Abstract
Currently, there are no reliable biomarkers for autism diagnosis. The heterogeneity of autism and several co-occurring conditions are key challenges to establish these. Here, we used untargeted mass spectrometry-based urine metabolomics to investigate metabolic differences for autism diagnosis and autistic traits in a well-characterized twin cohort (N=105). We identified 208 metabolites in the urine samples of the twins. No clear, significant metabolic drivers for autism diagnosis were detected when controlling for other neurodevelopmental conditions. However, we identified nominally significant changes for several metabolites. For instance, phenylpyruvate (p=0.019) and taurine (p=0.032) were elevated in the autism group, while carnitine (p=0.047) was reduced. We furthermore accounted for the shared factors, such as genetics within the twin-pairs, and report additional metabolite differences. Based on the nominally significant metabolites for autism diagnosis, arginine and proline metabolism pathway (p=0.024) was enriched. We also investigated the association between quantitative autistic traits, as measured by the Social Responsiveness Scale 2nd Edition, and metabolite differences, identifying a greater number of nominally significant metabolites and pathways. A significant positive association between indole-3-acetate and autistic traits was observed within twin-pairs (adjusted p=0.031). The utility of urine biomarkers in autism, therefore, remains unclear, with mixed findings from different study populations.
Lay Abstract Earlier literature has suggested that there are chemical molecules (metabolites) in the urine samples of autistic individuals that could be linked with their diagnosis of autism. However, there are still mixed results and uncertainty if any of these metabolites could be used as biomarkers (chemical molecules that indicate a biological condition). In this study, we analysed urine samples of 105 twins from Sweden, of which 48 were identical or fraternal twins, and the analysed metabolite levels were compared between those twins diagnosed with autism and those without a diagnosis. Additionally, we tested if the metabolite levels were different in relation to the level of autistic behaviour. The metabolite levels were measured using mass spectrometry, allowing us to identify hundreds of chemical molecules present in a urine sample followed with testing their significance for autism diagnosis or autistic traits. No single metabolite was found to be highly associated with a diagnosis of autism. However, indole-3-acetate, a metabolite produced by the breakdown of the amino acid tryptophan, was associated with the level of autistic behaviour within the twins. Nevertheless, several metabolites showed some association with autism or the level of autistic behaviour, while revealing interesting pathways for studies in the future.
Introduction
Autism is a neurodevelopmental condition with a heterogeneous presentation that is diagnosed in 1 to 2% of the world population (Lord et al., 2020). Although several genes and genetic variants have been associated with autism phenotypes (Lord et al., 2020; Satterstrom et al., 2020; Zhou et al., 2022), there is no evidence of a single biomarker that can be of benefit to the diagnostic procedure of the condition (Cortese et al., 2023). Finding biomarkers that could aid either in the diagnosis or specifying a subgroup of individuals could help, for instance, in the early diagnosis of autism (Amaral et al., 2019).
Prior research has suggested an association of metabolic anomalies with autism that could potentially aid in the search for biomarkers (Jensen et al., 2022). Metabolomic analyses using urine samples provide a non-invasive method with greater ease of sample collection and handling in comparison to other biospecimens. This is particularly advantageous in clinical settings where the process of diagnosing autism can begin at a younger age (Lord, 2019). Moreso, the ability to detect metabolites in urine samples is well-established using mass spectrometry (MS) and nuclear magnetic resonance spectroscopy (NMR), and the typical human urine metabolome has been mapped (Bouatra et al., 2013). Early efforts to understand the urine metabolome in autism began several decades ago, with initial biochemical studies searching for associations (Himwich et al., 1972; Young et al., 1978). The investigations have shifted to more untargeted and high-precision analytical approaches using NMR (Ma et al., 2021; Mavel et al., 2013) and MS (Mussap et al., 2020; Noto et al., 2014; Timperio et al., 2022).
