thought to play a role in ASD. In a recent ASD study, an experimental treatment that led to increased gamma-glutamyl AAs and other redox pathway metabolites correlated with improved behavior metrics in children (54). Here, we observe significant negative correlations and trends between many gamma-glutamyl amino acids and the ADOS-SS (Figure 2D, Table S1). We also observe some perturbations in the urea cycle, which processes the amino group of amino acids for excretion in urine (Figure S3C–S3D). Abnormalities in this pathway can be indicative of altered amino acid degradation observed in ASD and can lead to neurotoxic accumulation of nitrogen-containing compounds in the blood (61).

Additionally, the 5 plasma metabolites most positively correlated with ADOS-SS are all involved in cellular energy pathways (Figure 2E, Table S1). Differences in energy markers could indicate a neurodevelopmental phenotype during periods when the high lipid and energy requirement in the brain is crucial (62–64). Overall, these correlations support the involvement of lipid, amino acid, and xenobiotic metabolism in the etiology of ASD, as previously described (10, 50, 58), and reveal new candidates for ASD biomarkers that correlate with symptom severity.

Altered levels of Cellular Energy and Oxidative Stress Metabolites in ASD

In line with the observed correlations between oxidative stress and cellular energy with the ADOS-SS, we further observed altered levels of related metabolites between ASD and TD samples. Metabolites that are markers of mitochondrial and oxidative stress can offer a snapshot into cellular metabolic states. These markers include acyl-carnitines, which have been highlighted in various ASD studies and are established indicators of mitochondrial dysfunction (13, 14, 16–20, 26, 27, 35, 48, 55, 65–69). Acyl-carnitines are formed to allow transport of lipids across the mitochondrial membranes for beta-oxidation, and abnormal levels of these conjugated lipids accumulate to higher levels with a decrease in beta-oxidation. Interestingly, high levels of short acyl-carnitines are found in rodent models where ASD-like behaviors are induced with the short chain fatty acids, valproic and propionic acids (70).

We found differential levels of acyl-carnitines in ASD, creating a pattern of more abundant short chain acyl-carnitines and less abundant long chain acyl-carnitines in the ASD-GI samples compared to TD samples (Figure 3A). Acyl-carnitines trend toward positive correlations with more severe social behavior, an effect driven by structures with shorter moieties (C2–C14) (Figure S2C, Table S1). In fecal samples, acetylcarnitine (C2) and free carnitine were elevated in ASD (Figures 3B–3C) and were highly discriminatory (Figure 1F). Other mitochondrial markers in both plasma and feces were also differentially abundant in ASD (Figure 3D) along with markers of phospholipid metabolism, which occurs largely in the mitochondria and was an enriched pathway in fecal comparisons (Figures 3D and 1D). Such defects in cellular metabolism support the theory that mitochondrial dysfunction may not only be comorbid with ASD but also a potential contributing factor, as suggested by numerous previous reports (8, 15, 66, 68, 70), and alterations to levels of tricarboxylic acid (TCA) cycle intermediates have been observed in human ASD prefrontal cortex samples (51). Together with the observed correlations between metabolites and ADOSS SS, these results corroborate and extend a growing body of research into altered mitochondrial metabolism and oxidative stress in ASD.

Figure 3. ASD Abnormalities within Cellular Energy and Oxidative Stress Metabolites.

(A) Log2 fold change of acyl-carnitines in the plasma of ASD-GI samples compared to controls. Significance indicated by color according to legend below, determined by ANOVA contrasts. Star indicates that a trend or significance was observed but only in the comparison between all samples. (B-C) Scaled intensity values indicating relative levels of acetylcarnitine(C2) and carnitine, respectively, in ASD fecal samples compared to TD controls (all samples). Data are represented as mean ± SEM. Asterisks indicate significance in ANOVA contrasts (FDR cutoff q-value<0.1) performed on total metabolomics dataset (p-values: **p<0.01, ****p<0.0001. (D) Schematic of mitochondrial markers and other metabolites closely associated with cellular energy in plasma (within center box) and feces (boxed to left). *=significant only in ASD-GI. Color of text indicates direction and significance of change according to legend above. PC, phosphatidylcholine; PE, phosphatidylethanolamine; GPG, glycerophosphoglycerol; GPC, glycerophosphocholine; GPE, glycerophosphoethanolamine; GPS, glycerophosphoserine; GPI, glycerophosphoinositol; TAG, triacylglycerol; DAG, diacylglycerol; FFA, free fatty acid; SAT, saturated fatty acid; PUFA, polyunsaturated fatty acid.

