Bacterial gut microbiome differences in adults with ADHD and in children with ADHD on psychostimulant medication

Recent evidence suggests that there is a link between neurodevelopmental disorders, such as attention-deficit hyperactivity disorder (ADHD)


Introduction
Attention-deficit hyperactivity disorder (ADHD) is a common neuropsychiatric disorder characterized by persistent features of inattention and/or hyperactivity-impulsivity, leading to functional impairments in areas such as social relations, family settings and academic/occupational functioning (Austerman, 2015;Faraone et al., 2015). ADHD has a prevalence of 7.2% among children and 5% among adults (Thomas et al., 2015) and is often associated with emotion dysregulation (Beheshti et al., 2020;Craig B.H. Surman et al., 2011;Reimherr et al., 2005). In addition, ADHD is highly comorbid with other neurodevelopmental disorders, such as autism spectrum disorder (ASD) (Antshel et al., 2016;Jensen & Steinhausen, 2015). Psychostimulant medications are generally well-tolerated and reduce the core symptoms of ADHD (Riera et al., 2017). However, in 10-30% of patients, treatment is discontinued due to non-response or the emergence of adverse events (Mohammadi & Akhondzadeh, 2007;Storebø et al., 2018).
In the past years, there has been increased interest in the relationship between neurodevelopmental disorders, such as ADHD and ASD, and the gut microbiome, a complex dynamic ecosystem with trillions of microbes colonizing the gastrointestinal (GI) tract . The increasing interest is partly due to an overrepresentation of GI-symptoms in ADHD and ASD (McKeown et al., 2013;Vargason et al., 2019), and partly due to studies in rodent models suggesting microbiome-mediated behavioral changes related to ASD or ADHD (Hsiao et al., 2013;Mathee et al., 2020;Sharon et al., 2019;Tengeler et al., 2020). Specifically, fecal microbiota transplanted from ADHD patients into mice were found to result in changes not only in fecal microbial composition but also in behavior, brain structure and brain function (Tengeler et al., 2020). In addition, studies have proposed a link between the gut microbiome composition and neurodevelopment in very early childhood (Cowan et al., 2020;Kelsey et al., 2021;Vaher et al., 2022), including a small randomized trial suggesting that earlylife probiotic supplementation could reduce the risk for developing ADHD and ASD later in life (Pärtty et al., 2015). However, the gut microbiome difference of ADHD patients compared to controls is not well-characterized, e.g., there is a lack of taxa-mediated molecular investigations, and there is also a need for larger studies (Hiergeist et al., 2020;Mathee et al., 2020). Specifically, although several studies have examined the fecal gut microbiome in ADHD patients, the findings to date have been conflicting where some studies reported differences compared to controls, and others did not (Aarts et al., 2017;Jiang et al., 2018;Prehn-Kristensen et al., 2018;Richarte et al., 2021;Szopinska-Tokov et al., 2020;Wan et al., 2020;. This lack of consistent findings could be attributed to (i) studies being underpowered, (ii) analyses not taking into account reported putative fecal microbiome confounders, such as diet (Yap et al., 2021), stool consistency (Vandeputte et al., 2016a), antibiotic use (Bokulich et al., 2016) and other microbiome altering medications, (iii) previously reported methodological and /or technical biases (Hiergeist et al., 2020), and (iv) incorrect use of statistical methods, which are not suited for compositional data (Gloor et al., 2017;Morton et al., 2019). In particular, the analysis of differential abundance has been challenging, with several studies using statistical methods that are not controlling for the false discovery rate (FDR) in microbiome data, or not considering the differences in microbial load between samples (Lin & Peddada, 2020;Morton et al., 2019). Thus, further studies are needed to examine the gut microbiome in ADHD using large sample sizes and well-characterized participants with information on putative confounding factors.
To this end, we undertook the largest, to our knowledge, fecal shotgun metagenomics study in ADHD, including 147 ADHD patients (84 adults and 63 children) and 52 adult controls. All individuals had been extensively clinically characterized and had provided information on e.g., diet, dietary supplements, Bristol Stool Scale (BSS) scores, delivery mode, and intake of antibiotic drugs or other medications known to affect the fecal microbiota. To do a more proper case control comparison we adjusted for these confounding variables. In the present study we investigated (i) the relationships between clinical variables and bacterial strain diversities, (ii) the differences in bacterial strain (taxonomic) and bacterial gene (functional) diversities between adult ADHD patients and adult controls, and (iii) the strains and enzymeencoding bacterial genes that are differentially abundant between adult ADHD patients and adult controls. Moreover, we examined the impact of psychostimulant medication on the fecal microbiome and the associations with plasma levels of inflammatory markers and shortchain fatty acids (SCFAs).

Participants
Patient recruitment was carried out between January 2016 and June 2018 via preselected psychiatric outpatient clinics and through advertisements in a local newspaper in Stockholm County, Sweden. All participants were living in Stockholm County. Eligibility criteria for participation were the following: confirmed ADHD-diagnosis (International statistical Classification of Disease 10th revision (ICD-10) or DSM-V), stable pharmacological treatment if there was any (no changes in medication during the last 4 weeks), age of 5-55 years, and ability to read Swedish. Exclusion criteria were autism diagnosis, diabetes diagnosis, celiac disease, GI-disorder other than irritable bowel syndrome (IBS), and antibiotic intake during the last six weeks. Healthy controls in this study had no ADHD diagnosis but fulfilled the other above-listed criteria, and were house-hold family members of, or nonrelated to the recruited patients. Blood was collected and was used to measure six SCFAs (i.e., formic, acetic, propionic, butyric, succinic and isovaleric acid) and two inflammatory markers (i.e., soluble intercellular adhesion molecule-1 (sICAM-1) and soluble vascular cell adhesion molecule-1 (sVCAM-1)). Further information about the analyses of inflammatory markers and SCFAs in plasma can be found in Supplementary Methods. The participants were interviewed regarding delivery route, breastfeeding, body mass index (BMI), pharmacological treatment, antibiotic use, and other microbiome-affecting drugs such as proton pump inhibitors and laxatives. Self-reported psychiatric scales, as well as diet and GI-symptom questionnaires, were also completed by the participants. The interviews and blood samplings were conducted by experienced research nurses in psychiatric care. A more detailed description of the participants has been published elsewhere (Skott et al., 2020), and information about 'Patient recruitment' and 'Questionnaires and scoring' can be found in the Supplementary Methods. The data are collected at baseline in the randomized placebo-controlled trial ISRCTN57795429 (https://doi.org/10.1186/ISRCTN57795429).

