Microbiome-gut-brain axis in brain development, cognition and behavior during infancy and early childhood

The gut microbiota is increasingly recognized as a modulator of brain and behavior but its role in early childhood, when the microbiome and the brain are both undergoing rapid development, is poorly understood. Preclinical work suggests there are critical windows during early life when bacterial signals are required for normal neurobehavioral development, whereas gut microbial dysfunction has been observed in patients with certain neurodevelopmental disorders. Here, we review the evidence that gut bacterial diversity and community composition affect brain structure/function and behavior in typically developing pre-school children. Following narrative synthesis, we report that twenty studies suggest the microbiome-gut-brain axis may operate across three domains in infancy and early childhood: general neurocognitive development, socio-emotional behaviors, and brain structure and function inferred from neuroimaging. However, there is substantial variation in the bacteria-brain/behavior relationships reported. We identified sources of clinical and methodological heterogeneity in the studies, including participant characteristics, small sample sizes, variations in DNA extraction and sequencing, and statistical analysis approaches. We propose that harmonization of sample collection and data processing pipelines, longitudinal assessments, and mechanistic insights from whole metagenome analyses could improve understanding of the role of gut microbiome in brain development during early development. This will also promote comparability between studies and increase study power by allowing for meta-analyses. Greater knowledge of the role of gut microbiome in brain development may ultimately offer new avenues for promoting brain health in early life.


Introduction
Human life has never existed without microbes.The human body hosts an array of microorganisms, with the majority and most diverse set of microbes, including bacteria, viruses and fungi, present in the gastrointestinal tract.These, especially bacteria, play an important part in a range of physiological functions within the body, including host metabolism and immunity.The past few decades have seen an explosion of research into the bacterial influence on the brain via the microbiome-gut-brain axis (Cryan et al., 2019).
During the first few years of postnatal life the brain undergoes substantial changes, including increases in tissue volume and cortical complexity, organization of white matter fibers, and myelination, with different regions and networks maturing at different rates (Gilmore et al., 2018;Fig. 1).These changes underpin the acquisition of language, motor and social skills.In parallel to these intense Fig. 1.Development of the brain and microbiome during the first five years of life.The top panel illustrates the time course of key neurodevelopmental processes that underpin the acquisition of language, motor and social skills, and that could serve as neurobiological mechanisms linking gut microbiome with neurodevelopmental outcomes; adapted from (Casey et al., 2005;Semple et al., 2013;Shonkoff & Phillips, 2000); the height of the diamonds signifies the intensity/peak of the process.The bottom panel illustrates the developmental trajectories of the gut microbiome alpha diversity and community composition.Colonization pattern based on data presented in (Roswall et al., 2021), focusing on the most abundant taxa across the first five years of life.Created with BioRender.com.
K. Vaher et al. periods in brain and cognitive development, the gut microbiome, which is rapidly populated at birth, similarly follows a dynamic developmental trajectory (Bäckhed et al., 2015;Reyman et al., 2019;Roswall et al., 2021;Yatsunenko et al., 2012;Fig. 1).This has led to the notion of nested sensitive periods when brain development interacts with peripheral system development, in this case the gut microbiome, to shape the emergence of complex behaviors including learning and memory (Callaghan, 2020).It has been suggested that signals from the microbiome are required for certain aspects of brain development (Cowan et al., 2019;Stilling et al., 2014) and recent data from preclinical models support the theory that bacterial signals during critical windows in early life are required for typical neurobehavioral development (Buffington et al., 2016;Chu et al., 2019;Clarke et al., 2013;Diaz Heijtz et al., 2011;Sudo et al., 2004).Conversely, disruption of the microbiome development during these periods could impact concurrently developing cognitive functions (Callaghan, 2020;Cowan et al., 2019).
Clinical investigation provides indirect evidence for a putative role of the gut microbiome in human brain development.First, disrupted gut microbiome profiles are reported in individuals with autism spectrum disorder and attention deficit-hyperactivity disorder (Jurek et al., 2020).Second, important drivers of the microbiome development in early life (e.g.breastfeeding, delivery via Caesarean section (C-section)) are associated with cognition as well as aspects of brain development as assessed by magnetic resonance imaging (MRI) (Anderson et al., 1999;Bauer et al., 2019;Belfort et al., 2016;Blesa et al., 2019;Deoni et al., 2013Deoni et al., , 2018Deoni et al., ,2019;;Kar et al., 2021;Luby et al., 2016;Ottolini et al., 2020;Polidano et al., 2017).However, explorations into direct associations between gut microbial composition and neurocognitive and behavioral measures in early childhood have only emerged in the last 5-6 years.In this review, we evaluate the evidence that gut bacterial diversity and composition may affect human brain development.Here, we specifically examine features of the gut microbiome derived from bacterial DNA sequencing that are associated with brain development, indexed by quantitative neuroimaging, and/or neurobehavioral outcomes in typically developing children from infancy to 5 years.

Material and methods
We searched PubMed, Web of Science and Embase to identify studies that integrated bacterial DNA sequencing and quantitative neuroimaging or neurobehavioral assessments until June 2021.The search strategy used combined keywords describing gut microbiome with terms targeting: i) neuroimaging and brain features; or ii) child neurocognitive and behavioral outcomes (see Appendix A for search details).Studies were included if they reported data from human participants between birth and 5 years without a diagnosis of a neurodevelopmental disorder at the time of recruitment, and if they directly investigated relationships between gut microbiome features derived from bacterial DNA sequencing (metagenomics or 16S-based) and quantitative neuroimaging or neurobehavioral assessments.We considered publications using any neuroimaging modality, and neurobehavior could include general cognitive ability, motor development, social cognition and development, attention, language, communication, emotion recognition or temperament derived from direct observation or parent/teacher report.Single case studies, case-control studies solely focusing on microbiota differences between healthy/neurotypical and children with neurodevelopmental disorders, non-English language studies, and abstracts were excluded.Results are synthesized in tabular format for microbiota associations with measures of i) neurocognitive development, ii) socio-emotional behaviors/temperament, and iii) neuroimaging.

Risk of bias assessment
In the absence of a validated quality assessment tool for studies linking microbiome data with neurobehavioral and neuroimaging data, we summarized potential sources of bias arising from methodological heterogeneity between studies based on the STORMS Table 1 Overview of studies.Note that some studies included multiple measures of neurodevelopment: of the 7 studies that assessed general neurocognitive development, one also reported microbiome correlations with socio-emotional behaviors and one reported microbiome correlations with neuroimaging; of the 13 studies that assessed socio-emotional behaviors, two also reported microbiome correlations with neuroimaging.guidelines (Mirzayi et al., 2021).We extracted data on the following study characteristics: age of participants at microbiome and neurodevelopmental assessments, participant exclusion criteria, bacterial DNA extraction method, sequencing methodology, microbiota features of interest, use of validated neurodevelopmental outcome assessments, neuroimaging modality and use of region-ofinterest or whole-brain analyses, statistical models linking gut microbiota features with neurodevelopmental outcomes, adjustment for covariates, and correction for multiple comparisons (Supplementary Table 1).

Results and discussion
Our search yielded a total of 20 publications that associated features of gut bacterial composition with neurodevelopment indexed by neuroimaging and/or behavioral assessments.Due to the emerging nature of the field, all studies were published after 2015.We provide a synthesis of the studies structured around three domains: general neurocognitive development, socio-emotional behavior and temperament, and brain structure and function inferred from neuroimaging.An overview of the methodologies in the included studies is presented in Table 1.We also refer the reader to Table 2 for a brief explanation of the terminology commonly used for gut microbiome measurement and characterization.

Gut microbiome associations with general neurocognitive development in early childhood
We identified seven studies which investigated associations between microbiota composition and general neurocognitive development in typically developing children (Table 3).In all seven studies, the gut microbiota was assessed using 16S rRNA amplicon sequencing.The studies used different measures to assess neurocognitive development, including direct observation with the Mullen Scales of Early Learning (MSEL), Gesell Development Inventory (GDI), and the Bayley Scales of Infant Development (BSID); and parent/caregiver report using the Ages and Stages Questionnaire (ASQ).Two of the studies investigated the potential longitudinal effects of the gut microbiota composition in the first six months after birth on neurodevelopmental outcomes at age 2 to 3 years (Rozé et al., 2020;Sordillo et al., 2019), three studies investigated cross-sectional associations between gut microbiota and neurodevelopment in 18-month-old (Acuña et al., 2021) or 3-year-old children (Rothenberg et al., 2021;Zhang et al., 2021), and two studies used a combination of longitudinal and cross-sectional approaches where either cognitive outcomes (Carlson et al., 2018) or gut microbiota (Kort et al., 2021) were assessed at multiple timepoints.Six of these studies included children born within a normal range of gestation and one studied preterm infants (Rozé et al., 2020).We discuss the results below, starting from the earliest timepoints at which microbiome sampling was conducted.

Table 2
Glossary of terminology used for measuring and characterizing the gut microbiome.

Sequencing methods
Gut microbiome is most commonly sequenced using one of two methods: • 16S ribosomal RNA (rRNA) amplicon-based sequencing utilizes the conserved and hypervariable regions within the bacterial 16S rRNA gene.The conserved regions serve as primer binding sites to amplify the (fragments of) 16S rRNA gene and the hypervariable regions that contain species-specific information are used to differentiate between different bacteria and identify the composition of the microbiota (what bacteria are there).• Metagenomic shotgun sequencing involves sequencing the entire DNA as opposed to the 16S rRNA marker gene.This method enables both classification of bacteria as well as identification of their functional potential (what they are doing).