The interest in autism metabolomics has continued to expand; however, more accessible data are required across different populations to identify robust associations. Such an approach will also make it possible to characterise how both, situational and long-term environmental factors interplay with the genetic backgrounds of autism (Bai et al., 2019). Twin studies are an appropriate tool to investigate the contribution of environmental and genetic factors (Colvert et al., 2015; Frazier et al., 2014) and can be extended to metabolic profiles. It is especially valuable to analyse differences between monozygotic (MZ) twins who are discordant for autism, to tease out genetic contributions.
In this study, we utilised a rare twin sample, the Roots of Autism and ADHD Twin Study in Sweden (RATSS) (Bölte et al., 2014; Myers et al., 2021), enriched for MZ twins discordant for autism, and their urine sample-based metabolomics to search and validate potential metabolite biomarkers. Furthermore, on the basis of these findings we have mapped the affected metabolic pathways. Our study adds on to the literature exploring the metabolites as possible biomarkers for autism and expands the search for autistic traits.
Methods
Study participants
In this study, individuals (N=105) and 48 complete twin pairs were selected from the RATSS cohort (Bölte et al., 2014; Myers et al., 2021) for untargeted mass spectrometry-based urine metabolomics. Study participants were age- and sex-matched based on the autism diagnosis status and availability of urine samples. The study was approved by the Swedish Ethical Review Authority (2016/1452-31). Informed consent was obtained from all participants or their caregivers, depending on their age.
The autism diagnosis is based on a comprehensive psychodiagnostic assessment during the study visit, administered by a team of experienced clinicians (Bölte et al., 2014). Briefly, diagnoses were based on the Diagnostic and Statistical Manual of Mental Disorders, 5th Edition (DSM-5) (American Psychiatric Association, 2013), and behavioural assessment tools such as the Autism Diagnostic Interview-Revised (ADI-R) (Rutter et al., 2003), the Autism Diagnostic Observation Schedule 2nd Edition (ADOS-2) (Lord et al., 2012). Autistic traits were evaluated with the parent-report version of the Social Responsiveness Scale 2nd Edition (SRS-2) (Constantino, 2005, 2011; Constantino & Gruber, 2012), consisting of 65 items. Intelligence quotient (IQ) was measured by using the Wechsler Intelligence Scale for Children or Adults - IV General Ability Index (GAI) (Tulsky et al., 2001).
Urine collection and metabolite extraction
The urine samples were collected at the visit from the participants, aliquoted and stored in the Karolinska Institutet biobank at -80 °C, until sent for analysis. Urinary specific gravity was measured following centrifugation at 13,000g for 10 minutes. Urine aliquots (200 μl) were mixed with 200 μl of methanol:acetonitrile:water (50:30:20), vortexed for 30 minutes, max speed at 4 °C and then centrifuged at 16,000 g for 15 minutes at 4 °C. Supernatants were collected for metabolomic analysis.
Ultra-High Performance Liquid Chromatography (UHPLC)
For the experiments, 20 μL of samples were injected into a UPLC system (Ultimate 3000, Thermo Scientific) and were analysed on positive mode: samples were loaded onto a Reprosil C18 column (2.0 mm × 150 mm, 2.5 μm - Dr Maisch, Germany) for metabolite separation. Chromatographic separations were achieved at a column temperature of 30 °C and flow rate of 0.2 mL/min. A linear gradient (0–100%) of solvent A (ddH2O, 0.1% formic acid) to B (acetonitrile, 0.1% formic acid) was employed over 20 minutes, returning to 100% solvent A in 2 minutes and a 6-minute post-time solvent A hold. Acetonitrile, formic acid, and HPLC-grade water were purchased from Sigma Aldrich.
High Resolution Mass Spectrometry (HRMS)
The UPLC system was coupled online with a mass spectrometer, Q Exactive (Thermo Scientific), scanning in full MS mode (2 μ scans) at a resolution of 70,000 in the 67 to 1000 m/z range, target of 1 × 106 ions and a maximum ion injection time (IT) of 35 ms, 3.8 kV spray voltage, 40 sheath gas, and 25 auxiliary gas, operated in negative and then positive ion mode. Source ionization parameters were: spray voltage, 3.8 kV; capillary temperature, 300 ° C; and S-Lens level, 45. Calibration was performed before each analysis against positive or negative ion mode calibration mixes (Piercenet, Thermo Fisher, Rockford, IL) to ensure sub-ppm error of the intact mass.