Stool collection, DNA extraction and sequencing
Stool samples were self-collected by the patients in their homes using the OMNIgene gut kit (DNAgenotek, Ottawa, Canada), stabilizing the microbial DNA, and were then handed over to the research nurses. In total, there were N = 307 fecal samples collected from patients and controls. However, samples from patients using medications known to affect the microbiome were removed according to the following exclusion criteria: 1) use of antibiotics during the last 2 years, and 2) use of proton pump inhibitors or laxatives during the last 3 months. Following these exclusion criteria, N = 209 fecal samples remained and were sent for analysis (see suppl. Fig. S1). All samples were kept at − 80C until analyses, including one freeze-thaw cycle. A total of 400 μl of OM-LQR/ P-190 liquefaction (DNAgenotek, Ottawa, Canada) was added to fecal samples that had clumps > 1 cm in diameter after being vortexed vigorously for 45-60 s. DNA extraction, DNA sequencing, and taxonomic/functional profilings were performed by Diversigen Inc. using the shallow shotgun metagenomic sequencing service Boostershot® (Minneapolis, USA). For DNA extraction the PowerSoil Pro extraction kit was used (Qiagen, Germany) and sequencing was performed on a NextSeq 1x150 flow cell (Illumina, USA). The samples were sequenced in two separate rounds (1st round: June 2019: 128 ADHD samples, 2nd round December 2019: 23 ADHD samples and 53 control samples). The sequencing depth was greater in the 2nd round compared to the 1st round, p wilcoxon = 2.2e -16 , IQR 2nd batch = 4.1 M (3.7 M-5.0 M), IQR 1st batch = 2.5 M (2.1 M -2.9 M). Thus, in all statistical analyses, we adjusted for the library size (total count). Two samples from the 1st round were also sequenced in the 2nd round, and clustered with themselves on a PCoA plot based on Bray-Curtis explaining 99.7% of the variance.

Taxonomic and functional annotation
Taxonomic: DNA sequences were aligned in 2019 to Diversigen's (previously CoreBiome) Venti database, a curated database created in 2017 containing all representative genomes in RefSeq for bacteria with additional manually curated strains (in total n = 19,840). Alignments were performed with 97% identity against all reference genomes. Every input sequence was compared to every reference sequence in the Venti using fully gapped alignment with BURST (Al-Ghalith & Knights, 2020). Ties were broken by minimizing the overall number of unique Operational Taxonomic Units (OTUs). For taxonomic assignment, each input sequence was assigned the lowest common ancestor that was consistent across, at least, 80% of all reference sequences tied for best hit. Functional: DNA sequences were aligned to a gene database derived from the Venti database. Using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database (https://www.genome.jp/kegg/ko.html), KEGG Orthologs (KOs) were observed directly using alignment at 97% identity against the genes in the Venti database. Ortholog and paralog groups, based on sequence similarity scores, were grouped into KOs linking these genes to high-level functions (Kanehisa, 2017;Kanehisa et al., 2016). In our study, KOs that represent functional enzyme orthologs were used in the functional analyses and were labelled 'bacterial genes encoding enzymes', 'enzyme-encoding bacterial genes' or 'functional enzyme'.

Pre-processing of annotated data
The count tables for beta and alpha diversity were rarefied, with the lowest depth on 500 000 counts for the taxonomic data and 100 000 counts for the functional data, using Quantitative Insights Into Microbial Ecology (QIIME 2) version 2020.6 (Bolyen et al., 2019). This was decided according to a generated rarefaction graph. For the functional rarefaction, n = 4 fecal samples had an inadequate library size, i.e., below 100 000 counts, and were excluded from statistical analyses. In the taxonomical analysis, n = 4 additional individuals were removed due to having less than 500 000 counts (see Supplementary Fig. S1 for more information).

Statistical methods
Due to the different read depths in the two sequencing rounds, the total count of strains/enzymes for each individual (library size, microbial load) was always used as a covariate in all statistical analyses, except for in the Analysis of Compositions of Microbiome with Bias Correction (ANCOM-BC) since this method takes the library size into consideration.
Statistical significance was set at α = 0.05 and was corrected, when appropriate, for multiple comparisons using the Benjamini-Hochberg (BH) false discovery rate (FDR) or the Holm-Bonferroni correction (a corrected p-value is written as p FDR if corrected by FDR). In the Results section, we report p-values below 0.1, and for differences at p ≤ 0.05 we also provide the standardized beta and/or partial R 2 . All statistical analyses were performed using the R Studio version 1.4.1106 using the packages vegan, LmPerm, qiime2R, biomformat, ANCOMBC (Lin & Peddada, 2020), phyloseq (McMurdie & Holmes, 2013), dplyr and ggplot2 (Wickham, 2016).
Both the Bristol Stool Scale (BSS) score and the diet intake scores were created using self-reported retrospective questionnaires. In short, the diet data were created from a questionnaire consisting of intake frequencies of 57 food items. The questionnaire data were translated from energy (kcal/week) per food item for each individual into 12 core/ non-core food groups with energy percentages of the total energy intake for each individual, as previously described (Yap et al., 2021). To capture how diverse the diet was, the richness and evenness measure Shannon was applied on the data presented by the 12 core/non-core food groups. In addition, a Principal Component Analysis (PCA), based on the 12 core/non-core food groups, was carried out to capture the diet intake and the three principal component (PC) axes capturing the most variation (i.e., dietary axes PC1, PC2 and PC3) were used in this study. However, since mainly PC1 was found to be significantly associated with baseline beta diversity in adults with ADHD, only PC1 was used for case-control adjustments on beta diversity. More information on the diet score generation and the loadings of the PCs can be found in the Supplementary Methods. In addition, information about how many participants that had information on each clinical variable can be found in Supplementary Table S1.