Alpha diversity
The diversity of bacteria within a sample.Alpha diversity can be calculated using different measures that differentially take into account the richness (i.e.number of different bacterial groups; higher number = higher diversity) and/or evenness (i.e.distribution of abundances of the groups; more similar distributions = higher diversity).Some measures additionally take into account the phylogenetic relatedness of the community members (closely related bacteria = less diversity).Common measures of alpha diversity include Shannon index, Simpson index, Chao1, Faith's Phylogenetic Diversity.

Beta diversity
The diversity of bacteria between samples.Similar to alpha diversity, beta diversity can be calculated using different measures that take into account the presence/absence of species, abundance of species and/or value placed on rare species.Beta-diversity is calculated as a similarity/dissimilarity matrix and commonly presented in a principal component analysis (or similar reduced dimensionality) plot: if two samples are further apart on this plot, they have high beta diversity.High beta diversity between samples generally means that the community composition is highly different between these samples.Beta diversity is commonly used as a marker of overall bacteria compositional differences between study participants.Common measures of beta diversity calculation include Bray-Curtis dissimilarity, UniFrac distances, Aitchison distance.

Bacterial composition
In most instances, researchers are interested in which bacteria correlate with the outcomes of interest.This question is most commonly answered using one of the following approaches: • Individual taxa level analyses where relative abundances of specific bacterial taxa are associated with variables of interest.This analysis can be conducted on different hierarchical levels in the taxonomic classification (Phylum->Class->Order->Family->Genus->Species->Strain).• Dimensionality reduction analyses where: ○ the samples from study participants are clustered into groups based on the similarities in community composition; these clusters can then in turn be described by the dominant and/or discriminative bacteria ○ the abundances of bacteria are represented as factors or co-abundance groups where each can be described by the increasing/ decreasing abundance of specific taxa; study samples/participants in turn receive a score along these factors/co-abundance groups which are assessed for correlations with outcomes of interest  Associations between bacteria phyla and GDI scores (unadjusted analyses): • Firmicutes ~ Gross motor (rho = 0.327) With bacterial genera c : Specific bacterial taxa correlate with ASQ scores g : • Staphylococcus caprae r = − 0.15, p = 0.14 • Escherichia coli r = 0.12, p = 0.03 Rothenberg et al., 2021 Neurodevelopment correlates with gut microbiota in a cross-sectional analysis of children at 3 years of age in rural China China • n = 46 (28 male, 18 female) at 3 years (range 36.0-37.9months)  The model that best predicted m language scores at 36 months includes the three predictors which are all positively associated with language scores at 36 months (the standardized estimates from a linear regression model including the three predictors is given in brackets; adjusted R 2 = 0.31): • Language scores at 24 months (β = 0.44)  k Analysed using Fisher's exact test; covariates: maternal pregestational BMI and type of breastmilk up to 3 months of life.l Analysed DESeq2 using non-normalised raw count tables; covariates: maternal pregestational BMI and type of breastmilk up to 3 months of life; LFC = log 2 fold change (estimated from Fig. 3).m Best predictors selected using Mixed Integer Optimisation; predictors were selected from a total of 1170 potential predictors: one parameter indicating whether or not the mother of the child was included in the education intervention group, six anthropometric and cognitive parameters at 24 months, and 542 gut microbiota composition related parameters at 24 months and 621 parameters at 36 months; out of 60 best models including 2-4 parameters, language score at 24 months was included in 52 models, Coprococcus eutactus at 36 months in 42 models, and Bifidobacterium at 24 months in 19 models.
n Analysed using two-tailed Mann-Whitney U test for bacterial relative abundances; unadjusted analyses.

K. Vaher et al.
The two longitudinal studies showed that gut microbiome composition in early infancy associates with neurocognitive outcomes later in childhood.Specifically, by applying data reduction to bacterial composition data from young infants (3-6 months of life), Sordillo et al. identified four bacterial co-abundance groups and found that the factor representing increased abundance of Bacteroides and decreased abundance of Escherichia/Shigella and Bifidobacterium negatively correlated with fine motor skills, while the factor representing increased abundance of different Lachnospiraceae and Clostridiales taxa and decreased abundance Bacteroides negatively correlated with communication as well as personal and social skills at 3 years of life (Sordillo et al., 2019).Their taxa-level analyses suggested similar results on the order-and family-level, although the genera-level associations were often different.In particular, similar to the associations with the co-abundance groups, different genera in Clostridiales order (e.g.Ruminococcus, Oscillospira, Acidaminococcus) were less abundant in children with typically developing communication and social/personal skills compared to those with potential atypical or delayed development.However, the results for motor development were different from the coabundance group results, showing Streptococcus (Lactobacillales order) and Klebsiella (Enterobacteriales order) to have the strongest positive associations with motor scores, highlighting the effect of methodological variability on results.These latter findings are interesting as increased abundance of Streptococcus and Klebsiella in early infancy has also been associated with C-section birth (Reyman et al., 2019;Shao et al., 2019).The second longitudinal study was carried out in very and extremely preterm infants: Rozé and colleagues identified 5 clusters of preterm infants defined by their bacterial composition in the neonatal period, of which the Enterococcus-or Staphylococcus-dominated clusters were at higher risk for non-optimal 2-year neurodevelopmental outcome (defined as death or developmental delay (ASQ score < 185) compared to Escherichia/Shigella-, Enterobacter-or Clostridium-dominated clusters (Rozé et al., 2020).These cluster-level results were also in part paralleled by taxa-level analyses as a Staphylococcus species (Staphylococcus caprae) negatively and Escherichia coli positively correlated with ASQ scores (Rozé et al., 2020).
These results together suggest a potential role of increased abundance of Escherichia/Shigella and of bacteria in Enterobacteriales order (Klebsiella, Enterobacter) in early infancy for optimal neurocognitive development later in childhood, while the relationships with other bacteria (e.g. with members of Clostridiales order (Ruminococcus, Clostridium)) in this time period differ between the two studies.The differences could, at least in part, be explained by differences in study populations, timing of stool sampling, or other design features.Thus, further studies are required to determine the generalizability of these observations across populations.
Interestingly, the clusters characterized as having high abundance of Enterococcus or Staphylococcus mostly included infants born at an earlier gestational week (Rozé et al., 2020).This suggests that the composition of extremely preterm infants' gut microbiota might be more predictive of adverse neurodevelopmental outcomes (as compared to very preterm infants), and that the disrupted gut microbiota composition in extremely preterm neonates might have a larger effect on later neurodevelopmental outcomes compared to infants born at a later gestational week (Rozé et al., 2020).
Moving on from early infancy, Carlson et al. investigated whether gut bacterial composition in 1-year-old children correlates with neurocognitive development at the same time and/or at 2 years of life (Carlson et al., 2018).They identified three clusters of infants defined by their bacterial composition at 1 year of life, and found that the MSEL Early Learning Composite as well as expressive and receptive language subscale scores at 2 years of life were highest in the Bacteroides-and lowest in the Faecalibacterium-dominated clusters, but they found no differences between the clusters in cross-sectional, 1-year MSEL scores.These results are consistent with the findings in Sordillo et al. (2019), where the co-abundance group characterized by decreased abundance of Bacteroides showed negative correlations with communication and social development scores, suggesting a positive effect of increased Bacteroides during the first year of life on language/communication development later in childhood.Bacteroides is one of the most abundant bacterial taxa in the human gut, reaching its highest abundance around 1 year of life and then stabilizing to adult levels around 3-5 years of life (Fig. 1; Roswall et al., 2021).Increased abundance of Bacteroides in early infancy has also been associated with vaginal delivery (Bäckhed et al., 2015;Reyman et al., 2019;Shao et al., 2019) as well as with the cessation of breastfeeding around the first year of life (Bäckhed et al., 2015).Taken together, these results suggest that gut microbiome composition during the first year of life may be associated with neurocognitive outcomes at 2 and 3 years of life although the specific bacterial composition underlying these relationships varies between studies, possibly due to the different ages and populations studied.
In a cross-sectional study of 18-month-old children, Acuña et al. (2021) found that gut microbiota community structure was associated only with fine motor skills as assessed by BSID (3rd edition).By clustering infants into two community types, they observed that infants with higher fine motor scores were more likely to have Firmicutes-dominant (high abundance of Lachnospiraceae, Streptococcus and Blautia) compared to Bacteroides-dominant community type.Acuña et al. additionally demonstrated positive associations of fine motor scores with the abundance of Bifidobacterium, Corprococcus, Enterococcus, Lactobacillus, Holdemanella, Roseburia and Veillonella, and negative associations with the abundance of Parabacteroides.These results are partly consistent with the results by Sordillo et al. (2019), suggesting that increased abundance of Bacteroides in infancy and toddlerhood has negative effects on motor development, while Bifidobacterium and Streptococcus abundances may have beneficial effects on motor development in early childhood (Sordillo et al., 2019).These observations raise the possibility that some bacteria (e.g.Bacteroides) could have differential effects according to cognitive domain (e.g.language vs motor) and over time.
Gut microbiota composition during the second year of life may also be predictive of cognitive outcomes at 3 years.To understand how gut microbiota affects language development in 3-year-old children, Kort and colleagues used prediction modelling and identified that the best predictive model for 3-year language scores included language scores at 2 years, abundance of Coprococcus (from Lachnospiraceae family) at 2 years, and Bifidobacterium at 3 years of life, which all positively correlated with 3-year language scores (Kort et al., 2021).In individual taxa-level analyses, improved language at 3 years correlated with increased abundance of species in the genera of Coprococcus, Bifidobacterium, Faecalibacterium, Roseburia, Bacteroides and Clostridium, and decreased abundance of Parabacteroides and Escherichia/Shigella, among others, at 2 years of life.Importantly, these results provide further evidence that increased abundance of Bacteroides in the first two years of life positively correlates with language development, as reported previously (Carlson et al., 2018;Sordillo et al., 2019).
In a sample of 3-year-old children, Rothenberg et al. (2021) took a similar approach to Sordillo et al. (2019) in microbiome data reduction and found that the factor representing increased abundance of Faecalibacterium, Clostridium cluster XIVa, Gemmiger, Phascolarctobacterium, Alistipes, Oscillibacter, and Sutterella, and decreased abundance of Blautia, Anaerostipes, Clostridium cluster XVIII, and Streptococcus positively associated with both mental and psychomotor development assessed by the BSID (2nd edition).These results were partly consistent with taxa-level analyses as Faecalibacterium and Gemminger positively associated with psychomotor development, and Faecalibacterium and Clostridium XlVb positively associated with mental development.These results are also partly consistent with those observed by Kort et al. (2021) who reported that increased abundance of Faecalibacterium associates with improved cognitive (language) development.However, these results are in contrast to those observed in the study by Carlson et al. (2018), where high abundance of Faecalibacterium pointed towards lower cognitive and language scores, potentially suggesting that increasing abundance of Faecalibacterium after the first year of life has a positive effect on cognitive development while higher abundance of Faecalibacterium before that time may be unfavorable.Indeed, Faecalibacterium abundance in the gut starts to increase after 6 months of life, reaching adult levels around 3 years (Fig. 1; Roswall et al., 2021), allowing the speculation that exposure to a higher abundance of Faecalibacterium too early could be unfavorable for cognitive outcomes.
Interestingly, taxa-level analyses in Rothenberg et al. (2021) additionally identified Lachnospiraceae abundance as negatively associated with cognitive development, which is in line with the findings in the study by Sordillo et al. (2019) where the co-abundance factor represented by increased abundance of Lachnospiraceae species negatively associated with communication and social skills.However, these results conflict those observed by Kort et al., who reported that members of Lachnospiraceae family (Coprococcus, Roseburia) at 2 years of life positively associated with language development at 3 years (Kort et al., 2021).The results regarding Streptococcus demonstrated by Rothenberg et al. (2021) also differ to those reported by Sordillo et al. (2019) and Acuña et al. (2021), who showed that higher abundance of Streptococcus associated with improved motor development, suggesting that early colonization with Streptococcus correlates with optimal neurocognitive development, but later in development increased abundance of Streptococcus might have a negative effect on neurodevelopment.Streptococcus abundance is relatively high in newborns, but decreases rapidly over the first few months of life (Fig. 1; Roswall et al., 2021), suggesting that higher abundance of Streptococcus later in childhood is suboptimal.
A small study of 38 3-year-old children explored the associations between gut microbiome composition and neurodevelopment focused on different bacterial taxonomy levels and identified members of the Firmicutes phylum (e.g.Ruminococcus, Hungatella) to be positively and members of Bacteroidetes phylum (e.g.Porphyromonas, Butyricimonas) to be negatively associated with different subscales, particularly the motor scale, in the Gesell Developmental Inventory (Zhang et al., 2021).These results thus partly parallel those reported by Acuña et al. (2021) and suggest that members of the Firmicutes phylum may be positively associated with motor development, however, the specific genera-and species-level associations vary.
Four studies additionally evaluated associations between bacterial alpha diversity (within-sample diversity) and neurocognitive outcomes.Carlson et al. (2018) studied microbiota diversity in 1-year-old children and reported a negative association between different measures of alpha diversity and MSEL Early Learning Composite as well as expressive language and visual reception subscale scores at 2 years of life while, similarly to cluster analyses, they did not observe any cross-correlational associations with 1-year cognitive scores.Relationships between bacterial alpha diversity and neurodevelopment were tested in three other studies (Acuña et al., 2021;Rothenberg et al., 2021;Sordillo et al., 2019), and no significant associations were found between 3-year neurocognitive scores as assessed by the parent-reported ASQ (Sordillo et al., 2019) or observational BSID at 18 months (Acuña et al., 2021) or 3 years (Rothenberg et al., 2021) and 5-month, 18-month or 3-year microbial diversity, respectively.However, the interpretation of these results is complicated as different alpha diversity metrics differentially take into account the richness and evenness (see Table 2) of microbial community composition.