Metabolite quantification
Data were normalized by urinary specific gravity because creatinine excretion may be abnormally reduced in autistic children (Whiteley et al., 2006). Replicates were exported as mzXML files and processed through MAVEN (Melamud et al., 2010). Mass spectrometry chromatograms were elaborated for peak alignment, matching and comparison of parent and fragment ions, and tentative metabolite identification (within a 10-ppm mass deviation range between observed and expected results against the imported Kyoto Encyclopaedia of Genes and Genomes (KEGG) database (Kanehisa et al., 2023).
Differential metabolomics
Differential metabolomic analysis for 208 metabolites was performed using the drgee package (Zetterqvist & Sjölander, 2015) in R (R Core Team, 2020), for overall effects and differences between twin-pairs using a generalised estimating equations model (GEE) model, with suitable covariates. In the GEE model, the metabolites were individually considered as the response variable, while the autism diagnosis status was the predictor variable. Age, sex, body mass index (BMI), medication status and diagnosis status of other neurodevelopmental conditions (NDC) were added as covariates to the model: Metabolite ∼ Autism Diagnosis + Other NDC Diagnosis + age + sex + BMI + medication status. The model was applied for the whole cohort (Model A) and for differences within twin-pairs (Model B) by using the drgee package (Zetterqvist & Sjölander, 2015) in R (R Core Team, 2020). The same model was then applied to a subset of the cohort based on zygosity. Here, age and sex were removed as covariates when testing for differences between twin-pairs as these were intrinsically controlled.
In addition to testing autism diagnosis status as a predictor variable, we used autistic traits measured by the SRS-2 (Constantino, 2005, 2011; Constantino & Gruber, 2012) total raw scores. As IQ levels are known to have an impact on the SRS-2 scores (Hirosawa et al., 2020), along with the previously stated covariates (other than diagnostic status of other NDCs), IQ scores from GAI were added to the models: Metabolite ∼ SRS total score + age + sex + BMI + medication status + IQ GAI. Also, for SRS-2 total raw scores, the models were applied to subsets of the cohort based on zygosity. As previously stated, age and sex were removed as covariates when testing for differences between twin-pairs as these were intrinsically controlled. For all instances, p values were extracted from the results of each model and adjusted for multiple comparisons using the false discovery rate (FDR) method (Benjamini & Hochberg, 1995) in R (R Core Team, 2020).
Pathway enrichment analysis
Significant metabolites from the GEE models were tested for enrichment in the Small Molecule Pathway Database (SMPDB) (Jewison et al., 2014) using over-representation analysis (ORA) from the enrichment analysis module of MetaboAnalystR (Pang et al., 2022). The cut-off threshold for entries from metabolite sets to be included was set to 2. ORA was implemented using the hypergeometric test to evaluate whether a particular metabolite set is represented more than expected by chance within the given list of metabolites. One-tailed p values were provided after adjusting for multiple testing. Based on the statistical testing, enrichment dot-plots and pathway networks were generated. In the pathway networks, 2 or more nodes were only connected by an edge when the number of shared metabolites was >25% of the combined metabolites contributing to each node.
Statistical analysis
All statistical analyses were performed in R (v4.1.2) (R Core Team, 2020). The statistical models and tests used for the analyses are described in the methodology relevant to the experimental technique, in the sections above.
Code availability
The utilised code is available on GitHub (https://github.com/Tammimies-Lab/RATSS-Metabolomics) or available upon request from the corresponding author (kristiina.tammimies{at}ki.se).
Community involvement
Autistic individuals and other community stakeholders were not involved in developing the research question and other aspects of this study.