Diversity analysis
The beta diversity distances Jaccard, Bray-Curtis and Aitchison, and the alpha diversity metrices 'observed number of features', Shannon, and Pielou's evenness, were generated using QIIME 2 2020.6 (Bolyen et al., 2019). In the analysis of the influence of variables on fecal strain beta diversity in ADHD patients (Fig. 1B), and in the case-control analyses, Adonis was used, which is a non-parametric Permutational Multivariate Analysis of Variance Using Distance Matrices (Anderson, 2001). Total count was always adjusted for in these analyses and in the comparison between ADHD patients and healthy controls, Adonis was used with additional adjustments for age, sex, dietary axis PC1, diet diversity, BSS, and BMI (see further information in Suppl. Methods). Number of permutations used in Adonis was 999. When analyzing alpha diversity, as well as plasma levels of vascular inflammation markers and SCFAs, a linear model was used. If the data were not normally distributed, a linear model with permutation tests from the R-package lmPerm was used.

Differential abundance analysis
To evaluate differential abundances between adult ADHD patients and controls, we used ANCOM-BC that was developed for microbiome data and addresses the issue of unequal sampling fractions (microbial load) (Lin & Peddada, 2020). ANCOM-BC uses a sample-specific offset term in an observed abundance-based linear regression model to estimate the unknown sampling fractions and, through that, aims to correct for the microbial load between samples (see further information in Suppl. Methods).
Preprocessing using ANCOM-BC includes removal of features (enzymes or strains) that had counts in less than 10 samples per group (ADHD/controls) since the original authors stated that the FDR is not controlled for when using too few samples (Lin & Peddada, 2020). In addition, features with a median count less than 2 were removed (see Supplementary Table S3 and Supplementary Methods for more information). To correct for multiple testing in ANCOM-BC, Holm-Bonferroni correction was used in accordance with the recommendations by the original authors (Lin & Peddada, 2020). For features that were significantly different between groups, the bias-adjusted feature abundance was regressed against scales measuring ADHD symptoms and functionality, emotion regulation and ASD-symptoms. In the differential abundance analyses investigating use of psychostimulants in ADHD children, the bias-adjusted feature abundance of the significant features was regressed against scales measuring ADHD symptoms and functionality, ASD-symptoms, plasma levels of SCFAs, and the inflammatory markers sICAM-1 and sVCAM-1.

Study cohort characteristics
The patient cohort has been described in more detail elsewhere (Yang et al., 2020). Briefly, for the purposes of the present study, fecal samples from 147 patients diagnosed with ADHD (n children = 63, n adult = 84) and 52 adult controls (n family = 20, n non-related = 32) were collected and analyzed (Table 1, Fig. 1A). The pre-processing step was performed separately for the taxonomic and the functional analyses and samples with too low sequencing depth were removed (see Methods and Supplementary Fig. S1). Thus, the functional analysis was performed on a slightly larger sample compared to the taxonomical analysis (Supplementary Fig. S1). Demographic and clinical characteristics of the subjects are shown in Table 1. Adult ADHD patients experienced GI-pain significantly more compared to controls. However, the stool consistency score measured by BSS did not differ significantly between adult ADHD patients and controls. There were also no significant differences between adult ADHD patients and controls in terms of diet intake (measured by dietary PCA based on percent of energy from 12 different food groups) or diet diversity (Shannon diversity index based on percent of energy from 12 different food groups) (Table 1). Additional analyses examining diet intake in relation to age and sex can be found in the Supplementary Results. As expected, adult ADHD patients had compared to controls significantly higher scores of ADHD symptoms and difficulties in emotion regulation scored by the Adult ADHD Self-Report Scale (ASRS) and the Difficulties in Emotion Regulation Scale-16 (DERS), respectively. The child ADHD group consisted of significantly more males compared to the adult ADHD group (Table 1). Moreover, compared to adults with ADHD, children with ADHD had significantly different diet intake (dietary PC1 and PC2) and a higher proportion of plasma sVCAM-1 levels that was above the median of the ADHD group (i.e., 519520 ng/ L).
(caption on next page) Fig. 1. The ADHD cohort and the association of clinical characteristics and ADHD diagnosis with fecal microbiome diversity metrices. A) The experimental design. In total, 199 individuals were included in the analyses: Adult ADHD patients N = 84, child ADHD patients N = 63 and adult healthy controls N = 52. Shallow shotgun sequencing was used to sequence the fecal samples resulting in taxonomic (bacterial strains) and functional (bacterial genes encoding enzymes) data. In addition, psychiatric, GI and diet-scale scores, as well as plasma levels of inflammatory markers and SCFAs, were collected. B) Demographic and clinical characteristics in child ADHD patients, n = 62, and their association with the fecal taxonomic strain beta diversity metrices (Jaccard, Aitchison & Bray-Curtis distance) using Adonis. C) Demographic and clinical characteristics in adult ADHD patients, n = 81, and their association with the fecal taxonomic strain beta diversity metrices using Adonis. D) Taxonomic strain beta diversity measured by Aitchison distance was different comparing adults with ADHD, n = 81, and adult controls, n = 52. The difference was assessed using Adonis adjusted for age, sex and BMI. E) Functional enzyme beta diversity measured by Jaccard distance was different comparing adults with ADHD, N = 84, and adult controls, N = 52. The difference was assessed using Adonis adjusted for age, sex and BMI. F) The strain Inediibacterium massiliense Mt12 was significantly lower abundant in adult ADHD patients, n = 81, compared to controls, n = 52), estimated using ANCOM-BC and Holm-corrected. The y-axis represents the microbial load bias-adjusted abundances of the strain I. massiliense Mt12. G) The bacterial gene encoding enzyme 16S rRNA-methyltransferase (2.1.1.170), was Holm-significantly higher abundant in adult ADHD patients, n = 83, compared to controls, n = 52, estimated using ANCOM-BC. The y-axis represents the microbial load bias adjusted abundances of the bacterial gene-encoded enzyme 2.1.1.170. *p FDR ≤ 0.05, **p FDR ≤ 0.01, ***p FDR ≤ 0.001. SCFAs = short chain fatty acids; GI = Gastro-intestinal; BMI = body mass index; BSS = Bristol stool scale; ASRS = Adult ADHD self-report scale; DERS = Difficulties in emotion regulation scale-16; FDR = false discovery rate; SNAP = Parent report, Swanson, Nolan and Pelham scale; ANCOM-BC = Analysis of Compositions of Microbiomes with Bias Correction; EC = Enzyme Commission Dietary PC = Principal component (PC) coordinates from a Principal component analysis (PCA) based on the food questionnaire data to capture diet intake. Diet diversity, Shannon = Shannon diversity index based on % of energy from 12 different food groups. Vascular inflammation status = Above or below the plasma level 519520 ng/L of soluble vascular cell adhesion molecule-1 (sVCAM-1). This cut-off is the median sVCAM-1 plasma level in the original ADHD cohort (Skott et al., 2020).  Medication use refers to current use for children, and current use or used in the last three months for adults. Among those without psychostimulant medication use, 9 children and 9 adults used psychostimulant medication anytime 4-24 months before examination. Information on medication use more than 24 months before examination is unavailable. Psychostimulants for children include Methylphenidate and Lisdexamphetamine. For adults psychostimulants include Methylphenidate, Lisdexamphetamine and Dexamphetamine. Vascular inflammation status = having a plasma sVCAM-1 level > 519520 ng/L, which is the median of sVCAM-1 levels in the original ADHD cohort (Skott et al., 2020). * Functional total count data is defined by genes encoding Kyoto Encyclopedia of Genes and Genomes Orthology groups (KOs); in preprocessing (before Table data) n = 4 individuals were removed due to having less than 100 000 functional counts. ** Taxonomical total count data is defined by bacterial strains; in preprocessing n = 8 individuals were removed due to having less than 500 000 taxonomical counts.