Gut microbiome associations with socio-emotional behaviors
Thirteen studies have investigated associations between gut microbiome and socio-emotional development, including temperament, behavioral dysregulation, social behavior, and stress (Table 4).Eleven of these studies have assessed the microbiota composition using 16S rRNA sequencing (Aatsinki et al., 2019(Aatsinki et al., , 2020;;Carlson et al., 2021;Christian et al., 2015;Fox et al., 2021;Loughman, Ponsonby, et al., 2020;Loughman, Quinn, et al., 2020;Sobko et al., 2020;Sun et al., 2020;Wang et al., 2020;Zhang et al., 2021), one study used whole metagenome sequencing (Kelsey et al., 2021), and one study used both approaches (Laue et al., 2020).The majority of these studies used questionnaires to characterize behavioral development (Infant and Early Childhood Behavior Questionnaire, Child Behavior Checklist (CBCL), Social Responsiveness Scale, Perceived Stress Scale for Children); and three have used direct observational behavioral measures (NICU Network Neurobehavioral Scale, emotional attention paradigm using eye-tracking, and laboratory fear paradigms (Mask Task and Strange Situation)).The results are discussed below, arranged based on the domain of socioemotional behavior studied: temperament, fear, behavioral dysfunction, and other traits (social development and stress).

Temperament
Infant temperament, defined as stable traits of reactivity and regulation (Rothbart & Gartstein, 2008), predicts social and attention behaviors in childhood (Abulizi et al., 2017) as well as personality and psychopathology in adulthood (Tang et al., 2020).Five studies have investigated correlations between gut microbiome composition and temperament using the Infant or Early Childhood Behavior Questionnaires (IBQ or ECBQ).Three of these studies used cross-sectional approaches whereby gut microbiome and temperament were        Beta-diversity at 1 year of life associated with SRS-2 scores y .
Bacterial taxa associate with SRS-2 scores z : • at 6 weeks: ○ from shotgun metagenomic sequence data: Flavonifactor plautii (Ruminococcaceae family; β = 0.002) ○ from 16S: no significant associations   measured at the same age, ranging from 1 month of life to 1-and 2-year-old children (Christian et al., 2015;Kelsey et al., 2021;Wang et al., 2020), one study aimed to characterize the predictive value of gut microbiota at early infancy (2.5 months) to temperament in later infancy (6 months) (Aatsinki et al., 2019), and one study investigated the relationship between gut microbiota at multiple timepoints during the first year of life and temperament in 1-year-old children (Fox et al., 2021).
During the first month of life, Kelsey et al. (2021) demonstrated that different Bifidobacterium species were more abundant in infants with higher levels of regulation/orienting and negative emotionality, while Streptococcus and Schaalia were more abundant in infants with lower levels of negative emotionality.The abundance of Bifidobacterium species in early infancy (2.5 months of life) may also predict later temperament traits: Aatsinki and colleagues (2019) used cluster analysis and showed that a cluster of infants characterized by high abundance of Bifidobacterium and Enterobacteriaceae had the highest levels of orienting/regulation compared to the clusters characterized by high abundance of Bacteroides or Veillonella.These cluster-based results are consistent with bacterial-taxalevel analyses as unidentified Bifidobacterium species positively correlated with regulation/orienting, surgency, and fear reactivity.Aatsinki et al. found several other bacterial taxa associated with temperament traits, including negative correlations between species in Clostridiaceae family or Bacteroides and negative emotionality, and between and Streptococcus and fear reactivity.These results lend further support to the notion that Streptococcus, Bacteroides as well as members of the families Enterobacteriaceae (e.g.Enterobacter) and Clostridiaceae (e.g.Clostridium) may play a role in neurocognitive development as discussed in the section Gut microbiota associations with neurocognitive development in early childhood.
Taking a longitudinal approach in exploring microbiota associations with temperament during the first year of life, Fox et al. identified early (1-3 weeks of life) and late infancy (12 months of life) as sensitive periods during which gut microbial community composition associated with temperament traits in 1-year-old children (Fox et al., 2021).Specifically, at 1-3 weeks of life, the abundance of Bifidobacterium, Lachnospiraceae and Collinsella positively and the abundance of Klebsiella negatively associated with surgency/extraversion at 1 year of life, while the abundance of Megamonas, Acidaminococcus and Ruminococcus at 1 year of life positively and the abundance of Lactobacillus negatively associated with negative affectivity at the same age.In contrast, microbiota composition at 2 or 6 months of life was not associated with temperament traits in 1-year-old children (Fox et al., 2021), suggesting that gut microbiota composition during periods of early colonization and at 1 year of life, around cessation of breastfeeding, may exert effects on emotional/temperament development.
In line with this, Wang et al. additionally identified specific bacterial associations with temperament traits in 1-year-old children: the abundance of Bifidobacterium positively and the abundance of Hungatella negatively associated with different regulation/orienting subscales (Wang et al., 2020).These results thus parallel the results in early infancy and suggest that the abundance of Bifidobacterium during the first few months of life (Aatsinki et al., 2019;Fox et al., 2021;Kelsey et al., 2021) as well as at 1 year of life (Wang et al., 2020) correlates with temperament traits (especially regulation/orienting and surgency/extraversion) during the first year of life.In contrast, Bifidobacterium levels did not correlate with temperament traits in 2-year-old children (Christian et al., 2015), suggesting that there may be temporal effects of specific bacterial taxa on temperament, in line with the temporal succession of gut microbiome (Fig. 1); however, larger studies using longitudinal assessments of both microbiome and temperament are needed to test this hypothesis.
There may also be sex-specific associations between gut microbiota composition and temperament traits.For example, in 2-yearold children, several bacteria (Ruminococacceae, Parabacteroides, Dialister, Rikenellaceae) abundances correlated with surgency/extraversion only in boys (Christian et al., 2015), and in early infancy, the reported effects of Bifidobacterium on regulation and surgency (Aatsinki et al., 2019) were not identified in girls in sub-group analyses.Conversely, in both studies gut microbiome composition in girls appears to be particularly correlated with fear reactivity, although different species-level associations were identified (Aatsinki et al., 2019;Christian et al., 2015).Sex-specific effects of microbiota disruption on brain and behavior in other contexts have been reviewed in detail elsewhere (for example see (Jaggar et al., 2020)).
In terms of the relationships between bacterial diversity and temperament, the studies report discrepant findings.While alpha diversity has been positively associated with surgency/extraversion in 2-year-old children (Christian et al., 2015), alpha diversity at 2 months negatively correlated with negative affectivity in 6-month-old infants (Aatsinki et al., 2019) as well as in 1-year-old infants (Fox et al., 2021), although the latter result did not reach statistical significance.Conversely, there were no significant associations observed between alpha diversity and temperament traits during the first month of life (Kelsey et al., 2021).It could thus be hypothesized that, similarly to bacterial compositional effects, there may be temporal effects of alpha diversity on temperament though additional research is needed.Additionally, alpha diversity of the gut microbiota increases greatly over the first few years of life, especially around 4-6 months of life when solid foods are introduced to the diet (Reyman et al., 2019;Roswall et al., 2021), making comparisons of alpha diversity correlations with temperament traits at different ages difficult.