Results
Metabolites do not clearly explain sample differences
UHPLC was coupled with HRMS for the untargeted analysis of metabolites. In the study cohort (N=105 twins), of which 48 form complete twin-pairs, we identified 208 metabolites with high reliability following mass spectrometry of urine samples. Principal component analysis (PCA) of the metabolomics data demonstrated no clear drivers of effects across PC1 (23.2%) and PC2 (10.8%) (Figure 1A). The first 10 principal components were able to explain 68.2% of the variation (Figure 1B). The loadings for PC1 (Figure 1C) and PC2 (Figure 1D) did not identify a metabolite that was a major driver of the observed effects, with the top 10 loadings having a contribution of ∼1.25% to ∼1.5 % each in PC1 and ∼1.75 to ∼2.5% each in PC2. There were 9 individuals outside the 95% CI ellipses (Figure 1A), who were designated as outliers and excluded from further analysis, leaving a study cohort with 96 individuals, of which 42 form complete twin-pairs.
(A) Scatter plot for individuals based on metabolite contributions and grouped as per autism diagnosis status. (B) Scree plot of eigenvalues of principal components from A. (C) Loading plot of metabolites contributing to principal component 1 (PC1). (D) Loading plot of metabolites contributing to principal component 2 (PC2).
Of these 96 individuals, the autistic participants had a mean age of 14 years, 11 being female and 24 having a clinical diagnosis of a NDC other than autism. The demographic and clinical features of the study cohort used for further analyses are summarised in Table 1. Specific information on race/ethnicity, socioeconomic status and educational attainment were not recorded.
Demographic characteristics of study cohort.
Subtle changes in metabolite status across study cohort
The peak areas of each identified metabolite from participants of the study cohort (N=96) were analysed for association with autism diagnosis using a GEE model, first for the whole cohort (Model A) and then within twin-pairs (Model B). We also adjusted the models for the diagnosis status of other NDCs. As expected, from the prior PCA analysis, no metabolites had significant association with autism after FDR correction for 208 tests. Based on nominal significance, a handful of metabolites were changed (Table S1A). For instance, deoxycholic acid (p=0.048), orotate (p=0.042), phenylpyruvate (p=0.019) and taurine (p=0.032) were elevated, while carnitine (p= 0.047) was reduced in autism (Figure 2A and Table S1A).
(A) Autism diagnosis across full cohort (N=96), monozygotic twins (n=52) and dizygotic twins (n=44) tested using GEE Models A and B. (B) Autistic traits across full cohort, monozygotic twins and dizygotic twins tested using GEE Models A and B. (* p<0.05, ** p<0.01, *** p<0.001)
When analysing for differences within twin-pairs, i.e., which corrects for all the shared variation between the twins such as genetics and shared environment, using the GEE model (Model B), eight metabolites were significantly modulated in the autism group based on the nominal p value (Figure 2A and Table S1A). These were increased hypoxanthine (p=0.019), diiodothyronine (p=0.049) and kynurenic acid (p=0.003), and decreased guanidoacetic acid (p=0.001), indole (p=0.001) and L-arginosuccinate (p=0.031).
The study cohort was then subsetted as per zygosity status and the same GEE models were applied for analysis. In the subset of MZ twins (n=52), 17 metabolites were identified to be significantly different between autistic and other twins (Model A, Figure 2A and Table S1B), including an increase in unique metabolites such as alanine (p=0.029), asparagine (p=0.025) and 2-oxo-4-methylthiobutanoate (p=0.007), and a decrease in creatine (p=0.001) and deoxyinosine (p=0.020). When testing for differences within the MZ twin-pairs (Model B), 17 metabolites were differentially modulated (Figure 2A and Table S1B), including additional metabolites that were elevated like glutamine (p=0.008), proline (p=0.022) and homocysteine (p=0.037). A decrease was detected in several metabolites, such as glucose-6-phosphate (p=0.032), L-kynurenine (p=0.043) and D-histidine (p=0.014), among others.
In the subset for dizygotic (DZ) twins (n=44), peak areas of only diidothyronine (p=0.039) were elevated (Model A, Figure 2A and Table S1C). After accounting for differences within twin-pairs in the GEE model (Model B), ten metabolites were significantly modulated (Figure 2A and Table S1C), such as an increase in valine (p=0.018), quinoline (p=0.034) and N-acetyl-glutamate (p=0.046). Reductions were detected in four metabolites including isocitrate (p=0.018) and indole (p=0.009).