Clinical variables associated with microbial strain diversity in child and adult ADHD patients
In the ADHD cohort with taxonomical data of 143 patients (n children = 62, n adults = 81), we investigated which clinical variables predicted strain alpha and beta diversities. The alpha diversity was examined using the 'number of observed strains' that measures richness, the Pielou's evenness index that measures evenness, and the Shannon diversity index that measures evenness combined with richness (Kim et al., 2017). The beta diversity was examined using (i) Jaccard distance, a qualitative strain measure based on strain presence/absence between samples that does not consider the abundance of the strains (Yang et al., 2021), (ii) Aitchison distance, a measure based on Euclidean distance between samples with center log ratio (clr)-transformed strain counts (Otero et al., 2005), and (iii) Bray-Curtis distance, a quantitative measure based on the difference in the abundance of strains between samples (Yang et al., 2021). All diversity analyses were adjusted for the total bacterial count. When regressing each clinical variable against the different alpha diversity measures, Pielou's evenness and Shannon diversity were significantly associated with the BSS score (p Evenness,FDR = 0.0024, Beta Evenness = -0.44, partial R 2 Evenness = 18%; p Shannon,FDR < 0.0010, Beta Shannon = -0.46, partial R 2 Shannon = 21%) with looser stool being associated with lower evenness (Supplementary output file 1) in child ADHD patients. No clinical variable was significantly associated with alpha diversity (p FDR > 0.05) in adults with ADHD (Supplementary output file 1). Analysis using Adonis revealed that delivery mode, BSS score, psychostimulant medication and probiotic use were significantly associated with beta diversity in children with ADHD (p FDR < 0.05, Fig. 1B). Furthermore, diet (i.e., diet diversity, dietary PC1 and dietary PC3) was significantly associated with beta diversity in adult ADHD patients (p FDR < 0.05, Fig. 1C). In adult controls, BMI and dietary PC1 were FDR-significantly associated with beta diversity (Fig. S2A).
Notably, none of these clinical variables were significantly associated with all of the three distance metrics (Jaccard, Aitchison and Bray-Curtis) in any of the three groups (i.e., adults with ADHD, children with ADHD and adult controls).

Differences in diversity between adult ADHD patients and adult controls
In the case-control analyses we included only adults since (i) age influences the microbial fecal composition and (ii) there were only 6 fecal samples collected from child controls. Thus, the taxonomic casecontrol analyses included 133 individuals (n Adult ADHD = 81, n Adult controls = 52). To account for potential confounders when comparing the gut microbiome diversity of adult ADHD patients with adult controls we adjusted for age, sex, dietary axis PC1, diet diversity, BSS score, and BMI. When comparing the adult ADHD patients to the adult controls, we found no significant differences for any of the three alpha diversity measures (p Observed strains = 0.060). Significant taxonomical strain beta diversity differences between ADHD patients and controls were found for Jaccard and Aitchison when adjusting for age, sex, and BMI (R 2 Jaccard = 0.99%, p Jaccard = 0.018; R 2 Aitchison = 1.1%, p Aitchison = 0.019; n = 131; Fig. 1D). For Aitchison, the significance remained when also adjusting for BSS score, dietary PC1 and diet diversity (R 2 Aitchison = 1.2%, p Aitchison = 0.027; n = 110). However, there was no significant difference between patients and controls for Bray-Curtis distance (Supplementary output file 2).
The functional case-control diversity analyses examining enzymeencoding bacterial genes included 136 individuals (n Adult ADHD = 84, n Adult Controls = 52). No significant difference between adult controls and adult ADHD patients in functional enzyme alpha diversity was found. Significant functional enzyme beta diversity differences between ADHD patients and controls were found for Jaccard distances when adjusting for age, sex and BMI (R 2 Jaccard = 1.1%, p Jaccard = 0.019; n = 134; Fig. 1E), as well as when adjusting for BSS score, dietary PC1 and diet diversity (R 2 Jaccard = 1.3%, p Jaccard = 0.031; n = 112). There were no other significant differences between patients and controls for Aitchison or Bray-Curtis distances (Supplementary output file 2). Unadjusted case-control analyses as well as bacterial diversity comparisons between adults with ADHD, children with ADHD and adult controls can be found in Supplementary Results.