Fear-related behavior
Whilst IBQ and ECBQ include fear reactivity as one of the subscales for negative affectivity trait, experimental approaches to study fear behavior may reveal more specific information about development of fear processing during infancy.Indeed, preclinical studies suggest that microbiome is involved in fear-and anxiety-related behaviors (Diaz Heijtz et al., 2011;Hoban et al., 2018).Aatsinki et al. (2020) reported that increased attentional bias to fearful faces assessed by an emotional faces eye-tracking task correlated with decreased abundance of Bifidobacterium, Lactobacillus, Prevotella and Haemophilus and increased abundance of Clostridium at 2.5 months of life.These results seem in contrast to those observed by the same group previously (Aatsinki et al., 2019) where abundance of Bifidobacterium positively and members of Clostridiaceae family negatively associated with parent-reported fear reactivity in early infancy, however, it is unclear to what extent negative emotionality/fear reactivity in temperament scales relates to attentional bias to fearful faces.
Moreover, in a small (n = 14-19) pilot study, Carlson et al. (2021) reported that gut microbiota composition in 1-year-old children is associated with fear reactions in a non-social, but not social, fear paradigm.Specifically, they showed positive correlations between members of Clostridiales order, Sutterella, Dialister and Erysipelotrichaceae family and fear reactions.Interestingly, they did not observe significant associations between gut microbiota and parent-reported fear behavior (IBQ questionnaire), further suggesting that comprehensive assessment of fear processing is needed to fully understand the effects of gut microbiota on neurobehavioral and emotional development.Nevertheless, given the exploratory nature and small sample of this study, the results need to be replicated in a future study.

Behavioral problems
The CBCL is a parent reported questionnaire assessing a broad range of emotional and behavioral problems, and has been used in three studies investigating the relationships between gut microbiome and child development (Loughman, Ponsonby, et al., 2020;Loughman, Quinn, et al., 2020;Zhang et al., 2021).
Loughman and colleagues used two cohorts in Australia to describe the relations between gut microbiome composition during various points during the first year of life (at 1, less than 3, and at 6, and 12 months) and 2-year CBCL scores (Loughman, Ponsonby, et al., 2020;Loughman, Quinn, et al., 2020).Gut microbiome composition at 1 month after birth did not significantly associate with behavioral dysfunction at 2 years (Loughman, Ponsonby, et al., 2020), while several bacteria at around 2 months of life in a separate study were more or less abundant in children with behavioral problems at 2 years of life compared to those without behavioral problems (see Table 4) (Loughman, Quinn, et al., 2020).For example, several Streptococcus and Bacteroides species were less abundant in children with behavioral problems (i.e.increased abundance of these bacteria associated with fewer behavior problems).This is in line with the studies discussed in the section on Temperament as the increased abundance of these genera were also associated with decreased negative emotionality and fear reactivity in infancy (Aatsinki et al., 2019;Kelsey et al., 2021) (see section Temperament).Interestingly, bacteria in Bifidobacterium genera had both positive and negative associations with 2-year problem behavior, suggesting that species-level associations may be important.
The discrepancies between the findings in two studies investigating relationships between gut microbiome in early infancy and 2year behavior problems (Loughman, Ponsonby, et al., 2020;Loughman, Quinn, et al., 2020) could be due to different characteristics of the study populations: the study reporting associations between bacterial genera and problem behavior was carried out in infants with colic who had a higher baseline prevalence of behavioral problems (Loughman, Quinn, et al., 2020).This is relevant because infantile colic has also been associated with aberrant intestinal microbiome (Dubois & Gregory, 2016), which raises the possibility that the potential to detect microbiome-behavior associations is higher in children with divergent gut microbiome and/or behavior compared to the general population.
Although there was no association between the gut microbiota at 6 months and problem behavior at 2 years after adjustment for covariates, at 12 months the abundance of Prevotella was significantly lower in children with behavioral problems at 2 years (Loughman, Ponsonby, et al., 2020).Interestingly, reduced abundance of Prevotella has also been observed in patients with autism spectrum disorder (Liu et al., 2019).
A small preliminary study exploring the associations between gut microbiome composition and problem behavior in 38 3-year-old children cross-sectionally identified the phyla Bacteroidetes and Actinobacteria to be positively, and Proteobacteria and Verrumicrobia (e. g.Akkermansia) to be negatively correlated with different CBCL subscales (Zhang et al., 2021).This suggests that different bacteria may be important for behavioral development after the first year of life compared to those reported in early infancy (Loughman, Ponsonby, et al., 2020) though additional research is needed to test this hypothesis.
Regarding bacterial alpha diversity, increased alpha diversity in early infancy positively associated with problem behavior at 2 years in infants with colic (Loughman, Quinn, et al., 2020).However, in typical development, alpha diversity during the first year of life does not appear to be statistically significantly associated with behavioral problems at 2 years of life, though the direction of effect was similar at all timepoints to that observed in infants with colic (Loughman, Ponsonby, et al., 2020).These results further suggest population-specific effects of gut microbiota on behavioral development and highlight the possibility that the effects of microbiome on behavioral development may be greater (and more detectable) in vulnerable populations as also suggested above.

Other behavioral traits
Three studies have directly assessed the association between gut microbiome composition and other behavioral traits: social impairment (Laue et al., 2020), stress-related behavior in neonatal unit (Sun et al., 2020), and perceived stress (Sobko et al., 2020).
Using a prospective, longitudinal assessment of gut microbiome composition via both 16S rRNA and shotgun metagenomic sequencing, Laue et al. (2020) demonstrated that increased abundance of several species within the Lachnospiraceae family (e.g.Ruminococcus gnavus and torques, Blautia producta) at 1 and 2 years associated with worse social behaviors in 3-year-old children assessed using the Social Responsiveness Scale (SRS); the associations between 3-year social behavior and gut microbiome at 6 weeks and 3 years were less pronounced.Importantly, these results are in line with two other studies discussed above which reported a negative association between the abundance of Lachnospiraceae in early infancy and 3-year communication and personal/social skills (Sordillo et al., 2019), as well as a negative association between abundance of Lachnospiraceae and cognitive development at 3 years (Rothenberg et al., 2021).The longitudinal design of the study by Laue and colleagues additionally highlights the possibility of temporal effects of microbiome on behavior and suggests that the timing of sampling and assessments is crucial in identifying potential correlations between the microbiome and behavior.The use of metagenomic data by Laue et al. (2020) additionally enabled inferences about functional pathways, revealing that pathways involved in the urea cycle (L-ornithine de novo biosynthesis) and vitamin B6 biosynthesis (superpathway of pyridoxal 5 ′ -phosphate biosynthesis and salvage) in 6-week and 1-year-old infants are positively associated with improved social behaviors.Bacterial alpha diversity, on the other hand, was not associated with SRS scores, further suggesting unknown effects of gut bacterial alpha diversity on neurodevelopment, as discussed in previous sections.
Two studies have investigated aspects of stress in relation to gut microbiome composition.Sun et al. (2020) used longitudinal sampling of gut microbiome in preterm infants during their stay in the neonatal intensive care unit and associated the composition of the gut microbiome to the stress score of the NICU Network Neurobehavioral Scale (NNNS) at term-equivalent age.The study identified four genera which had dynamic associations with the stress score during the neonatal period.The direction of effect for these genera-stress associations changed over the course of the first month of life, further implying that the specific relationships between microbiome and behavior detected is age-or time-dependent.The strongest effects for all of these genera were observed around the latest timepoints assessed (i.e., at 4 weeks of life) when the abundance of Enterococcus, Shigella, and an unknown bacteria in Enterobacteriales order negatively, while genus Veillonella positively correlated with stress scores.Interestingly, these results have some consistency with those observed in (Rozé et al., 2020) which reported that the clusters dominated by Escherichia/Shigella or Enterobacter (Enterobacteriales order) during the first postnatal month in preterm infants were less at risk for non-optimal neurodevelopmental outcomes at 2 years of life, and the abundance of species in Veillonella genera in 2.5-month-old infants have been both negatively and positively associated with 6-month temperament traits of regulation/orienting, surgency and fear reactivity (Aatsinki et al., 2019).These results together suggest a potential role for these bacteria (Enterococcus, Enterobacteriales (Enterobacter), Veillonella) during the first few months of life on outcomes both in the neonatal period as well as later in infancy and childhood.
The second study looking at the relationships between gut microbiome and stress in childhood primarily focused on investigating the effects of an outdoor nature-related intervention on gut microbiome composition and perceived stress in 2-5-year-old children, however, their secondary analyses also included direct assessment of gut microbiome profiles and perceived stress scores (Sobko et al., 2020).In this study, bacterial alpha diversity negatively associated with perceived stress levels, and the diversity within the Bacteroidetes phyla negatively associated with stress scores.

