Randomized control trial layout
This 10-week pilot study followed a standard double-blind, randomized control trial (RCT) design, where participants were randomized to either placebo or control. The cohort consisted of 18 consenting children diagnosed with ADHD, who were a sub-cohort of participants from a larger study that investigated the effects of dietary micronutrient supplementation on ADHD outcomes36. Faecal microbiome analysis was not intended at the initialisation of this large study36, and this sub-cohort represents the later male recruits. There was no a priori selection of candidates for participation in the microbiome analysis (Fig. 1) and the randomization was not influenced.
Study design and RCT layout. The 10-week pilot study was performed with a sub-sample of participants from a larger study36. Study design followed a standard double-blind, randomized control trial (RCT), and there was no a priori selection of candidates for participation in the microbiome analysis.
The placebo group contained eight participants and the micronutrient group contained ten participants. One participant in the placebo group had taken oral antibiotics during the RCT and was subsequently excluded at the data analysis stage, giving a final total of 34 samples (Fig. 1). For each participant, a pre-RCT, and a post-RCT faecal sample was collected, and microbiome sequencing analysis was performed on DNA extracted from each sample, giving a total of 36 samples (Supplementary File 1).
All participants were under supervised administration of capsules, where the micronutrient group received a formulation containing a blend of vitamins, minerals, amino acids and antioxidants (see Supplementary Table 1 for ingredients of both treatment and placebo capsules). Participants began by taking one capsule, three times each day, increasing the dose by three capsules every two days up to a target dose of 12 capsules per day: four taken at three different intervals. Further information regarding the dosing, randomization and blinding procedures is available in Rucklidge et al.36.
Due to the RCT layout of this study, the dataset consisted of paired samples for each individual that had been collected pre and post-RCT. Where possible, the data were analyzed using a paired-sample approach to appropriately incorporate the baseline data for each individual. This ensured that the study specifically investigated changes that occurred during the trial period. Significant changes were only considered true, if they were specific to the micronutrient-treatment, or if the magnitude of change was in the opposite direction to the placebo. In addition to the below methods, repeated measures correlation was performed using the R package, rmcorr37, and power analysis was performed using the R package, PWR38. Correlation analysis of the abundance data was performed within R, using the “Pearson” correlation coefficient.
Efficacy and safety assessments
All participants were monitored by a clinical psychologist or psychology graduate student, under the supervision of a psychologist. This was performed at the screening visit, baseline, and weeks 2, 4, 6, 8 and 10 (or end of study), via face-to-face meetings or via phone contact. At each visit, the Children’s Global Assessment Scale (CGAS)39 and ADHD Rating Scale IV (ADHD-RS-IV) – clinician40,41 were completed. The CGAS was used by the clinician to assess the overall level of the children’s functioning based on all the information gathered since the last visit. Scoring is based on a single numerical scale from 1 to 100, where the higher score is indicative of better functioning. At pre and post-RCT time-points, the clinician also completed the ADHD-RS-IV, which contains 18 items directly linked to DSM-IV diagnostic criteria for ADHD and provides a total score and two subscale scores for inattention and hyperactivity/impulsivity, assessing ADHD symptoms based on frequency (0 ‘never or rarely’ to 3 ‘very often’). The clinician took into account observations from visits and formal cognitive testing, information from others, as well as parent report in determining ratings. However, frequency of behaviours was the main focus of the rating, considering how often the behaviours were present.
At baseline and study completion, hematological variables, biochemical variables, thyroid function, prolactin, fasting glucose, homocysteine, iron, zinc, vitamin D, vitamin B12, copper, blood pressure, height and weight were also recorded. To assess the child’s dietary patterns, including consumption of fruit, vegetables, breakfast, and fast foods, a brief diet intake questionnaire was performed at baseline and end of RCT with a higher score indicative of a healthier eating pattern (modified from Baker, et al.42). Demographic variables, including participant’s ethnicity and parents’ occupation were also collected at baseline.
Informed consent and ethics
Written informed consent was obtained from all of the participants’ parents or legal guardians and assent was obtained from the participants. The trial was prospectively registered with the Australia and New Zealand Clinical Trial Registry ACTRN12613000896774, on the 12/08/2013. The exploratory nature of the study, as well as other treatment options for ADHD, was explained to participants and their parents prior to enrolling. This study was approved by the University of Canterbury (New Zealand) and national institutional review boards. Ethical approval for this study was approved by the Southern Health and Disability Ethics Committee (New Zealand), Ethics ref: 13/STH/45/AM05, and all research was performed in accordance with the relevant guidelines.