Differential abundance analysis revealed one strain and one bacterial gene encoding enzyme that differed in abundance between adult ADHD patients and controls
Differential abundance testing using ANCOM-BC identified one strain, Inediibacterium massiliense Mt12 (p Holm = 0.012), to be significantly lower abundant in adult ADHD patients (n = 81) compared to adult controls (n = 52), adjusted for sex, age, and BMI ( Fig. 1F and Supplementary output file 3). Inediibacterium massiliense remained significantly different even when performing the analysis on the species level (p Holm = 0.010). To examine whether the abundance of this strain associated with psychiatric symptom severity, the bias-adjusted abundance of the significant strain was regressed against scale-scores measuring ADHD symptoms and functionality, emotion regulation (DERS), and ASD-symptoms. However, no FDR-significant associations were found.
ANCOM-BC identified one bacterial gene encoding enzyme, i.e., 16S rRNA-methyltransferase (2.1.1.170), to be significantly more abundant in the adult ADHD patients (n = 84) compared to adult controls (n = 52) (p Holm = 0.0064; Fig. 1G and Supplementary output file 3). When regressing the bias-adjusted abundances of the 16S rRNAmethyltransferase to symptom scale scores, no FDR-significant associations were found. KOs were also analysed, and the results can be found in the Supplementary output file 3 and the Supplementary Results. The Supplementary Results also include an analysis in a present/non-present setting for the features (strains/enzymes) found in less than 10 ADHD patients or controls.

Psychostimulant medication is associated with alterations in fecal bacteria in children with ADHD
In our previous investigations using the complete study cohort, i.e., not only individuals with microbiome data, we found that children with ADHD on (vs. not on) psychostimulant medication had (i) higher plasma levels of vascular inflammatory markers (Yang et al., 2020), and (ii) lower plasma levels of the SCFAs acetic acid and propionic acid (Yang et al., 2023;Yang et al., 2022). However, these associations were not observed in the adult ADHD group. Since there is a reported link between the immune system and gut microbiota (Fung, 2020;Geuking et al., 2014), and since SCFAs are downstream products of the gut microbiota (Boets et al., 2017), we performed additional analyses in children with ADHD (n = 63) by investigating the association between fecal bacterial diversity and psychostimulant medication. There were no significant differences in clinical characteristics between children on vs. not on psychostimulant medication (Supplementary Table S2). However, there was a suggestive significant difference in diet diversity (p = 0.06) between the two groups. Therefore, diet diversity was used as an adjustment in downstream analyses.
Analysis of the taxonomical strain beta diversity, adjusted for age, sex and diet diversity, revealed a significant difference between children with ADHD on psychostimulant medication (n = 33) and those not on psychostimulant medication (n = 29) for both Aitchison distance (R 2 Aitchison = 2.6%, p Aitchison = 0.004) and Bray-Curtis distance (R 2 Bray-Curtis = 3.1%, p Bray-Curtis = 0.047) ( Fig. 2A). There was no significant difference between the two groups using Jaccard distances. Similarly, no significant difference was found in terms of functional enzyme beta diversity. Examination of the alpha diversity showed that being on psychostimulant medication was associated with lower strain evenness, measured by Pielou's evenness (p = 0.034, Beta = 0.16, partial R 2 = 9.0%), as well as lower functional enzyme evenness (p = 0.007, Beta = 0.18, partial R 2 = 12%), both adjusted for age, sex, and diet diversity ( Fig. 2B-C). However, there was no significant difference between the two groups in terms of richness for taxonomical strain or functional enzyme data. No differences in microbiome diversities were found in adults with ADHD on psychostimulants compared to those not on psychostimulants. ANCOM-BC identified the strain Bacteroides stercoris CL09T03C01 to be significantly lower in children with ADHD on psychostimulant medication (n = 30) compared to those not on medication (n = 29) (p Holm = 0.023; Fig. 2D). In addition, the bacterial gene encoding the enzyme adenosylcobinamide-phosphate synthase (6.3.1.10) involved in Cobalamin (Vitamin B 12 ) biosynthesis was found to be significantly lower in children with ADHD on psychostimulant medication (n = 33) compared to children not on medication (n = 30) (p Holm = 0.012; Fig. 2E, Supplementary output file 3). When regressing the bias-adjusted abundances of B. stercoris CL09T03C01 and adenosylcobinamidephosphate synthase against symptom scale scores, plasma levels of SCFAs and plasma levels of the inflammatory markers sICAM-1 and sVCAM-1 in children with ADHD, we found that B. stercoris CL09T03C01 Fig. 2. The fecal microbiome diversity in children with ADHD and with or without psychostimulant medication. A) Taxonomic (bacterial strains) Aitchison distance in children with ADHD on psychostimulant medication, n = 33, was different compared to children with ADHD not on psychostimulant medication, n = 29. The difference was assessed using Adonis adjusted for age, sex and diet diversity. B) Taxonomic Pielou's evenness was lower in children with ADHD on psychostimulant medication, n = 33, compared to those not on psychostimulant medication, n = 29. The difference was assessed using a permutational linear model adjusted for age, sex and diet diversity. C) Functional (bacterial genes encoding enzymes) pielou's evenness was lower in children with ADHD on psychostimulant medication, n = 33, compared to those not using psychostimulant medication, n = 30. The difference was assessed using a permutational linear model adjusted for age, sex and diet diversity. D) The strain Bacteroides stercoris CL09T03C01 was Holm-significantly less abundant in children with ADHD on psychostimulant medication n = 30, compared to those not using psychostimulant medication, n = 29, using ANCOM-BC. The y-axis shows microbial load bias-adjusted abundances of B. stercoris CL09T03C01. E) The bacterial gene encoding vitamin B 12 synthesis enzyme adenosylcobinamide-phosphate synthase (6.3.1.10) was Holm-significantly lower abundant in children with ADHD on psychostimulant medication n = 33, compared to those not using psychostimulant medication, n = 30, using ANCOM-BC. The yaxis shows microbial load bias-adjusted abundances of 6.3.1.10. F) The plasma levels of SCFAs formic and propionic acid [uM] were FDR-significant positively associated with bias-adjusted abundances of Bacteroides stercoris CL09T03C01 in children with ADHD, n = 53. The associations were assessed using linear models. G) Plasma levels of the vascular inflammatory markers sICAM-1 and sVCAM-1, measured in a subsample, were FDR-significantly higher in children with ADHD on psychostimulant medication, n = 19, compared to those not on psychostimulant medication, n = 13. The difference was assessed using a permutational linear model adjusted for age, sex and diet diversity. sVCAM-1 = soluble vascular cell adhesion molecule-1; sICAM-1 = soluble intracellular cell adhesion molecule-1; Adonis = Permutational Multivariate Analysis of Variance Using Distance Matrices; EC = Enzyme Commission. had significant positive associations with the SCFAs formic (p FDR = 0.013, Beta = 0.41, partial R 2 = 17%) and propionic (p FDR = 0.013, Beta = 0.42, partial R 2 = 18%) acids (n = 53) (Fig. 2F). No significant associations were found for adenosylcobinamide-phosphate synthase. Since in our original cohort (Yang et al., 2022) SCFAs were lower in children with ADHD on psychostimulants, compared to those not on psychostimulants, we performed a sensitivity analysis by adjusting for psychostimulant use to account for its mediating effect when regressing the SCFA levels on the bias-adjusted abundances of B. stercoris CL09T03C01. However, even after adjusting for psychostimulant use, the formic (p FDR = 0.014, Beta = 0.36, partial R 2 = 13%) and propionic (p FDR = 0.014, Beta = 0.36, partial R 2 = 14%) acids had significant positive associations with the bias-adjusted abundances of B. stercoris CL09T03C01.
Analyses focusing on vascular inflammation and the concentrations of acetic and propionic acids were also performed to examine whether the prior reported differences (Yang et al., 2022;Yang et al., 2020) could be observed in this sub-cohort with data showing alterations in the taxonomic and functional fecal microbiome. We found higher plasma levels of sICAM-1 (p FDR = 0.004, Beta = -0.25, partial R 2 = 27%) and sVCAM-1 (p FDR = 0.005, Beta = -0.26, partial R 2 = 27%) in children with ADHD on psychostimulant medication (n = 22) compared to children not on psychostimulant medication (n = 10), after adjusting for age, sex, and diet diversity (Fig. 2G). No differences in levels of propionic (p = 0.078) or acetic acids was found between medicated (n = 32) and non-medicated (n = 25) children. Using Adonis, we found no significant associations between strain Aitchison and Bray-Curtis beta diversity, and sICAM-1 or sVCAM-1 (p Aitchison = 0.076, p Bray-Curtis = 0.081) levels. An analysis of microbiome-deviating samples in the psychostimulant group can be found in the Supplementary Results and Supplementary Fig. S4.