Microbiome associations with structural and functional neuroimaging features in childhood
We identified four studies reporting associations between gut microbiota profiles and brain imaging features in early childhood (Table 5).In three studies, the gut microbiota was assessed using 16S rRNA gene-based amplicon sequencing (Carlson et al., 2018(Carlson et al., , 2021;;Gao et al., 2019) and one study used metagenome sequencing (Kelsey et al., 2021).Brain structural and functional properties have been assessed in these studies using structural MRI (T1-and T2-weighted) (Carlson et al., 2018(Carlson et al., , 2021)), functional MRI (Gao et al., 2019), and functional Near-Infrared Spectroscopy (fNIRS) (Kelsey et al., 2021).Although these studies to date are limited by small sample sizes (maximum sample size is 63 for participants with both measures), they suggest potential relationships between the gut microbiome and brain structure and function.
All four studies used alpha diversity as a microbiome feature of interest.As an exploratory analysis, Carlson et al. demonstrated that in a sample of 27 children with structural T1-weighted MRI scans, alpha diversity of the gut microbial taxa at 1 year of life positively correlated with the volume of left precentral gyrus, right angular gyrus and left amygdala at 2 years of life; while alpha diversity was uncorrelated with global measures of brain volume such as total gray and white matter or ventricular volume at 1 year of life (Carlson et al., 2018).However, as Carlson et al. observed negative correlations between alpha diversity and cognition (i.e. in opposite direction to those observed with regional volumes; see section Gut microbiota associations with general neurocognitive development in early childhood), it is difficult to interpret the gut-brain structure findings.A more recent pilot study from the same group identified positive correlations between gut microbiota and fear responses (see section Fear-related behavior), but did not observe a correlation between bacterial alpha diversity at 1 month or at 1 year of life and volume of structures involved in fear processing (amygdala, hippocampus or medial prefrontal cortex (mPFC)) at 1 month or at 1 year (though note that n = 13-14 for these analyses) (Carlson et al., 2021).These results together may potentially suggest that bacterial alpha diversity correlations with brain structure become apparent later in infancy/toddlerhoodbut larger, longitudinal studies are needed to test this hypothesis.
Gut bacterial alpha diversity has also been a marker of interest in studies investigating associations between gut microbiota and brain function.Using resting state functional MRI Gao and colleagues demonstrated that gut bacterial alpha diversity in 39 1-year-old infants negatively correlated with functional connectivity between the left amygdala and thalamus as well as anterior cingulate cortex and insula at the same age, and positively correlated with functional connectivity between supplemental motor area and the inferior parietal lobule (Gao et al., 2019).Interestingly, the correlations of gut bacterial alpha diversity with the development of the connectivity between parietal and frontal brain areas may already be apparent earlier in infancy.Using fNIRS, Kelsey et al. reported that gut bacterial diversity positively correlated with connectivity within the fronto-parietal network in newborn infants (Kelsey et al., 2021).They also found taxa diversity to be linked with increased connectivity of the homologous-interhemispheric network.Making use of shotgun metagenomic data, which enables insight into functional properties of the microbiome, Kelsey et al. (2021) additionally demonstrated that an increase in the diversity of virulence factors correlated with increased connectivity of the homologousinterhemispheric network, suggesting that the taxa diversity association with functional connectivity may be driven by increases in virulence factors.
These findings together raise the possibility that bacterial alpha diversity in early life affects the structure and function of the amygdala, which plays a central role in fear and emotion regulation, frontal precentral areas involved in motor processing (e.g.precentral gyrus and supplemental motor area), and parietal areas involved in language, sensory and emotional processing (e.g.

Table 5
Summary of studies investigating associations between gut microbiome and structural and functional neuroimaging features.. • the left amygdala and thalamus (negative association; correlation coefficients between − 0.33 and − 0.55) • the anterior cingulate cortex and right anterior insula (negative association; correlation coefficients between − 0.31 and − 0.54) • the supplemental motor area and left parietal cortex (positive association; correlation coefficients between 0.29 and 0.41) Kelsey et al., 2021 Gut microbiota composition is associated with newborn functional brain connectivity and behavioral temperament USA • n = 63 (37 male, 26 female) for microbiome and brain scan; mean age [range]: 25 days [9-56] • resting state functional NIRS • three functional connectivity networks of interest based on signal correlations in regions of interest: ○ fronto-parietal network ○ default mode network ○ homologous-interhemispheric network Same as in Table 4 Taxa-level alpha diversity associates positively with: •   4 Nominally significant associations: • 1-month Weighted Unifrac PC1 was positively associated with 1-year mPFC volume (β = 22701, r 2 = 0.34, p = 0.046, n = 14) • 1-year Weighted Unifrac PC1 was negatively associated with 1-year amygdala volume (β = -201.3r 2 = 0.29, p = 0.034, n = 13) Nominally significant associations between bacterial genera at 1 month and brain volumes at 1 month h : Nominally significant associations between bacterial genera at 1 month and brain volumes at 1 year i : *overlapping cohorts.a covariates: older siblings, paternal ethnicity, and total intracranial volume.b covariates: caesarean section, paternal ethnicity, currently breastfeeding, total intracranial volume.c covariates: caesarean section, paternal ethnicity, currently breastfeeding, total intracranial volume.d For each voxel ANCOVA was used to test for microbiome effects; covariates: older sibling, paternal ethnicity, birth weight, postnatal age at scan, sex, twin status, maternal/paternal education, residual frame-wise displacement; effect determined as significant if voxel-wise p < 0.001 and cluster size threshold of 7 face-connected voxels.Post hoc cluster-level responses were generated using the average functional connectivity per subject for each cluster detected and then calculated Pearson correlation with alpha diversity.e covariates: antibiotics, delivery method, breastfeeding, infant age, birthweight and weight at study visit, gestational age, income, sex, head circumference at birth.f covariates: antibiotics, delivery method, breastfeeding, infant age, infant weigh at study, gestational age, income, sex, maternal depression, head circumference at birth.g analysed using LefSE; covariate adjustment not specified; ↓ and ↑ indicate whether a bacteria is enriched in low connectivity or high connectivity groups, respectively (groupings based on median split); LFC: log-2-fold change.
angular gyrus and inferior parietal lobule).However, current literature does not rule out the effects of microbial diversity on other brain networks or on global/generalized properties.In addition, the correlations of microbial alpha diversity with brain structure and function may be age-dependent as the effects of alpha diversity on brain functional connectivity appear to be present already in early infancy while the effects on brain structure become evident after the first year of life.Whilst all studies investigating the relationships between gut microbiome and brain development indexed by neuroimaging use bacterial alpha diversity as a feature of interest, the associations with bacterial community composition and specific bacterial taxa or their functionalities are less explored.Carlson and colleagues used a cluster analysis and reported that infants with Bacteroidesdominated microbial communities had the largest and infants with Ruminococcaceae-dominated microbial communities had the smallest volume of superior occipital gyrus at 1 year of life, and, conversely, infants with Bacteroides-dominated microbial communities had the smallest and infants with Ruminococcaceae-dominated microbial communities the largest volume of caudate nuclei at 2 years of life (Carlson et al., 2018).These results thus mirror the correlations between the microbiome community clusters and MSEL scores which were also only observed at 2 years (see section Gut microbiota associations with general neurocognitive development in early childhood).As with alpha diversity, there were no effects of microbiota clusters on global measures of brain volume, which implies that microbiota-brain interactions may affect multiple discrete neural systems rather than global brain growth.Another study by the same group also found suggestive correlations between gut microbiota composition and the volume of fear-and emotion-related brain structures at 1 month and 1 year (amygdala, hippocampus and mPFC) (Carlson et al., 2021).Namely, at 1 month, the relative abundance of Streptococcus associated with smaller hippocampal, amygdala and mPFC volumes, decreased abundance of Staphylococcus associated with larger amygdala and mPFC volumes, and increased abundance of Bacteroides associated with larger amygdala volume.In contrast, increased abundance of Bacteroides at 1 month associated with smaller mPFC volume at 1 year, whilst larger amygdala volume at 1 year was associated with increased abundance of Lachnospiraceae and decreased abundance of Enterobacteriaceae at 1 month of life.Therefore, the abundance of specific bacterial species may influence regional brain volumes throughout infancy, however, the specific bacteria-brain region relationships need to be replicated and expanded upon in larger studies.Kelsey et al. (2021) additionally identified several bacterial taxa associations with brain functional connectivity (as measured with fNIRS), including positive associations of Clostridium, Enterococcus and Bacteroides species with the fronto-parietal network connectivity, and negative associations of Bifidobacterium species with homologous-interhemispheric network connectivity (i.e., connectivity between bilateral channels on an fNIRS cap plausibly covering frontal, temporal, and parietal regions).However, it remains to be determined in future longitudinal studies whether the relationships reported by Kelsey et al. replicate and persist beyond the newborn period.
The brain areas associated with bacterial diversity or composition in childhood have also been of interest in adult studies.For example, gut microbiota profiles have been associated with structural connectivity in limbic circuitry involving areas of anterior cingulate cortex and amygdala (Tillisch et al., 2017), and functional connectivity of the insula (Curtis et al., 2019) and the amygdala (Zheng et al., 2020).Furthermore, preclinical studies have also shown neurobiological alterations in the brains of animals with disrupted gut microbiota, including volumetric and morphometric changes in the amygdala, hippocampus, and anterior cingulate cortex (Luczynski et al., 2016(Luczynski et al., , 2017;;Ogbonnaya et al., 2015), coupled with changes in myelination (Hoban et al., 2016;Lu et al., 2018) and expression of signaling molecules and neurotransmitters (Clarke et al., 2013;Desbonnet et al., 2015;Diaz Heijtz et al., 2011;Neufeld et al., 2011;Sudo et al., 2004).These studies can guide the hypotheses of future research with pediatric populations.
Importantly, incorporating brain imaging measures to the studies of microbiome-gut-brain axis may reveal the neurobiological pathways of gut microbiome influence on behavior.For example, Kelsey and colleagues suggest that bacterial taxa and virulence factor diversity associations with behavioral temperament (see above section Temperament) were mediated via homologousinterhemispheric connectivity (Kelsey et al., 2021).In addition, in the study by Gao et al., the connectivity between supplemental motor area and parietal cortex, which positively correlated with alpha diversity, also negatively correlated with Mullen Early Learning Composite scores at 2 years; however, they did not formally test for the possible mediation effect (Gao et al., 2019).These initial findings highlight the potential for this kind of research to reveal mechanistic pathways by which brain development links gut microbiome with behavioral and cognitive outcomes.