Association between autistic traits and metabolite differences
We also investigated association between the metabolites and quantitative autistic traits measured by the SRS-2 total raw scores (Constantino, 2005, 2011; Constantino & Gruber, 2012). We analysed the association between the peak areas of each metabolite identified in the study cohort (N=96) with the SRS-2 using the GEE models and adjusted for IQ (Wechsler GAI) (Tulsky et al., 2001). Again, no metabolite reached statistical significance after FDR multiple test correction. From Model A, 30 nominally significant metabolites were detected (Figure 2B and Table S2A). Amongst those elevated were acetylcarnitine (p=0.023), aspartate (p=0.002) and hydroxyphenylpyruvate (p=0.016); and the only metabolite to be reduced was creatine (p=0.027). When testing for differences within twin-pairs (Model B), five metabolites crossed the nominal significance threshold (Figure 2B and Table S2A), including a decrease in adenosine (p=0.025) and 6-phospho-d-gluconate (p=0.033), and an increase in indole-3-acetate (p=0.0001, adjusted p=0.031).
We also investigated the association in the zygosity groups for the autistic traits and the metabolites. In the MZ subset (n=52), 12 metabolites demonstrated nominally significant changes when applying the GEE model (Model A, Figure 2B and Table S2B). An elevation was detected in coenzyme A (p=0.044), AMP (p=0.041) and a decline in levels of DL-pipecolic acid (p=0.024). Following testing for differences within the twin-pairs (Model B), six metabolites were identified (Figure 2B and Table S2B), which included an increase in prephenate (p=0.027) and D-glycerylaldehyde 3-phosphate (p=0.003) and a decrease in 5’-S-Methyl-5’-thioadenosine (p=0.001) and thiamine (p=0.032).
In the DZ subset (n=44), 23 metabolites were significantly elevated (p<0.05) in association with the autistic traits (Model A, Figure 2B and Table S2C). These were, to name a few, D-methionine (p=0.047), D-gluconate (p=0.038), myo-inositol (p=0.047) and lysine (p=0.043). When accounting for differences between twin-pairs (Model B), 16 metabolites demonstrated significant changes (Figure 2B and Table S2C) where none were reduced, including glycolate (p=0.009), 4-aminobutyrate (p=0.035) and IMP (p=0.012).
Enrichment of identified metabolites in biochemical pathways
The nominally significant metabolites identified from the GEE models were used for ORA as described in the methods. Ten biochemical pathways were identified to be enriched in the autistic twins based on the full cohort analysis (N=96, Model A) (Figure 3A and 3B), out of which there was significant enrichment only in the bile acid biosynthesis pathway (p=0.035, Table S3A). When ORA analysis was done for the metabolites found to be nominally significant for autism diagnosis between twin-pairs (Model B), ten pathways were identified (Figure 3C and 3D), where only the arginine and proline metabolism was significant (p=0.024, Table S3B).
(A) Enrichment dot-plot for top 10 enriched pathways from GEE Model A. (B) Interaction network for enriched pathways from GEE Model A. (C) Enrichment dot-plot for top 10 enriched pathways from GEE Model B. (D) Interaction network for enriched pathways from GEE Model B.
For the models based on autistic traits in the full cohort (N=96, Model A), 43 biochemical pathways were enriched (Figure 4B), out of which two crossed the significance threshold (Table S3C): glutathione metabolism (p=0.0503) and glutamate metabolism (p=0.0506). The enrichment dot-plot of the top ten pathways is depicted in Figure 4A. On the other hand, eight biochemical pathways were identified using significant metabolites identified after testing for autistic traits between twin-pairs (Model B, Table S3D). of which methionine metabolism (p=0.001) was significantly enriched (Figure 4C and 4D).
(A) Enrichment dot-plot for top 10 enriched pathways from GEE Model A. (B) Interaction network for enriched pathways from GEE Model A. (C) Enrichment dot-plot for top 10 enriched pathways from GEE Model B. (D) Interaction network for enriched pathways from GEE Model B.