Discussion
In this largest-to-date fecal metagenomics study in ADHD, we investigated the associations between the gut bacterial microbiome and (i) clinical variables in 81 adult ADHD patients and 62 child ADHD patients, (ii) ADHD diagnosis in 163 adults (adult ADHD patients compared to adult controls), and (iii) psychostimulant medication. The findings are summarized in Supplementary Table S4. As expected, dietary axis PCs and dietary diversity were significantly associated with bacterial-strain beta diversity in both adult ADHD patients and controls. Diet has already been shown, or proposed, to influence the human fecal bacterial composition (Bolte et al., 2021;David et al., 2014;Turnbaugh et al., 2009), demonstrating that our generated data has adequate quality and power to detect these influences. In child ADHD patients, probiotic use, delivery mode and BSS scores were significantly associated with bacterial strain beta diversity. Interestingly, we could not detect any association between probiotic use or delivery mode with the beta diversity in adults. It has been reported that the microbiome in very early childhood is less stable than the adult microbiome, although less in known about the microbiome in 3-18 year old individuals (Derrien et al., 2019). Since our group of children has a median age of 13 years, our findings suggest that delivery mode and the use of probiotics may affect the microbiome into late childhood but not in adulthood. This could be attributed to a less stable/not-yet normalized gut microbiome in younger individuals. However, since the group sizes for C-section and probiotic use in our cohort were small (n less than 12), replications of these findings are warranted in larger studies. We also found that none of the clinical variables were significantly associated with all three distance metrics measuring bacterial strain beta diversity. This reinforces the importance of using several beta diversity distances that measure different aspects of the microbiome composition.
When comparing adult ADHD patients to adult controls, in which putative confounders were carefully considered, we found a significant difference in (i) the beta diversity distance measures Jaccard and Aitchison for taxonomical strain data, and (ii) the Jaccard distance measure for functional enzyme data. The Bray-Curtis distance measure was not significantly different in the strain or enzyme beta diversity analyses, suggesting that the less abundant strains and enzymes were the ones driving the difference in beta diversity between ADHD patients and controls, since Jaccard gives more weight to low abundant features compared to Bray-Curtis (Yang et al., 2021).
ANCOM-BC revealed the strain Inediibacterium massiliense Mt12 and the enzyme 16S rRNA-methyltransferase (2.1.1.170) to be differentially abundant in adult ADHD patients compared to adult controls. Specifically, I. massiliense Mt12 was lower abundant and the enzyme 16S rRNA-methyltransferase was higher abundant in adult ADHD patients compared to adult controls. Only one study, to our knowledge, has reported on I. massiliense Mt12, where it was discovered and isolated from the fecal sample of a severely malnourished infant (Alou et al., 2017). 16S rRNA-methyltransferase is an enzyme implicated in posttranscriptional modifications of ribosomes (Wong et al., 2013) and mutations in the gene coding for the enzyme have been found to result in antimicrobial resistance (Okamoto et al., 2007). However, future studies are needed to understand whether this bacterial enzyme, including its non-mutated form, could affect the human microbiome.
In our study cohort, 57% of the children with ADHD were on psychostimulant medication. Prior studies using the complete cohort found that children with ADHD on psychostimulant medication had (i) lower plasma concentrations of acetic and propionic acids (Yang et al., 2022), and (ii) higher plasma levels of the inflammatory markers, sVCAM-1 and sICAM-1 (Yang et al., 2020;Yang et al., 2023). Interestingly, in our fecal sample sub-cohort, where we excluded those on antibiotic or laxative medication and adjusted for age, sex and diet diversity, children with ADHD on (vs. not on) psychostimulant medication were found to have significantly (i) different strain beta diversity profile, measured by Bray-Curtis distance and Aitchison distance, (ii) lower alpha diversity enzyme-and strain-evenness, (iii) lower abundance of the strain Bacteroides stercoris CL09T03C01 and bacterial genes encoding an enzyme involved in vitamin B 12 synthesis, and (iv) higher plasma levels of sICAM-1 and sVCAM-1. sICAM-1 and sVCAM-1 are two vascular inflammatory markers implicated in migration of leukocytes from the blood into the surrounding tissues that facilitate inflammatory responses and have been suggested to be associated with clinical features of depression, schizophrenia, and bipolar disorder (Müller, 2019;Pantović-Stefanović et al., 2018;Radu et al., 2020;Tchalla et al., 2015). Moreover, sVCAM-1 has been suggested to be a factor involved in brain endothelial barrier function (Haarmann et al., 2015). However, our observational measures do not allow us to untangle the biological relationships between psychostimulant medication and changes in the fecal bacterial microbiome. Thus, we cannot establish if the microbiome is being affected primarily by psychostimulants, in turn increasing vascular inflammatory markers, or if increased vascular inflammation induced by psychostimulants affects the microbiome. Strain Aitchison beta diversity was not significantly associated with plasma levels of sICAM-1 and sVCAM-1. However, children with ADHD with both taxonomical analyzable samples and vascular inflammatory data available were few (N = 32), thus lowering the statistical power to detect significant associations. Future mechanistic studies investigating causal relationships between psychostimulants, fecal microbiome and vascular inflammation are therefore warranted. Interestingly, we found no difference in microbiome diversity in adults with ADHD on psychostimulant medication compared to those not on medication. However, we cannot exclude the possibility that this lack of difference was due to effects of past psychostimulant medication in those patients currently not receiving medication, where 9 of the adults not on medication had recieved medication in the last 4-24 months.
ANCOM-BC identified one strain (Bacteroides stercoris CL09T03C01) and one enzyme encoded by bacterial genes (adenosylcobinamidephosphate synthase) to be lower abundant in children with ADHD on psychostimulants compared those not taking psychostimulants. The available information on the strain Bacteroides stercoris CL09T03C01 is rather limited. One study characterized the B. stercoris CL09T03C01 strain in a human gut when exploring genes encoding for the toxin secretion system in Bacteroidales (Coyne et al., 2016). Interestingly, in our cohort we found the B. stercoris CL09T03C01 strain to be positively associated with plasma levels of the SCFAs formic and propionic acids in children with ADHD. Indeed, the genus Bacteroides has been reported to produce formic (Song et al., 2004) and propionic acids (Louis & Flint, 2017) and the species, B. stercoris, has been reported to produce propionic acid in the human gut (Johnson et al., 1986;Lan et al., 2006). Thus, the lower abundance of B. stercoris CL09T03C01 in ADHD children on psychostimulants is in accordance with the lower plasma levels of SCFAs reported in the original full cohort (Yang et al., 2022). Notably, in the current sub-cohort we could not detect higher SCFA levels associated with psychostimulant use, which may be due to a smaller sample size (n original cohort = 88, n used sub cohort = 57).