Transfer of ASD Fecal Microbiota into Mice is Accompanied by Metabolic Signatures

Since microbial metabolites ranked highly in the Random Forest machine learning analysis, we tested if any of the observed metabolite differences in humans could be transferred to mice via fecal microbial transplant. We selected 4 male donor samples from each of the ASD and TD groups and colonized 2–3 male germ-free mice per donor for three weeks before collecting plasma and fecal samples for metabolite profiling and bacterial DNA sequencing, respectively (Figure S4A). Global metabolomic analysis revealed that colonized mice modestly cluster by donor and group when differential metabolites are considered in principle component analysis (Figure S4B). We selected the human donors based on 4EPS levels (Figure S4C), due to its involvement in an ASD mouse model (52) and dysregulation of similar phenolic compounds in human ASD (24, 26, 30, 31, 71). 4EPS is not produced by the host and is strictly a bacterial metabolite (52, 72). Surprisingly, we observed 4EPS levels in a bimodal distribution in mouse samples (Figure S4D–S4E). In spite of the surprising results with 4EPS, many of the metabolites with the highest fold change and lowest p-value are indeed other phenolic molecules such as metabolites of hippurate, tyrosine, and diet-derived phenols, some of which have been similarly measured in a previous study transferring fecal microbes from ASD individuals into mice (73) (Figures S4F, Table S10). While preliminary, these studies reveal dysregulated microbiome-mediated effects on xenobiotic pathways and phenolic metabolites in ASD.

Steroid Hormone Levels are Elevated in ASD

Multiple human ASD studies have examined the levels of specific steroid metabolites within androgenic, pregnenolone and progesterone metabolism, with some finding aberrant levels, positive correlations with ASD severity, and behavioral improvement following treatments that lower levels of certain hormones (49, 65, 74–81). On the other hand, in a recent clinical trial of ASD children given an antioxidant treatment, levels of pregnenolones and androgens increased and correlated with improved behavior (54). In our dataset, we found increases of many detected metabolites within the pregnenolone and androgen pathways (Figures 1C, 4A, Table S1). This is an indicator that the physiological pathways associated with the downstream metabolism of cholesterol are significantly altered between ASD and TD populations. There does not appear to be a global change in steroid metabolism, as most primary bile acid and sterol metabolites were unaffected (Tables S1 and S4). We observed some elevation of these hormone levels independent of sex, which is notable considering the male bias in ASD, reflected in our primarily male sample set (7–15% female) (82) (Figure S6A and S6B). Because a cluster of our samples are from older individuals in the ASD group, and to account for age-dependent increases in androgens (Table S7), we stratified by age and still observed heightened androgenic and pregnenolone metabolite levels in ASD subpopulations (Figure S6C). Taken together, these data indicate that steroidal hormone metabolism may be altered in the ASD population relative to TD samples and that these differences are not driven solely by sex or age differences in our cohort.

Figure 4. Steroid Hormone Levels are Elevated in ASD and other Lipid Metabolite Levels Differ in ASD.

(A) Significant alterations to levels of all metabolites detected in the pregnenolone, progestin, and androgen steroid pathways in plasma (P) and feces (F), with colors indicating significance and fold change direction according to legend at bottom right, and numerical fold change in text within the box. (B) Complex lipid panel results for all the ASD plasma samples compared to TD controls with acyl chain length of lipids across the top, described by chain length, degree of unsaturation and categorized by saturated (SAT), monounsaturated (MUFA), and polyunsaturated (PUFA) fatty acids. Lipid classes are listed along the left. Direction of change and significance are indicated by the legend. Significance determined by ANOVA contrasts, FDR cutoff q-value<0.1.

Lipid Metabolite Levels Differ in ASD

Lipids are crucial for energy storage, cellular membrane integrity and cell signaling. They play a variety of roles in the central nervous system, and their dysregulation has been linked to ASD (11, 13). Phospholipids have been measured at lower levels, while long chain fatty acids are reportedly elevated, but PUFAs have been measured at both higher and lower levels, depending on the cohort (11, 14, 15). Here, we performed a comprehensive, quantitative metabolite analysis on complex lipids. The concentrations of 999 lipids were quantified, and 13.7% of these were significantly different in the ASD samples, with 4.6% increased and 9.1% decreased compared to TD samples. Many of these differentially abundant lipids included phospholipids, cholesterol esters and glycerolipids (Figure S7A). In general, shorter (14–18 chain length) saturated fatty acids were less abundant in ASD throughout the lipid classes (Figures 4C, S7B).