Sample collection and storage
Fresh stool samples were collected at baseline and post RCT (10 weeks) using the OmnigeneGut faecal collection system (DNA Genotek, Ottawa, Canada), according to the manufacturer’s specifications. Using this system, homogenised samples are stable for 60 days at ambient temperature, however, all samples were stored at −4 °C for a maximum of 14 days, before aliquoting and storage at −80 °C.
DNA Isolation and sequencing
Total DNA was extracted from stool samples using the NucleoSpin DNA Stool isolation kit by Macherey-Nagel (Germany, Duren), according to the manufacturer’s recommendations for human stool samples. Extracted DNA was stored at −20 °C until PCR amplification and Illumina MiSeq DNA sequencing. PCR amplification and single-end DNA sequencing of the V3/V4 regions of the 16S rRNA gene was outsourced to New Zealand Genomics Limited (Dunedin, New Zealand) and performed according to their protocols.
Demultiplexed and pre-processed sequence reads were supplied by New Zealand Genomics Limited and these were imported into the Quantitative Insights Into Microbial Ecology (QIIME2, version 2017.12), python-based pipeline using CASAVA 1.8. Barcode and primer removal, quality control, amplicon sequence data correction, phiX filtering, and dereplication was verified using the DADA2 software package43. The total sequence length was truncated to 220 bp, and 13 bp of low quality data were trimmed from the start of the sequence. This processing was separately performed on the V3 and V4 sequence runs. Feature tables and representative sequence files were then merged for downstream analysis, within QIIME2.
Taxonomic classification was assigned using the GreenGenes database, specific for the V3/V4 16S region (version 13.8). All taxonomic classifications were implemented within QIIME2 and assigned using the naïve Bayesian algorithm, developed for sklearn classifier. For phylogenetic diversity analysis, sequences were aligned using the MAFFT44 programme plugin, and then filtered to remove highly variable positions. Reads that were unassigned at the Kingdom level were removed from the OTU table, as were OTUs that were observed less than 100 times (Supplementary File 2). The final taxonomic classification was made against the filtered feature table, which had 4474 distinct features (Supplementary File 3).
Alpha and Beta diversity were calculated using the q2-diversity plugin and included Faith’s Phylogenetic Diversity, and weighted and unweighted Unifrac distances. The feature table was rarefied to a sampling depth of 273,465, which retained 50.18% (9,297,810) of sequences and all samples (34). This sampling depth was selected as it was approaching the maximum depth which retained all samples for our analysis. Principal coordinate analysis (PCoA) was used to investigate the similarities between bacterial communities, based on treatment using Unweighted and Weighted Fast Unifrac methods. Pairwise distance and pairwise differences in alpha diversity values (observed operational taxonomic units) were calculated within QIIME2, using the q2-longitudinal plugin.
Computing species abundance using the q2-ANCOM plugin
Species abundance was assessed using the QIIME2 ANCOM plugin45, consecutively at each taxonomic level to detect significant changes in abundance between each of the four groups (micronutrient pre-RCT, micronutrient post-RCT, placebo pre-RCT and placebo post-RCT). Relative frequencies at each phylogenetic level were calculated by ANCOM, from the DADA2 feature table for assigned sequence variant.
Analysing paired differences in abundance using the q2-longitudinal plugin
Longitudinal analysis of feature data from paired samples (pre and post-RCT) was performed following the methods outlined by Bokulich et al.46. Briefly, relative abundance was manually calculated from the DADA2 generated (non-rarefied) feature table, filtered to contain Actinobacteria, and consecutively collapsed at each phylogenetic level (Supplementary File 4 and 5). The pairwise difference between baseline and post-RCT samples for each phylogenetic level was then compared for the micronutrient and placebo groups. This method measures within group pairwise difference using Wilcoxon signed-rank tests, and pairwise group comparison tests using Kruskal Wallis.
Functional predictions of metagenomics profiles were computed using two common approaches. The first approach involved analysis with the R package Tax4Fun47, and the second approach involved analysis with the package, Phylogenetic Investigation of Communities by Reconstruction of Unobserved States (PICRUSt)48, via the Galaxy platform49. For analysis with Tax4Fun, a QIIME2 closed reference feature table was generated using the QIIME2 vsearch plugin at 97% similarity to the SILVA50,51 database release version 132. For analysis using PICRUSt, the feature table was generated at 97% similarity to the Green Genes database version 13.5.
Change in the abundance of metagenomic pathway profiles generated by each approach was then calculated by subtracting the baseline measurement from the post-RCT measurement, for each individual. Differentially abundant pathways that correlated with treatment were identified using a linear regression model within the Limma package52 that accounted for treatment, age, and ethnicity.