Summary of microbiome features associated with neurodevelopment
All 20 studies report some significant associations between microbiome features (clusters, taxa abundance, alpha diversity) and neurodevelopmental measures.However, differences in the direction and magnitude of bacteria-brain/behavior relationships between studies are substantial.Fig. 2 summarizes several lines of convergence and divergence of the gut bacterial influence on brain and behavior.
For example, although increased gut bacterial alpha diversity is generally considered a proxy for a healthy/stable microbiome (Shade, 2016) and is associated with improved health outcomes (e.g.reduced alpha diversity has been observed in patients with attention deficit-hyperactivity disorder and diabetes; Kostic et al., 2015;Prehn-Kristensen et al., 2018), it remains unclear from the published literature whether alpha diversity is associated with neurodevelopmental measures.Whilst neuroimaging studies suggested that alpha diversity in early life plays a role in the structure and function of several brain regions (including the amygdala, frontal precentral areas, and parietal regions), the associations with behavioral measures vary, with several studies reporting no significant associations between alpha diversity and features of cognitive or socio-emotional development.Importantly, recent systematic reviews have also reported inconsistent relationships between alpha diversity and psychiatric and developmental disorders such as depression, anxiety, autism spectrum disorder, schizophrenia and bipolar disorder (Ho et al., 2020;Nguyen et al., 2018;Sanada et al., 2020;Simpson et al., 2021).
K. Vaher et al.Regarding bacterial taxa, we identified that the abundance of several bacteria is associated with multiple aspects of neurodevelopment.For instance, the abundance of Bacteroides appears to be positively associated with language and behavioral development, whilst it has negative effects on motor development.In early infancy, the abundance of Enterococcus and Staphylococcus seems to associate with poorer and the abundance of Escherichia/Shigella, Klebsiella and Enterobacter with improved neurocognition; in contrast, the abundance of Enterococcus and Escherichia/Shigella during the second year of life correlates with better and worse neurocognitive outcomes, respectively, suggesting potential age-dependent effects of specific bacteria on cognitive development.In parallel, the abundance of Escherichia/Shigella, Streptococcus, Staphylococcus, Enterococcus, and Bacteroides during the first month of life correlates with brain cortical connectivity and/or the volumes of fear-and emotion-related brain structures (amygdala, hippocampus and mPFC).On the other hand, the bacteria in Clostridiales order (e.g.Clostridium, Faecalibacterium) and Lachnospiraceae family (e.g.Coprococcus, Roseburia) do not have a clear direction of effect on neurodevelopmental measures, with studies reporting both positive and negative effects of these bacteria on language, motor, and socio-emotional development.Lastly, abundances of Bifidobacterium, Lactobacillus, Bacteroides and Veillonella have been correlated with temperament traits and other measures of socio-emotional behavioral development, however, the directions of effect vary substantially between studies and behavioral measures.These inconsistencies could be explained by the vast methodological heterogeneity between studies, as discussed in the next section.

Methodological and statistical considerations, and avenues for future research
Whilst this review of 20 studies suggests that variations in gut microbiota composition during the first few years of life may be associated with neurocognitive and socio-emotional development and features of brain structure and function in early childhood, quantitative synthesis of microbiota-brain/behavior associations was not possible due to clinical and methodological heterogeneity across studies.Comparison of sequencing-based human microbiome studies is especially complex due to the multitude of systemic biases introduced in various stages of research (Nearing et al., 2021).We extracted data on study characteristics to collate sources of potential bias arising from clinical and methodological heterogeneity (Supplementary Table 1), and we discuss these items in turn below, focusing on potential solutions for future studies.

Sample size
One of the limiting factors in existing studies is the sample size.Although almost half of the studies were only able to combine microbiome and brain/behavior data from around 15-90 participants, several others were considerably larger and enabled associations between microbiome and neurodevelopmental data from > 200 individuals.The small sample sizes are especially observed in the four studies that combined metagenomics data with neuroimaging, underlining the preliminary nature of these results and the need for replication in future studies.Small sample size is a particularly limiting factor in microbiota studies due to the high dimensionality of the data, i.e. much larger number of variables (bacterial taxa) than the number of cases/samples, which together with the compositional nature of the data impose challenges in applying classical statistical methods (Johnstone & Titterington, 2009;Tsilimigras & Fodor, 2016).Similarly, neuroimaging studies are often limited by small sample sizes due to the high cost.Neuroimaging data is often extremely noisy and variable, which coupled with a lot of analytic freedom may lead to challenges in downstream analyses and variability in results, especially when the sample size is insufficient and the hypotheses require examining individual differences in neural measures.We propose that future studies use rigorous power calculations to inform sample sizes in order to limit both Type I and Type II errors and, where possible, constrain hypotheses and analysis approaches based on prior research.In addition, harmonization of sampling protocols and processing pipelines in consortia, and other data-sharing initiatives, could enable scale-up of studies, increase study power and provide essential validation sets.Harmonization of sampling and data processing could additionally promote comparability between studies and increase study power by allowing for meta-and mega-analyses.

Participant characteristics
Gut microbiome composition, cognition, behavior and brain structure-function in the studies included in this review has been analyzed at various timepoints from birth until three years of life.This provides a source of methodological heterogeneity that limits the generalizability of results across studies.It remains to be determined whether reported differences in bacteria-brain/behavior relationships are dependent on the temporal development of the gut microbiota and/or the neurodevelopmental measures analyzed, or on the differential effects of different bacteria on the brain at different neurodevelopmental timepoints in early childhood.Both the brain and gut microbiome undergo extensive changes in the first five years of life (Fig. 1), thus, associations between the bacterial profiles and neurodevelopmental outcomes may depend on the timing of sampling and assessments.Therefore, the interacting temporal dynamics of these two systems via the microbiome-gut-brain axis and the potentially predictive value of gut microbiota profiles at different timepoints during infancy on neurocognitive outcomes later in childhood need to be explored in future studies with longitudinal sampling.
In addition to participant age at sampling and assessments, another source of heterogeneity in relation to participant characteristics is the inclusion or exclusion of participants based on specific clinical/demographic criteria.In light of the focus on typical development, studies included in this review excluded participants with neurodevelopmental delays, major illnesses, fetal ultrasound abnormalities, or visual/hearing impairments.Some studies additionally excluded participants based on specific criteria that could affect microbiome composition such as antibiotic use prior to study visits, C-section delivery, or formula feeding.Finally, several studies excluded participants based on low gestational age at birth while a few others focused exclusively on preterm infants.We chose not to exclude these studies because preterm birth is not a neurodevelopmental disorder per se and many survivors have typical neurodevelopment.The use of different exclusion criteria impacts the ability to make direct comparisons of the results across studies and participant heterogeneities may need to be taken into account when drawing comparisons between studies.

DNA extraction and sequencing
DNA extraction has been suggested to be one of the steps that is at most risk of bias in microbiome research (Brooks et al., 2015;McLaren et al., 2019).Specifically, extraction efficiencies vary both between bacteria as well as extraction kits, resulting in higher or lower yields of DNA and biasing the detection of specific bacteria (Nearing et al., 2021).Moreover, systematic bias due to differences in DNA extraction protocols is increased in low biomass samples (Davis et al., 2019), such as those collected from young infants.The studies discussed in the current review have used a range of different DNA extraction kits and protocols, however, some studies have not specified the kit/protocol.This heterogeneity could thus partly underly the varying results between studies.
Nineteen studies discussed in this review used sequencing based on the amplification of hypervariable regions of the bacterial 16S rRNA gene and only two studies applied whole metagenome sequencing, which, in addition to genus-level identification of bacteria achieved by 16S rRNA-based sequencing, can more accurately identify bacterial species and strains, as well as the functional pathways (Kelsey et al., 2021;Laue et al., 2020).The higher cost and need for a larger amount of input DNA for shotgun metagenomic sequencing contribute to the continued prevalence of 16S rRNA sequencing.In addition, 16S rRNA sequencing could have higher sensitivity compared to metagenomics in low-microbial biomass samples (Hillmann et al., 2018;Pereira-Marques et al., 2019).Nevertheless, more widespread use of whole metagenome sequencing would support greater inference about the functional features, including metabolic pathways through which the gut microbiome influences brain development.Alternative methods have been introduced, however, to also infer functional information from 16S rRNA-based sequence data (e.g.PICRUSt (Langille et al., 2013)).In particular, the gut-brain modules developed by Valles-Colomer and colleagues enables assembly of pathways with neuroactive potential in bacteria and could be used in future studies for functional insights inferred from 16S rRNA sequencing (Valles-Colomer et al., 2019).
Another source of variation in the sequencing methodologies is the use of a range of different hypervariable regions for 16S rRNAbased sequencing.This could explain variations of results between the different studies because the choice of 16S rRNA region can significantly alter the estimation of taxonomic composition and diversity (Clooney et al., 2016;Rintala et al., 2017;Yang et al., 2016).