PUFA lipid levels were also different, with elevations in diacylglycerols and free fatty acids, and an enrichment of the 18:2 (linolenic) chain length in most lipid classes (Figure 4C). Fecal samples generally trended in the opposite direction from plasma in PUFA lipids (Figure S7C, Table S4). PUFA lipids containing linolenic acids are precursors to other PUFAs (arachidonic and docosahexaenoic acids) important for brain development, function, and structural integrity (14). Future studies are needed to determine if these changes in lipid levels contribute to ASD.

Considering the prevalence of intestinal issues in ASD, GI-dependent metabolite analysis between individuals may provide useful insights. Accordingly, we found a total of 87 and 24 metabolites in plasma and feces, respectively, that were potentially differential within the ASD+GI compared to ASD-GI individuals by ANOVA contrasts, but none passed the false-discovery rate cutoff (Figures S5A–S5B). At the pathway level, we found enrichments in free fatty acids between ASD+/−GI plasma (Figure S5C), but none in fecal samples (Table S4). In plasma, free fatty acids of multiple chain lengths trended lower in ASD+GI compared to ASD–GI samples (Figure S5D). Some fatty acids, like PUFAs, are anti-inflammatory and their lower levels may contribute to GI dysfunction directly (14). We did not observe significant correlations between lipid (or other) metabolite patterns and GI symptoms or disease, including diarrhea, constipation, or IBS. While these results are interesting, the implications of correlating an altered metabolome and GI symptoms in ASD remain to be determined.

Differential Phenolic Xenobiotic Metabolite Levels in ASD

Phenolic metabolites, a diverse structural class of thousands of molecules containing a phenol moiety, come from dietary ingredients or biotransformation of aromatic amino acids by the gut microbiota. In their free phenolic forms they can be readily absorbed through intestinal tissues; however, microbial modification of these molecules can significantly alter their absorption, bioavailability, and bioactivity (83), leading to various benefits or harm to the host (83, 84). In fact, altered levels of phenolics have been highlighted in many ASD metabolomic studies (20, 22–28, 30, 33, 34, 54, 71, 85, 86). However, a consensus of these metabolite changes in ASD has yet to be reached, and most studies have only measured a subset of these metabolites.

We observed altered or trending levels of phenolic metabolites belonging to several interrelated pathways, including tyrosine, benzoate, and food component/plant metabolites (Figure 5A). Some share structural similarity to para-cresol sulfate, a toxic molecule elevated in the urine of young ASD children (22, 24). These include, among others, 4-ethylphenyl sulfate (4EPS), 2-ethylphenylsulfate, cresol derivatives, 4-allylphenyl sulfate, and 4-methylbenzenesulfonate, the latter of which was elevated a remarkable 60-fold in the small subset of ASD samples in which it was detected (Figure 5A–5D, Table S1). Changes in the levels of these phenolic molecules are also observed in animal models of ASD (52, 87). Previously, 4EPS was observed with a 46-fold elevation in the maternal immune activation mouse model, and administration of synthetic 4EPS to wild type mice was sufficient to induce an anxiety-like phenotype (52). Here, in ASD plasma samples, 4EPS levels were increased 6.9-fold (Figure 5D). This is largely driven by a considerable subpopulation of ASD individuals with high levels of 4EPS, who also show higher levels of variation in related phenolic metabolites (Figure S8A). In many cases phenolic metabolite levels correlate with 4EPS in ASD samples, including 4-acetylphenyl sulfate, a derivative of 4EPS, and others (Figure 5E, S8B, and S8B). These observed alterations, combined with mounting evidence from previous studies, suggest that phenolic metabolites may be involved in ASD.

Figure 5. Differential Phenolic Xenobiotic Metabolite Levels in ASD.

(A) Phenolic metabolites, belonging to the benzoate, tyrosine, and food component/plant pathways that are significantly different between ASD vs TD groups in plasma (P), and feces (F). Directionality and significance defined in legend, and numerical fold change in text within the box. (B) Scaled intensity values indicating relative levels of 4-allylphenol sulfate in plasma (all samples). (C) Scaled intensity values indicating relative levels of 2-ethylphenyl sulfate in plasma (all samples). (D) Scaled intensity values indicating relative levels of 4-ethylphenyl sulfate (4EPS) in plasma (all samples). (E) Spearman correlation between plasma 4EPS and 4-acetylphenol sulfate, log₂ scale. Data in (B-D) are represented as mean ± SEM. Asterisks indicate significance in ANOVA contrasts with an FDR cutoff q-value<0.1 performed on total metabolomics dataset (p-values: **<0.01, ****<0.0001).