The bacterial genes encoding the enzyme adenosylcobinamidephosphate synthase (6.3.1.10) were also significantly lower abundant in children with ADHD taking psychostimulants. Interestingly, adenosylcobinamide-phosphate synthase is an enzyme involved in the prokaryotic synthesis of cobalamin (vitamin B 12 ) (Warren et al., 2002) (https://www.genome.jp/entry/6.3.1.10), known to play a vital role in normal brain functioning. Notably, studies suggest that the microbial communities in the gut are more likely competing with the host for dietary vitamin B 12 , rather than contributing vitamin B 12 to the host (Degnan et al., 2014;Roth & Mohamadzadeh, 2021). Nevertheless, vitamin B 12 is reported to be a modulator of the gut microbiome composition, such as enhancing SCFA-producing bacteria (Xu et al., 2018), increasing alpha diversity (Guetterman et al., 2021) and suppressing the virulence factor Shiga toxin 2 (Cordonnier et al., 2016).
Previous studies have found that several psychiatric medications are associated with the fecal microbiome (Cussotto et al., 2019;Maier et al., 2018). However, to date, there are very few reports on a possible association between psychostimulant medication and the fecal microbiome. The majority of studies investigating the fecal microbiome in ADHD have either treatment-naïve patients or are not comparing the microbiome between patients taking and not taking psychostimulants (Aarts et al., 2017;Jiang et al., 2018;Richarte et al., 2021;Wan et al., 2020;. A study in ADHD patients between 13 and 29 years (n = 41), with a mean age of 20 years, did not detect any significant association between beta diversity and psychostimulant use but found that three genera were lower abundant in patients on psychostimulants (Szopinska-Tokov et al., 2020). Since our study detected the association between psychostimulants and fecal microbiome only in children, it is hard to compare our findings to this previous study that did not include age group (child/adult) stratified analyses. Another study had only four patients in the non-medicated group, which makes a statistically adequate analysis not possible (Prehn-Kristensen et al., 2018). A limitation to our study is the lack of data on symptom severity prior to medication with psychostimulants in children with ADHD. As a result, we cannot rule out the possibility that the intensity of disease burden could mediate the observed differences in plasma levels and microbiome diversity in children on psychostimulant medication. However, all patients with an ADHD diagnosis in the participating clinics are offered psychostimulant medication. Further studies are warranted to test our findings and elucidate the biological relationship between psychostimulant medication and changes in the fecal microbiome, including whether stimulants have a long-term effect on the microbiome and whether supplement treatment could be used to sustain a healthy gut microbiota.
Several microbiome studies have indicated the importance of adjusting for confounding factors, such as drug use (Imhann et al., 2017), BSS score Vandeputte et al., 2016b) and diet (Bolte et al., 2021;David et al., 2014;Yap et al., 2021). A previous study found an upstream effect of diet on the altered gut microbiome in autism (Yap et al., 2021). However, we did not find significant differences in eating habits between adult ADHD patients and adult controls, thus eliminating diet as a clear mediator of the observed differences in our study. Furthermore, the dietary PC1 axis and diet diversity were used as adjustments in all case-control beta diversity analyses to account for individual differences. However, the use of a retrospective self-reported diet questionnaire still constitutes a limitation. Nonetheless, we found a significant association in both dietary diversity and dietary PC1 axis with beta diversity in two of the three distance measures for adults with ADHD, indicating that the diet data captured some gut microbiomeinfluencing eating habits.
In addition, since antibiotics, laxatives and proton pump inhibitors can influence the gut microbiome (Imhann et al., 2017;Vich Vila et al., 2020), we excluded all individuals that had used an antibiotics medication the last 2 years, or that had used proton pump inhibitors or laxatives the past three months.
Prior studies on the fecal gut microbiome in ADHD are inconclusive and heterogeneous, which could be attributed to low sample sizes, poor use of appropriate statistical and bioinformatic methods, or not taking potential confounders into consideration. In addition, differences in age, environment, geographic location, sampling procedure and technical methods (such as DNA extraction, sequencing platforms, bioinformatic pipelines, taxonomical and functional annotation) make it challenging to directly compare our results to that of other studies. Most studies, to date, investigating the gut microbiome in ADHD patients have used pediatric/adolescent cohorts and 16S rRNA sequencing to generate their data (Aarts et al., 2017;Jiang et al., 2018;Prehn-Kristensen et al., 2018;Richarte et al., 2021;Szopinska-Tokov et al., 2020;Wan et al., 2020;, which also complicates a direct comparison with our results that were generated using shotgun sequencing and only compared an adult ADHD patient population to healthy controls. To our knowledge, only one previously reported study used shotgun sequencing, looking at a smaller sample (N = 17) of children with ADHD (Wan et al., 2020) and found a KEGG ortholog, K06269 -implicated in dopaminergic synaptic pathways, that was upregulated in ADHD patients. However, according to the KEGG database, K06269 is only expressed by eukaryotic cells, thus making it difficult to draw any specific conclusions about the bacterial gut microbiome's role. One study used an adult population, which consisted of 100 treatment-naïve ADHD patients (Richarte et al., 2021) where no differences were found in alpha or beta diversity compared to healthy controls. The latter study only looked at Bray-Curtis distances and is therefore still in line with our findings since we also did not find any differences between diagnoses when using Bray-Curtis distances. The same study did not present the Jaccard and Aitchison distances, which were the distances that captured a difference between ADHD and healthy controls in our study. Although they found several genera showing significant differences in relative abundance, these genera did not replicate in our study. However, it is hard to draw conclusions about these discrepancies since the latter study used 16S rRNA sequencing and performed the analyses on family genus and phylum levels.