Microbiota features of interest and statistical approaches
The application of different statistical analyses to describe bacteria-brain/behavior relationships also varies between the studies.In particular, the choice of microbiome feature of interest differs between studies.
Several reports discussed above implemented different data reduction techniques to the microbiome data prior to associations with neurodevelopmental outcomes.These include clustering based on distance metrics applied to relative genus abundances (Aatsinki et al., 2019;Acuña et al., 2021;Carlson et al., 2018;Rozé et al., 2020) or factor representations/co-abundance groupings of bacteria based on the correlations of the abundances of bacterial taxa in the samples (Rothenberg et al., 2021;Sordillo et al., 2019).Whilst these methods are effective in reducing the number of multiple comparisons when associating the microbiota composition data with variables of interest (e.g., neurodevelopmental assessment scores), the comparisons across studies are complicated as the initial data reduction step (clustering or co-abundance grouping) is highly variable between studies.Additionally, even if the algorithms for initial data reduction are similar, the resulting microbiome groups or factors will be dependent on the samples present.For example, depending on the age of the children at the time of sample collection, the microbiome groups or factors are different due to the temporal nature of microbiome succession in the gut in early childhood (Bäckhed et al., 2015;Roswall et al., 2021;Stewart et al., 2018).Other participant characteristics (e.g., prematurity, infantile colic, exclusion of C-section-born children) might additionally affect the microbiome composition and thus the results of data reduction.
Clustering of the human gut microbiome samples initially led to the idea of enterotypes reflecting distinct compositions of microbial communities, however, it is debatable whether these enterotypes exist or how they should be defined (García-Jiménez & Wilkinson, 2019;Goodrich et al., 2014).Yet, the idea of combing existing large-scale microbiome datasets to define age-dependent bacterial clusters/factor compositions that could be applied in future studies is attractive as a means of standardization to facilitate comparisons between studies.Another interesting approach for data reduction of microbiome datasets is based on calculating microbiota age (Stewart et al., 2018;Subramanian et al., 2014), which captures the relative maturity of microbiota composition.These and similar approaches for data reduction of multidimensional microbiome data could be of particular utility in studies investigating the relationship between gut microbiome and multidimensional neuroimaging data (Jenkinson et al., 2012).
Analyses relating taxa-level bacterial data with neurodevelopmental outcomes offer additional, valuable insights as these are, in theory, easier to compare between different studies.However, reporting of the findings differs across studies, with descriptions of perfeature analyses on different taxonomical levels (operational taxonomic units/species, genus, family, order, phyla) and inconsistencies in reporting of effect sizes and adjustments for multiple comparisons.Moreover, there are a variety of per-feature association methods that have been applied to study brain/behavior associations with bacterial abundances (e.g., DESeq2, MaAsLin, voom, LefSE, standard correlation measures such as Pearson's or Spearman rank correlation, linear regression, Student's t-test or Wilcoxon signed rank test), which vary in options for normalization, data transformation, models, and associated statistical inferences (Mallick et al., 2017(Mallick et al., , 2021;;Nearing et al., 2022).Each of these methods have their limitations in terms of power, sensitivity, and rates of false positive and negative findings (Mallick et al., 2021;Nearing et al., 2022).Indeed, a recent study which compared 14 differential abundance testing methods demonstrated that the tools identified different sets of bacteria to be associated with features of interest, and that the tools have varying sensitivities to different aspects of the data (e.g.sample size, sequencing depth) and data pre-processing (e.g.transformation, normalization, rarefaction, prevalence filtering) (Nearing et al., 2022).These differences could underly some of the between-study variations in the bacteria-neurodevelopment relationships observed.In addition, metagenomic data has specific characteristics (noisy, non-normally distributed, sparse, zero-inflated) which call for special caution in using standard epidemiological multivariate statistical models.Importantly, microbiome data is compositional by definition and appropriate compositional data analytic methods in every step of the analyses (including transformation, normalization, distance calculations as well as correlation with variables of interest) are crucial to avoid spurious correlations (Gloor et al., 2017;Spichak et al., 2021;Tsilimigras & Fodor, 2016).
Nevertheless, irrespective of the method used (clusters, factor compositions, taxa-level analyses), biologically relevant microbiome-brain/behavior relationships could be expected to emerge from different analysis methods.In other words, the use of different analysis methodologies across studies could be seen as a strength if they produce comparable findings, demonstrating that these microbiome-brain/behavior relationships are robust and observable across methodologies.The implementation of multiverse analyses (Steegen et al., 2016) or consensus approaches (as suggested by Nearing et al. (2022)) could thus be valuable to help ensure results are robust to choices in statistical analysis pipelines.
Moreover, the studies discussed aimed to identify individual bacteria that correlate with brain properties or behavioral traits.However, the abundances of different bacterial species are often intertwined and groups of different bacteria are co-occurring, suggesting that use of methods that enable the identification of bacterial networks or consortia which associate with brain or behavior features may be important in future studies.Furthermore, trajectories and/or stability of microbiota composition during early life rather than composition at any specific timepoint may be important for neurocognitive outcomes, suggesting that longitudinal modelling of the microbiota profiles with neurocognitive outcome data could reveal novel associations within the microbiota-gutbrain axis in childhood.
In addition to identifying specific bacteria that correlate with brain and behavior, most studies have used within-sample microbial diversity (alpha diversity) as a measure of interest and these results are often discussed as separate or independent from the microbial community results.There are multiple indices that can be calculated as measures of alpha diversity (e.g., Chao, Shannon and Simpson indices, number of observed species, Faith's Phylogenetic diversity), which differently take into account the richness (number of taxonomic groups) and/or evenness (distribution of microbial abundances) of the microbial communities (see Table 2).Thus, depending on the measure used, alpha diversity may provide limited insights into the microbial community structure (Shade, 2016).In addition, alpha diversity of the gut microbiota increases greatly over the first few years of life, especially around 4-6 months of life when solid foods are introduced to the diet (Reyman et al., 2019;Roswall et al., 2021), making it difficult to compare microbial alpha diversity relationships with brain and behavior across different timepoints in early childhood.Alpha diversity measures may especially lack resolution in early infancy when the gut is heavily dominated by Bifidobacteria (Reyman et al., 2019).We propose that future studies discuss alpha diversity associations with brain and behavior in the context of the underlying microbial community structure in the samples.

Neurodevelopmental outcome assessments
To date, four studies have investigated the associations between childhood gut microbiota and brain development using neuroimaging.These studies took varying approaches for defining the outcome measures of interest and include both exploratory (e.g., volume of 90 gray matter regions) and hypothesis-driven (e.g., volume and connectivity of fear-and emotion-related brain structures, such as the amygdala and hippocampus) targets.Of note, current literature lacks explorations between gut microbiome and features of brain microstructure inferred from diffusion-weighted or magnetization transfer imaging, which enable inferences about the myelin content in the brain (Lazari & Lipp, 2021;Mohammadi & Callaghan, 2021).These investigations would be important as preclinical studies suggest changes in myelination (Hoban et al., 2016;Lu et al., 2018) and neuronal morphology (Luczynski et al., 2016) in rodents with disrupted gut microbiota.Measures of brain microstructure have previously been associated with diet-induced changes in the microbiome in rats (Ong et al., 2018;Tengeler et al., 2020) as well as with gut microbiome in adults (Fernandez-Real et al., 2015;Tillisch et al., 2017).Moreover, diffusion MRI in infancy and early childhood has a strong predictive value for neurocognitive outcomes (for example see (Ball et al., 2015;Barnett et al., 2018;Counsell et al., 2008;Duerden et al., 2015;Girault et al., 2019;Ouyang et al., 2020;Van Kooij et al., 2012)) and several factors that are established as important drivers of the microbiome development (e.g.breastfeeding, C-section delivery) are also associated with alterations in brain microstructure in infancy and childhood (Bauer et al., 2019;Blesa et al., 2019;Deoni et al., 2013Deoni et al., , 2018Deoni et al., ,2019;;Kar et al., 2021).However, it remains unexplored to what extent these associations may be mediated through the gut microbiome.In addition, magnetic resonance spectroscopy may provide another important avenue for exploration in microbiome-gut-brain axis studies as it can provide unique information about the metabolic and neurobiological substrates of the brain (Stovell et al., 2017), and has been applied in the study of the effects of gut microbiome alterations on the brain in preclinical (Janik et al., 2016;Menneson et al., 2019) as well as adult human studies (Ahluwalia et al., 2016;He et al., 2018).
Studies which have associated gut microbiota features with neurocognitive and behavioral outcomes have mostly relied on validated and standardized parent questionnaires (e.g.IBQ, ECBQ, ASQ, CBCL) or directly observed assessments (e.g.BSID, MSEL), spanning from one month to three years of life, which can be considered a major strength in these studies due to the standardized nature and wide applicability of these scales across populations.In addition, two other studies used experimental methods for the assessment of emotional attention (Aatsinki et al., 2020) and fear reactivity (Carlson et al., 2021), which revealed more nuanced bacteria-behavior relationships not captured by broad-scale assessments.  Sordillo et al., 2019, 4 Rozé et al., 2020, 5 Rothenberg et al., 2021,  In the light of the studies reporting microbiome differences in individuals with neurodevelopmental disorders (especially autism spectrum disorder and attention deficit-hyperactivity disorder (Jurek et al., 2020)), future studies could additionally benefit from incorporating early behavioral measures linked to those disorders.These may include experimental assessments of early social and attentional development, for example eye-tracking tasks measuring social cognition or attention maintenance/switching, or standardized tools such as the Autism Diagnostic Observation Schedule.The assessment of these behavioral measures in future studies would enable testing whether early life microbiome measures are predictive of later clinical diagnoses, and may also reveal novel mechanisms within the microbiome-gut-brain axis involved in the pathogenesis of certain neurodevelopmental disorders.
In addition, future studies looking into the relationships between the gut microbiome and early executive functions (e.g. using Behavior Rating Inventory of Executive Function or Early Executive Functions Questionnaire) may reveal information about the specificity vs generalizability of microbiome-behavior associations.In other words, these studies could help to understand whether the microbiome-behavior relationships are specific to certain domains of cognition (e.g., social cognition, language development) or whether these are reflective of a more domain-general impact (e.g., on executive functions).Understanding the specific/selective vs general impact of microbiome on behavior may additionally help to reveal the neural circuits/pathways linking microbiome development with neurodevelopment.
Furthermore, as suggested by Kelsey and colleagues, incorporating brain imaging measures to the studies of microbiome-gut-brain axis may reveal important mechanistic neurobiological pathways whereby the gut microbiome may influence brain and behavior that are difficult to detect when exclusively studying measures of overt behaviors (Kelsey et al., 2021).For example, by looking into the relationships between gut microbiome and brain structure/function inferred from neuroimaging, future studies may be able to clarify the timeline of the impact of the microbiome on child cognition and behavior as neuroimaging studies could reveal early neural signatures that would be detectable before certain behaviors become observable/testable in early childhood.In addition, this research could help elucidate whether the microbiome has a general or specific impact on neurodevelopment by testing whether the correlations between microbiome composition and brain structure/function are observed across the brain or only in particular brain regions.Thus, future studies integrating all three data types (bacterial sequencing, multimodal neuroimaging as well as standardized and experimental behavioral assessments) will be valuable to understand the links between the gut microbiome and neural circuits involved in cognition and behavior.