Discussion
In this study, we explored the urine metabolome of a subset of individuals from the RATSS cohort (Bölte et al., 2014; Myers et al., 2021) using untargeted mass-spectrometry. While only a single metabolite reached significance after FDR correction, we identified several metabolites as nominally significant for autism diagnosis and autistic traits. Furthermore, the identified metabolites were enriched in a few relevant biochemical pathways. Increased detection of significant metabolites were found for the autistic traits, rather than for the autism diagnoses.
Our study is the first to evaluate differences in the urine metabolome of an autism twin cohort and utilise information on co-occurring NDCs in our analyses. This is an important aspect to account for, when investigating suitable biomarkers in lieu of the high prevalence of co-occurring NDCs in autism (Khachadourian et al., 2023). The size of our study cohort (N=96, with n=38 diagnosed with autism) is larger when compared to several recently published studies using urine-based mass spectrometry, such as 60 available samples from an otherwise larger cohort (Liu et al., 2022), 57 participants (Mussap et al., 2020) and 14 autism discordant sibling pairs (Timperio et al., 2022). The twin cohort also gives us the possibility to analyse the effects of genetic and environmental factors on these potential biomarkers. Despite this, we were not able to find any metabolites that reached the multiple test correction significance nor nominally associated similarly in the whole cohort and within twin-pair analyses (Figure 2A). Our results show that the contribution of the shared factors, including genetics, in the metabolite profiles is strong and therefore it is difficult to find robust biomarkers of autism.
Our study is also the first to use autistic traits as measured by the SRS-2 as a predictor of metabolite changes. Predominantly, we found a large number of metabolites being associated with autistic traits rather than autism diagnosis. This shows potential that in the future, metabolite profiles could be used for detecting differences in continuous autism operationalisations. We observed a significant positive association between indole-3-acetate and autistic traits within the twin-pairs. Elevated levels have been previously detected in autistic individuals (Gevi et al., 2016; Mussap et al., 2020). However, a decline has also been reported (Emond et al., 2013). Indole-3-acetate is a known by-product of the tryptophan metabolism modulated by the gut microbiome (Krishnan et al., 2018; Pavlova et al., 2017), with growing evidence indicating a dysbiosis (Kang et al., 2013; Wan et al., 2022).
Despite the fact that we only found nominally associated metabolites, our findings align with previously reported urine-based metabolomic studies for autism diagnosis. For instance, we observed elevated levels of phenylpyruvate (Noto et al., 2014; Timperio et al., 2022) and taurine (Ma et al., 2021; Mavel et al., 2013; Nadal-Desbarats et al., 2014; Yap et al., 2010), as noted earlier in several studies. We also show decreased levels of carnitine, similarly to what was reported in other biological samples (Fahmy et al., 2013; Filipek et al., 2004). However, direct comparisons with metabolomic studies using different biological samples should be critically evaluated due to lack of information about any correlations between such reports, as observed in other multifactorial pathologies (Erben et al., 2021). Our observations are unique when it comes to the association with autistic traits, as these have not been used as a predictor in metabolomic studies of autism before. While individual metabolites provide a snapshot into specific altered metabolic processes, interpretation of pathway level changes may be more relevant in multifactorial conditions like autism (Arora et al., 2022).
Several of the enriched pathways that were detected in this study are related to energy metabolism and mitochondrial function, which are crucial for the developing brain (Bülow et al., 2022). There are earlier studies that have linked such metabolic changes in autism physiology. These include the genetics (Varga et al., 2018) and functional differences (Giulivi et al., 2010) that contribute to atypical mitochondrial function in autism (Frye, 2020; Rossignol & Frye, 2012). We have also identified pathways related to amino acid metabolism, although with relatively lower confidence. This may nevertheless still prove to be an interesting aspect to investigate further, particularly with the recently growing evidence in this direction (Chen et al., 2022; Tărlungeanu et al., 2016; Zhu et al., 2022).
