Long-term benefit of Microbiota Transfer Therapy on autism symptoms and gut microbiota

Overall, the most substantial improvements observed were on the CARS assessments, which was conducted by a professional evaluator and is less susceptible to placebo-effect²⁰. CARS is a stable and consistent diagnostic tool with high predictive validity²¹ and has been used to evaluate participants before and after therapeutic interventions in multiple studies^20,22,23. For the follow up CARS, the evaluator collected current information based on each question’s unique criteria. After the interview was complete for each question, the evaluator reviewed the information initially collected at baseline and used it for calibrating the final evaluation.

We performed statistical analyses to assess whether improvements in GI and ASD severity were correlated. As shown in Fig. 2c–e, percentage changes in CARS, SRS, and ABC scores were positively correlated with percent changes in GSRS scores (Spearman correlation test, 2-tailed p < 0.005 and r > 0.7), implying that GI relief provided by MTT may ameliorate behavioral severity in children with ASD, or vice versa, or that both may be similarly impacted by another factor. Another GI assessment, DSR, however, showed that there was no significant correlation. Although the direction of the influence is not clear, a potential clinical link between GI and behavior severity is consistent with what previous studies have reported^24,25.

Upon completion of the original MTT treatment, we observed that the unweighted UniFrac distance³⁰ between the gut microbiota of MTT recipients and their corresponding donors was smaller than before treatment, suggesting some engraftment of the donor microbiome into the recipients by MTT¹⁹. Interestingly, two years after the trial, the recipients were as different from the donor microbiome as they were pre-treatment as measured by unweighted UniFrac distance (Fig. 3b, Supplementary Fig. S4b) and several other metrics of community dissimilarity (Supplementary Fig. S5b–e). This suggests that the recipients didn’t retain completely the donated microbiome, but rather retained some features of it such as increased overall diversity, and increase in some important microbes such as Prevotella, while finding a new state.

To the best of our knowledge, long-term follow-up studies are rare for medical treatment of individuals with ASD. In treatments with vancomycin³⁷ or phytochemical sulforaphane³⁸, benefits were lost within two or four weeks, respectively, of the treatments being discontinued. Thus, the long-term benefits observed here two years after MTT stopped are very encouraging, and MTT-driven gut microbiota transformation seems robust and long-lasting for the treatment of ASD. Despite steady and continuous improvement in behaviors over two years, we must underscore that the original clinical trial and current follow-up study are open-label trials without a control for placebo effect. Autism symptoms are relatively stable over time without a major intervention: for example, a trajectory study with 345 children with ASD showed that more than 80% of participants with ASD retained unexpectedly stable core symptoms severity over 8 to 12 years³⁹. The VABS observations indicate that the improvements in adaptive behaviors observed here were substantially more than expected for children with ASD over two years. A limitation of this study is that 12 out of 18 participants made one or more changes to their medications, nutritional supplements, and diets between the end of the original MTT trial and the two-year follow-up since the treatment stopped (Supplementary Table S1). As described in detail in the methods section, participants were asked to rate the perceived effectiveness on GI and ASD symptoms (on a scale of 0–4) caused by changes in medications, diet, or nutritional supplements. Although the scale on the perceived effectiveness is still subjective and difficult to interpret, low scores received (1.1 for GI and 0.8 for ASD symptoms) suggest that these treatments on average could have only “slight effect”. Thus, it appears that most of the changes observed were probably due to the MTT, although we still need follow-up studies to understand whether the improvement by MTT were solely from vancomycin, MoviPrep, Prilosec, Standardized Human Gut Microbiota (SHGM), or a combination of these individual factors. For example, some participants in our study could have GI symptoms that were acid-peptic in nature, and their improvements on GI symptoms might be solely attributed to the administration of stomach-acid suppressant (Prilosec)⁴⁰. We hypothesize that MTT may also be beneficial for children with ASD who do not have obvious GI symptoms but have low diversity of gut bacteria, as our previous study²⁶ found that most children with ASD had low gut bacterial diversity, regardless of whether they have GI problems.

Here, we also would like to address a potential study limitation interpreting the improvement on GI symptoms after MTT, since these heterogeneous GI symptoms could reflect a wide range of underlying etiological GI pathologies. Although we reviewed medical histories to exclude children with known gastrointestinal diagnoses (such as ulcerative colitis, Crohn’s Disease, celiac disease, eosinophilic gastritis, or similar conditions)¹⁹, we did not conduct additional GI diagnostic evaluations, which is a limitation of this study. Thus, we want to underscore need of follow-up studies embracing more thorough examination of participants’ GI pathologies in order to better understand effectiveness of MTT.