Covariates
Choice of covariates differs between the studies.Several studies took a combination of theoretical and data-driven approaches to identify the covariates to be included in the analyses, and include variables that have been previously associated with the gut microbiome or the neurodevelopmental outcome measure, and/or that associate with the gut microbiome and/or the neurodevelopmental outcome in univariate analyses in the study data.The most commonly used covariates included factors of the child (e. g., age at assessment/fecal sample, sex, gestational age at birth, birthweight, exposure to antibiotics and probiotics, breastfeeding status or length, mode of delivery), parents (e.g., ethnicity, education level, age, income/socioeconomic status, maternal psychiatric history, BMI) as well as specific variables directly related to microbiota (e.g., storage conditions) or outcomes (e.g., head circumference, cognitive scores at an earlier age).Nevertheless, there were some studies that did not adjust for any potential confounders (Christian et al., 2015;Kort et al., 2021;Sobko et al., 2020;Zhang et al., 2021).
Covariate adjustment is necessary to identify the effects of gut microbiota on brain development and behavior that are independent from other neurodevelopmentally important factors (e.g., birthweight, feeding, gestational age) and collection of such metadata relating to participant exposures and history enables measuring the effects of these exposures on the brain mediated by the gut microbiota.Furthermore, standardization of metadata collection in future studies would facilitate comparisons between the studies.

Adjustment for multiple comparisons
As discussed above, the studies included in the current review have analyzed a variety of microbiota and neurodevelopmental features of interest, leading to a multiple statistical tests correlating microbiome and brain/behavior measures within each study.Thus, adjustment for multiple comparisons is important to avoid false positive results.Indeed, the majority of the included studies performed some correction for false positivity rate, but seven studies did not describe adjustment for multiple testing, raising the possibility of type 1 error.

Conclusion
This review synthesized findings from 20 studies which investigated relationships between gut microbiome profiles and neurodevelopmental outcomes in typically developing children from birth to five years of age.The studies used a range of outcome assessments including structural and functional neuroimaging, standardized neurocognitive assessments, and parent-reported temperament and behavioral dysfunction questionnaires; gut microbiome data were analyzed using 16S rRNA or whole metagenome sequencing.Most studies report significant associations of microbiota clusters, abundance of specific taxa or alpha diversity with neurodevelopmental outcomes, however, differences in the nature of bacteria-brain/behavior relationships between studies are substantial.
We identified several sources of methodological heterogeneity, including variations in gut microbiota sequencing, features of interest, statistical analysis decisions, and adjustment for confounding factors and multiple comparisons which could, at least partly, explain the differences observed in the results.We propose that standardization of sampling and data processing pipelines could improve study power and enable validation of the current results in terms of the direction and magnitude of the effects.Adoption of reporting guidelines such as STORMS (Mirzayi et al., 2021) and pre-registration of analyses could enhance reliability and facilitate meta-analyses in future work.Importantly, however, apparent conflicting findings in the studies need to be considered within the substantial temporal development of gut microbiota, brain and behavior during pre-school age.We propose that future large studies with longitudinal microbiota sampling and neurodevelopmental assessments across early years of life are required to identify the interacting temporal dynamics of the microbiome and the brain via the microbiome-gut-brain axis in childhood.

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.

Fig. 2 .
Fig. 2. Bacterial features associated with aspects of neurodevelopment.↑, ↓, and "ns" indicate whether the bacterial feature was positively, negatively, and/or not significantly associated with cognitive, behavioral or neuroimaging measures.The bacterial families and genera presented in this figure are those that were identified as correlated with neurodevelopment in at least two studies and/or with at least two different neurodevelopmental domains.The superscripts are study references: 1 Carlson et al., 2018, 2 Zhang et al., 2021,3 Sordillo et al., 2019, 4 Rozé et al., 2020, 5 Rothenberg et al., 2021,

Table 3
Summary of studies investigating associations between gut microbiome and general neurocognitive development.

Table 4
Summary of studies investigating associations between gut microbiome and socio-emotional behaviors.

Table 4
(continued ) The strongest effect estimates are observed at around 26 days of postnatal life: Enterococcus, the unidentified Enterobacteriales genera and Shigella negatively, while genus Veillonella positively associate with NSTRESS scores.(continued on next page) K.Vaher et al.

Table 4
(continued ) *Overlapping cohorts.aAnalysed in boys and girls separately using Pearson's correlation; unadjusted analyses.bAnalysed in boys and girls separately using PERMANOVA; unadjusted analyses.cAnalysed in boys and girls separately using Spearman rank correlation; unadjusted analyses.dAnalysedusingordinaryleastsquarespathanalysis;covariate adjustment not specified.eAnalysedusingLefSE,covariateadjustmentnotspecified;↓ and ↑ indicate whether a bacteria is enriched in low or high temperament trait groups, respectively (groupings based on median split); LFC: log-2-fold change.fAnalysedusinglinearregression;covariates:infantsexand mode of delivery.gAnalysedusinglinearregression;covariates:gestationalage, infant age, sex, mode of delivery, breastfeeding and antibiotics intake.hAnalysedusingDESeq2;covariates:infantage,sex and mode of delivery; LFC = log 2 fold change; FDR-corrected q-value < 0.25 was considered statistically significant.ilAnalysedusinglinearregression;covariates:delivery mode, feeding type, and probiotic consumption.mAnalysedusingmultivariatelinearregression;covariates:infant sex and breastfeeding duration.nAnalysedusingPERMANOVA;covariates:infantsexand breastfeeding duration.oAnalysedusingmultivariatenegativebinomialmixedmodels in DESeq2; covariates: infant sex and duration of breastfeeding; LFC = log 2 fold change; FDR-corrected p-value < 0.1 was considered statistically significant.pAnalysedusingDESeq2;LFC=log 2 fold change; covariates: mode of delivery, breastfeeding, infant age at faecal sampling, maternal depressive symptoms at the end of pregnancy, and infant sex.qAnalysed using two-level mixed effects structure accounting for the within-subject correlations among the fear outcomes and within-subject but between-mask correlations of the outcome.rAnalysed the same way as in q .sProblembehaviourcase group defined as 1 SD above the mean of reference population (T ≥ t Analysed using PERMANOVA.uAnalysedusingvoomon normalised abundance of taxa: using relative log expression with pseudo-counts, scaled the counts for each sample by the median across OTUs of the OTU count divided by the geometric mean of that OTU's count across all samples; unadjusted analyses, corrected for multiple comparisons; LFC = v Behaviour problem group was defined as a score of T ≥ 60 (1 SD above the population normed mean) on one or more of the three CBCL subscales (n = 20 of problematic behaviour case group in this study (17 %)); covariates: sex, probiotic randomisation group, child age at baseline and maternal mental illness (at child age 2 years).wAnalysedusingDESeq2; groups defined and same covariates as in v; results are FDR corrected.xbacterial genera reported here based on visual inspection of the heat map for correlations of the highest magnitude with CBCL scores; analyses unadjusted, effect sizes and significance level not reported.y Analysed using PERMANOVA; effect sizes not given; covariates: child age at follow-up, maternal education, maternal marital status, parity, maternal and paternal age at delivery, and child sex; results not significant if additionally adjusted for maternal smoking during pregnancy, early postnatal exclusive breastfeeding, delivery mode, perinatal antibiotic exposure, and continuous gestational age at delivery.z Analysed using MaAsLin; β-values are unstandardised (refer to % change of relative abundance per point increase in SRS scores); FDR-corrected q-value < 0.25 was considered statistically significant; covariates: child age at follow-up, maternal education, maternal marital status, parity, maternal and paternal age at delivery, and child sex.aa Analysed as in z .bb Analysed using sparse log-contrast regression with functional compositional predictors; covariates: sex, delivery type, premature rupture of membranes, score of SNAPPE-II (acute physiologyperinatal extension), birthweight, % of feeding with mother's breastmilk.cc Analysed using Spearman correlation; unadjusted analyses.K. Vaher et al.