Limitations
Our study has a number of limitations that need to be acknowledged: (i) There was only a very small sample of child controls (N = 6), which makes statistical comparisons unsuitable, and we can therefore not draw any conclusions about the children with ADHD in comparison to a healthy population. (ii) Our study's fecal samples represent the baseline samples collected in a clinical randomized placebo-controlled trial of a synbiotic treatment consisting of a mixture between dietary fibers and lactobacilli strains. Therefore, this study group may represent individuals motivated to try nonpharmacological supplements, thus affecting the generalizability of our findings. (iii) Since we only analyzed DNA data, the activity levels of, e.g., enzymes still need to be confirmed in meta-transcriptomic settings with RNA material. (iv) Self-reporting diet questionnaires do not fully capture real-life dietary habits and our questionnaire was not optimal to capture the habits of individuals with a vegetarian or vegan diet. (v) The samples were sequenced in two different batches, and although the pipelines and sequencing platforms used were the same, the second batch had higher sequencing depth which we, however, adjusted for in the statistical analyses. (vi) There is evidence suggesting that the genetic background of the host may impact the composition of the gut microbiome (Awany et al., 2019;Gilbert et al., 2018;Hughes et al., 2020), thus making it an important factor that should be taken into consideration in future studies. (vii) Studies investigating the possible link between the gut microbiome composition and neurodevelopment are mostly focusing on the first years of life (Cowan et al., 2020;Kelsey et al., 2021;Vaher et al., 2022). This makes age-group stratifications that are based on a microbiome rationale more complex, since the later years of childhood and adolescent are less explored. Therefore, separating adults and children with an 18-year-old cut-off could in terms of the microbiome be arbitrary. Nonetheless, this division was clinically relevant in our study since patients over 18 years of age were being evaluated using different psychiatric scales, clinics, and protocols.

Conclusion
We conducted the largest, to our knowledge, shotgun study investigating the fecal bacterial microbiome in ADHD using a wellcharacterized human cohort with extensive information about lifestyle, drug use, biological measures and confounders affecting the microbiome. Our main findings provide further evidence to support a significant difference in the fecal microbiome composition in adult ADHD patients compared to adult controls. In addition, our findings suggest a novel association between psychostimulant medication in children with ADHD and changes in fecal enzyme and strain bacterial diversity, and a lower abundance of bacterial genes encoding an enzyme in vitamin B 12 synthesis. Additional large-scale studies are needed to replicate these findings, including longitudinal large-scale multi-omics intervention studies aimed at uncovering the causal relationships between gut microbiota and ADHD.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability
Data will be made available on request to researchers with an ethical permit