The protocol for the original treatment study was approved by FDA (Investigational new drug number 15886) and the Institutional Review Board of Arizona State University (ASU IRB Protocol #: 00001053), as described in Kang et al.¹⁹. All research methods were performed in accordance with the relevant guidelines and regulations. For the follow-up study described in this paper, a new IRB protocol was approved by the IRB of Arizona State University (ASU IRB Protocol # 00004890). The follow-up study was only open to the 18 children with ASD who participated in the original Phase 1 trial study, and no further inclusion or exclusion criteria was applied to what we employed in the original trial¹⁹. For the consent process, we contacted past participants and their parents and explained the study in detail by phone and/or email, and provided them with a copy of the parent permission form and the child assent form. After the informed consent process, all 18 families agreed to participate and signed written parent permission and assent forms. We maintained confidentiality of participants results by de-identifying all samples for the entire analyses. The participant’s name and identifiers were removed and not used in all sections of the manuscript, including supplementary information. The trial was registered at the ClinicalTrials.gov (NCT02504554↗) on March 30, 2015.

The purpose of the study was to conduct a 2-year follow-up on children with ASD who participated in the original study and to determine long-term safety and efficacy of the Microbial Transfer Therapy (MTT). In brief, the previous study was designed to perform an open-label clinical trial for children with ASD and moderate to severe GI problems (ages 7–17 years)¹⁹. Eighteen children were enrolled in the previous study and treated with a combination of vancomycin (to reduce pathogenic bacteria) for 2 weeks, MoviPrep (a bowel cleanse to remove vancomycin and remaining bacteria), Prilosec (a proton pump inhibitor in order to reduce stomach acidity and to increase the survival rate of high and maintenance doses of SHGM), and Standardized Human Gut Microbiota (gut bacteria from healthy donors) for 7–8 weeks⁴¹. Participants were monitored during the 10 weeks of treatment, and again 8 weeks after MTT stopped. There were no major adverse effects, and all 18 participants completed the study. Most of the children had substantial reduction of their GI and ASD symptoms, and those improvements remained at the follow-up at 8 weeks after the treatment ended. Further information on the previous study design and outcomes are described in Kang et al.¹⁹. In this current study, we followed up past participants approximately 2 years since the treatment stopped, and requested parents to collect one stool sample from their child in order to investigate changes in bacterial composition. After the study ended, 6 participants did not make any changes in their treatments for the next two years, and 12 participants made one or more changes (Supplementary Table S1). The changes (and the number of participants making those changes) included dietary changes (gluten-free/casein-free diet (1), ketogenic diet (1), and removing eggs(1)), GI treatments (magnesium citrate (2), digestive enzymes (1), dulcolax (1), enteragam (1), probiotics (2)), medication changes (Abilify (1), birth control (1), IVIG (1), Lamictal (1), lithium (1), Tenex (1), thryroid (1), and Vyvanse (1), and nutritional supplements (amino acids (1), multivitamins (2), Bravo yogurt (1), citruline (1), fish oil (1), GABA (1), l-theanine (1), magnesium (1), Oraimmune (1), and Rerum (1). Participants were asked to rate the perceived effectiveness of each treatment on a scale of 0–4 for both GI and ASD symptoms, where 0 = “No effect”, 1 = “Slight benefit”, 2 = “Moderate benefit”, 3 = “Good benefit”, and 4 = “Great benefit”.

To be consistent with the previous assessments¹⁹, we evaluated GI symptoms using a revised version of Gastrointestinal Symptom Rating Scale (GSRS) that employs 7-point Likert scale⁴² in 5 domains: Abdominal Pain, Reflux, Indigestion, Diarrhea, and Constipation. We also collected Daily Stool Records (DSR) that rates the stool using the Bristol Stool Form scale (1 = very hard, 7 = liquid) over 14 days. For the assessments of autism and related symptoms, we also employed the same assessments that we used in the previous study including the Parent Global Impressions–III (PGI-III)⁴³, the Childhood Autism Rating Scale (CARS), the Aberrant Behavior Checklist (ABC), the Social Responsiveness Scale (SRS), and the Vineland Adaptive Behavior Scale II (VABS-II). The CARS evaluation was conducted by the same professional evaluator who conducted the previous CARS evaluation, and parents assessed the GSRS, DSR, PGI-III, ABC, SRS, and VABS-II at approximately 2 years after the therapy stopped.