Even with a relatively larger cohort, our study like several others is limited by the size of the clinical cohort to identify robust and reproducible urine-based biomarkers of autism, if there are any. Furthermore, since urine is an excretory by-product, dietary choices can influence the urine metabolome (Stratakis et al., 2022), both in neurotypical and autistic individuals. Future studies should aim to include this information in their analyses, accompanied with knowledge about the gut-microbiome status of study participants, as this can also be potentially associated with autism status (Taniya et al., 2022). Furthermore, there is a need to replicate such studies in several international cohorts to test the value of identified biomarkers in different populations.
In conclusion, we have identified several urinary metabolites that show nominal association with autism diagnosis status and autistic traits. We have also determined the enrichment of the significant features in metabolic pathways. While our findings highlight no robust urine-based biomarkers for autism, this ever-increasing knowledgebase can serve as signposts for investigations in the future.
Data Availability
The utilised code is available on GitHub (https://github.com/Tammimies-Lab/RATSS-Metabolomics) or available upon request from the corresponding author (kristiina.tammimies{at}ki.se).
Author contributions
Conceptualization: A.A., S.B. and K.T.; Methodology: A.A. and K.T.; Software development: A.A. and K.T.; Validation: A.A. and K.T.; Formal Analysis: A.A.; Investigation: A.A.; Resources: S.B. and K.T.; Data Curation: A.A.; Writing – Original Draft: A.A., F.M. and K.T.; Writing – Review & Editing: A.A., F.M., S.B. and K.T.; Visualization: A.A. and K.T.; Supervision: S.B. and K.T.; Project Administration: A.A. and K.T.; Funding Acquisition: S.B. and K.T.
Declaration of conflicting interests
A.A., F.M., and K.T. declare no competing interests. S.B. declares no direct conflict of interest related to this article. He discloses that he has in the last three years acted as an author, consultant, or lecturer for Medice and Roche. He receives royalties for textbooks and diagnostic tools from Hogrefe, Kohlhammer, UTB and Liber. S.B. is a partner in NeuroSupportSolutions International AB.
Funding
The project was supported by the Swedish Research Council (S.B., and K.T.), Swedish Foundation for Strategic Research (K.T.), the Swedish Brain Foundation – Hjärnfonden (K.T.), the Harald and Greta Jeanssons Foundations (K.T.), Åke Wiberg Foundation (K.T.), Strategic Research Area Neuroscience Stratneuro (K.T.), The Swedish Foundation for International Cooperation in Research and Higher Education STINT (K.T.), and Board of Research at Karolinska Institutet (K.T.). Open access funding provided by Karolinska Institutet.
Supplemental material
Supplemental material for this article is available online.
Supplementary Tables
Table S1. GEE model for metabolites and autism diagnosis. (A) Outcomes from modelling in full cohort. (B) Outcomes from modelling in monozygotic twins subsetted from full cohort. (C) outcome from modelling in dizygotic twins subsetted from full cohort.
Table S2. GEE model for metabolites and autistic traits. (A) Outcomes from modelling in full cohort. (B) Outcomes from modelling in monozygotic twins subsetted from full cohort. (C) outcome from modelling in dizygotic twins subsetted from full cohort.
Table S3. Over representation analysis of significant metabolites. (A) Enriched pathways from GEE Model A and autism diagnosis. (B) Enriched pathways from GEE Model B and autism diagnosis. (C) Enriched pathways from GEE Model A and autistic traits. (D) Enriched pathways for GEE Model B and autistic traits.
Acknowledgements
We would like to thank all twins and parents who have participated in this research. We would also like to thank the RATSS team, especially Anna Pilfalk, Karl Lundin Remnelius, Dr. Johan Isaksson and Dr. Janina Neufeld for their valuable contribution to the work presented in this study. Additionally, we would like thank Prof. Antonio Persico for early discussion on designing the project. We acknowledge the Swedish Twin Registry for data access and urine analysis by the Proteomics and Metabolomics Facility, Department of Ecological and Biological Sciences, University of Tuscia, Viterbo, Italy and Prof. Lello Zolla.
Footnotes
↵* email: kristiina.tammimies{at}ki.se
