Analysis of Fecal Biomarkers in Children with Autism Spectrum Disorder and Gastrointestinal Symptoms

Analysis of Fecal Biomarkers in Children with Autism Spectrum Disorder and Gastrointestinal Symptoms

Authors: Kathryn Hanagan^1,2, Juergen Hahn^2,3,4

¹Department of Computer Science, Purdue University, West Lafayette IN 47906

²Center for Biotechnology and Interdisciplinary Studies, Rensselaer Polytechnic Institute, Troy NY 12180 

³Department of Biomedical Engineering, Rensselaer Polytechnic Institute, Troy NY 12180 

⁴Department of Chemical and Biological Engineering, Rensselaer Polytechnic Institute, Troy NY 12180

Introduction: Autism spectrum disorder (ASD) encompasses a large group of early onset neurological conditions that result in impairments in social behavior and communication, which are estimated to affect 1 in 59 children in the United States [1]. Early diagnosis is extremely important to the treatment of individuals with ASD, since early diagnosis can lead to early intervention which may allow for the improvement of neural connectivity while brain plasticity is still high [2]. One current limitation on early intervention is that the diagnosis of ASD is only based upon observations. Earlier diagnosis may be possible by developing a quantifiable biomarker-based test for ASD. The high prevalence of gastrointestinal (GI) symptoms co-occurring with ASD and significant evidence for alterations in the gut microbiome show promise as a method for biomarker discovery [3]. Thus, this study attempts to develop a multivariate model of fecal metabolites that accurately classifies between ASD and typically developing (TD) children.

Materials and Methods: The data come from a study performed by Kang et al. [4] which examined the effects of gut microbiome transfer therapy (MTT) on children with ASD and severe GI problems. The study consisted of 38 children, ages 7-16 years: 18 professionally diagnosed with ASD and 20 considered to be typically developing. For 10 weeks, the ASD group was undergoing MTT and the concentrations of 669 fecal metabolites were measured at several points in time. Observations continued for the next 8 weeks following the treatment. The purpose of this statistical analysis was to generate a multivariate metabolite model that maximally separates the TD and pre-treatment ASD group utilizing the initial stool samples. To ensure continuous distribution of values, metabolites were removed if they had more than 40% of measurements below the detection limit over all participants at week 0, resulting in 583 useful metabolites. Univariate analysis was performed on each metabolite by computing the Area Under the Receiver Operator Characteristic curve (AUROC) for classifying between the ASD and TD cohorts at week 0. Subsequently, 60 metabolites had AUROC values of at least 0.65. To ensure the robustness of the model, either a Wilcoxon signed-rank test or a paired t-test was performed on each metabolite to determine if the TD cohort changed significantly from week 0 to 18. In total, 48 metabolites had a p-value > 0.1, remaining relatively unchanged over the course of the study. Fisher Discriminant Analysis (FDA) was performed on every combination of 2, 3, 4, or 5 metabolites from this subset to determine which were the best for classification. “Leave-one-out” (LOO) cross-validation was performed on the best models to assess their validity.

Results and Discussion: The optimized FDA models consisting of 4 and 5 metabolites achieved AUROC values of 0.95 and 0.98, respectively. During cross-validation, the Type II error rate (β) was modulated between 0.01, 0.05, 0.10, and 0.20. With β=0.20, a classification threshold is determined, resulting in a TPR=0.89 and TNR=0.80 for the 4-metabolite model. The 5-metabolite model with β=0.10 achieves a TPR=0.94 and TNR=0.75. The 4-model consists of sphingosine, indole, adenine, and theobromine, while the 5-model consists of indole, adenine, N-stearoyl-sphingosine (d18:1/18:0)*, N-acetylsphingosine, and theobromine.

Conclusions: The effectiveness of using multivariate techniques to distinguish between fecal metabolite profiles of ASD and TD groups was demonstrated by this study. As few as 4 or 5 metabolites can be used for classifying between the ASD and TD cohorts at their baseline measurements prior to MTT. Further investigation is needed to determine if these results apply to the general population, since this study is limited by a small number of participants and a somewhat inconsistent control group. Still, the significant differences may point to possible metabolite/cellular mechanisms that may be implicated in ASD. Furthermore, this work posits an avenue for the possibility of deriving a biomarker-based test for ASD in children at young ages.

References: [1] J. Baio, et al, MMWR Morb Mortal Wkly Rep, 2018 Nov 16, 67(45): 1279. [2] Pierce, K.et al., Journal of Pediatrics (2016), 176, 182–194. [3] Holingue et al. Autism res (2018), 24-26 [4] Kang, D. et al., Microbiome (2017) 5:10