DISCUSSION

Changes in the metabolome have been linked to a number of neurodevelopmental and neurodegenerative disorders (88, 89). The current study includes a comprehensive profiling of the metabolites from 231 plasma and 97 matched fecal samples from ASD and TD individuals who have been extensively behavior tested. We observed that specific individual metabolites and metabolic pathways in lipid, amino acid, and xenobiotic metabolism were altered between ASD and TD groups, and that many of these same molecules and pathways correlated to severity of ASD behavioral scores. The core findings are summarized in Figure 6. Various steroid hormone metabolite levels were elevated in ASD samples. PUFA levels, and short chain acyl-carnitines were generally elevated in ASD plasma, while saturated fatty acid levels and long-chain acyl-carnitines were decreased, contributing to a general picture of dysregulated cellular energy and oxidative state, with potential connections to mitochondrial dysfunction in ASD. Some of the lipid and xenobiotic metabolites also showed interesting changes within the ASD group when stratified by the presence of GI symptoms. Finally, many phenolic metabolites, derived largely from host and bacterial metabolism of amino acids, plant polyphenols, and other food components were detected at differential levels in plasma and feces between groups.

Figure 6. Summary chart of core findings.

Categorized into metabolites of lipid, mitochondrial function marker, and xenobiotic and phenolic pathways as well as potential biomarkers, the metabolites of most interest (left) and observations made from the data (right) are summarized.

ASD is diagnosed by behavioral tests, with extensive heterogeneity of symptoms, severity, and etiology between individuals and little consensus on molecular mechanisms. This enigmatic spectrum has a strong but complex genetic basis, with hundreds of reported risk genes (90). The contributions of risk alleles to behavior is a vast and active area of research. In addition to genetics, understanding of altered metabolite levels in blood, feces, and brains of ASD individuals may provide a glimpse into physiologic aspects of the disorder and hold the potential to advance diagnosis and/or stratification of sub-populations of ASD. In fact, our results broadly support the notion of multiple etiologies or endophenotypes within the ASD population.

Interestingly, multiple mouse models support the notion that gut metabolites are associated with brain development and function (50, 91, 92). Some metabolites are closely linked with neurological disorders, either to positive or negative outcomes (93, 94). Several examples have been reported of gut metabolites entering circulation and directly affecting the brain, as well as cases where metabolites stimulate pathways in the gut, immune, or autonomic nervous system and exert changes to the brain and to behavior (95–98). Comprehensive metabolic profiling in humans and animal models provides insight into the molecular status of disease and how genetic factors and environmental influences (dietary, microbial, and chemical) interact. Deeper analysis of our dataset along with additional studies, with future empirical studies to validate the relevance of our observations, could illuminate aspects of ASD pathophysiology. The intriguing correlations between ASD behaviors, altered levels of fecal and plasma metabolites, and GI symptoms contribute to the concept that ASD may be viewed as a whole-body condition, and argue for increased investigation into peripheral aspects of disease that may lead to advances in diagnosis and improved stratification of ASD populations.

KEY RESOURCES TABLE

Resource Type	Specific Reagent or Resource	Source or Reference	Identifiers	Additional Information
Add additional rows as needed for each resource type	Include species and sex when applicable.	Include name of manufacturer, company, repository, individual, or research lab. Include PMID or DOI for references; use “this paper” if new.	Include catalog numbers, stock numbers, database IDs or accession numbers, and/or RRIDs. RRIDs are highly encouraged; search for RRIDs at https://scicrunch.org/resources.	Include any additional information or notes if necessary.
Biological Sample	Human fecal and plasma samples	UC Davis Mind Institute
Biological Sample	Murine Fecal Samples	Mazmanian Laboratory Colony
Commercial Assay Or Kit	Qiagen DNeasy powersoil extraction kit	Qiagen	Cat No./ID: 12888-100
Deposited Data; Public Database	Data	Mendeley Data	https://data.mendeley.com/datasets/rmkrcdyng3/draft?a=b461830b-7151–45fa-a18f-37014315f33d
Organism/Strain	C57Bl6 germ-free mice	Mazmanian Laboratory Colony