In order to be consistent with the previous work in Kang et al.¹⁹, we employed the same DNA extraction assay, library preparation, and sequencing methods. We extracted microbial genomic DNA from stool samples using a PowerSoil^® DNA Isolation Kit (Mobio Carlsbard, CA) and constructed 16S rRNA library for MiSeq Illumina platform according to the protocol from Earth Microbiome Project (http://www.earthmicrobiome.org/protocols-and-standards/16s/↗). The barcoded primer set 515f -806r was used for pair-ended sequencing to target the 16 s rRNA V4 region⁴⁴. Library preparation and sequencing work were performed at the Microbiome Analysis Laboratory in the Biodesign Institute of Arizona State University (http://krajmalnik.environmentalbiotechnology.org/microbiome-lab.html↗).

The 16S rRNA gene data from our previous sequencing run was generated from four Illumina MiSeq runs, and the follow-up data (new to this study) was sequenced on a single MiSeq run. To ensure consistent processing across all of these runs, and to enable the use of the most recent bioinformatics methods for microbiome analysis, we processed all five of these sequencing runs together in a single QIIME 2 analysis⁴⁵.

Analyses were performed on the forward reads only, to retain as many sequences as possible (i.e., to not discard sequences where forward and reverse reads could not be joined). Sequence quality control was performed with two different processing pipelines, both implemented in QIIME 2.

The first processing pipeline used DADA2⁴⁶ via the q2-dada2 QIIME 2 plugin (we’ll refer to this as the DADA2 pipeline). In the DADA2 pipeline, the first 10 bases were trimmed from reads and reads were truncated at 150 bases prior to processing with the dada2-single method, as recommended by the DADA2 authors. All other DADA2 parameters were left with their default settings. DADA2 was run on a per-sequencing run basis using the same parameters for all runs of DADA2. The resulting feature tables were merged, such that we had one master “feature table” containing counts of features on a per sample basis, where our features were unique 16S rRNA gene sequence variants. The “feature sequence” that defined each feature was the unique 16S rRNA gene sequence variant. Microbiome analysis results based on this processing pipeline are presented in Figs 3, 4 and Supplementary Figs S4, S5.

The second processing pipeline used the quality filtering approach described in Bokulich et al.⁴⁷ as implemented in the q2-quality-filter plugin using a min-quality of 20 (all other parameters were defaults). We will refer to this pipeline as the Q20 pipeline. Sequence reads were subsequently clustered into 97% OTUs using open-reference OTU picking as implemented in the q2-vsearch plugin (which uses vsearch⁴⁸ for clustering). Again, we had one master “feature table” containing counts of features on a per sample basis, where in this case our features were 97% OTUs. The “feature sequence” that defined each feature was the OTU centroid sequence. As an additional quality control step, features that were only observed in a single sample and features that could not be assigned to a phylum against the Silva database⁴⁹ (described further below) were filtered from the feature table. Microbiome analysis results based on this processing pipeline are presented in Supplementary Fig. S6.

Following both the DADA2 and Q20 pipelines, feature sequences were aligned with mafft⁵⁰ (wrapped in the q2-alignment plugin), high entropy positions were filtered from the resulting alignment⁵¹, an unrooted tree was constructed with FastTree⁵² (wrapped in the q2-phylogeny plugin), and the tree was rooted using midpoint rooting. Taxonomy was assigned to each feature sequence against the Silva 119 database using a Naïve Bayes classifier implemented in the q2-feature-classifier plugin⁵³. Feature tables were rarefied to 5,298 sequences per sample, and diversity metrics were computed using the q2-diversity plugin.

For complete details on the bioinformatics methods, including the versions of all software and dependencies, and all commands and parameter settings, readers can view the provenance data contained in the Supplementary File S1. This information can be viewed visually at https://view.qiime2.org↗.

The specific values of the microbiome richness and distance metrics differ between the DADA2 and Q20 bioinformatics pipelines. For example, Faith’s Phylogenetic Diversity is considerably lower with the DADA2 pipeline, likely because it does a much better job at raw sequence quality control (as described in the original DADA2 paper). Despite this, it is encouraging to observe that the different quality control and feature definition pipelines illustrate the same trends in microbiome composition and richness.

We assumed the data as non-normally distributed because of a relatively small sample size, and performed nonparametric analyses of Mann-Whitney U-test, Wilcoxon signed-rank test, and Spearman correlation test. We reported all p values from two-tailed tests, and accepted p values lower than 0.05 as significant. For the boxes in Figs 1, 3 and 4, the cross line in the box indicates median values and the bottom and top edges of the box indicate the 25^th and 75^th percentiles of the sample.

Supplementary Figures S1-S6 Supplementary Table S1 Supplementary Table S2 Supplementary File S1