The associations between gut microbiota and fecal metabolites with intelligence quotient in preschoolers

Abstract

Background

The awareness of the association between the gut microbiota and human intelligence levels is increasing, but the findings are inconsistent. Furthermore, few research have explored the potential role of gut microbial metabolites in this association. This study aimed to investigate the associations of the gut microbiota and fecal metabolome with intelligence quotient (IQ) in preschoolers.

Methods

The 16 S rRNA sequencing and widely targeted metabolomics were applied to analyze the gut microbiota and fecal metabolites of 150 children aged 3–6 years. The Wechsler Preschool and Primary Scale of Intelligence, Fourth Edition (WPPSI-IV) was used to assess the cognitive competence.

Results

The observed species index, gut microbiome health index, and microbial dysbiosis index presented significant differences between children with full-scale IQ (FSIQ) below the borderline (G1) and those with average or above-average (all P < 0.05). The abundance of Acinetobacter, Blautia, Faecalibacterium, Prevotella_9, Subdoligranulum, Collinsella, Dialister, Holdemanella, and Methanobrevibacter was significantly associated with preschooler’s WPPSI-IV scores (P < 0.05). In all, 87 differential metabolites were identified, mainly including amino acid and its metabolites, fatty acyl, and benzene and substituted derivatives. The differential fecal metabolites carnitine C20:1-OH, 4-hydroxydebrisoquine, pantothenol, creatine, N,N-bis(2-hydroxyethyl) dodecanamide, FFA(20:5), zerumbone, (R)-(-)-2-phenylpropionic acid, M-toluene acetic acid, trans-cinnamaldehyde, isonicotinic acid, val-arg, traumatin, and 3-methyl-4-hydroxybenzaldehyde were significantly associated with the preschooler’s WPPSI-IV scores (P < 0.05). The combination of Acinetobacter, Isonicotinic acid, and 3-methyl-4-hydroxybenzaldehydenine may demonstrate increased discriminatory power for preschoolers in G1.

Conclusion

This study reveals a potential association between gut microbiome and metabolites with IQ in preschoolers, providing new directions for future research and practical applications. However, due to limitations such as the small sample size, unclear causality, and the complexity of metabolites, more validation studies are still needed to further elucidate the mechanisms and stability of these associations.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12866-024-03579-9.

Background

The gut microbiota has emerged as a critical factor in the regulation of neurodevelopment and cognitive functions through the microbiota-gut-brain axis (MGBA) [1]. This complex bidirectional communication system involves neural, endocrine, immune, and metabolic pathways that link gut microbiota with brain function [2]. The early years of life are particularly important, as this is when both the gut microbiota and brain experience significant development [3]. The dynamic establishment of the gut microbiota during this period has been suggested to play a crucial role in shaping cognitive outcomes, including intelligence quotient (IQ) [4, 5].

Recent studies have underscored the potential impact of gut microbiota composition on cognitive development. For instance, studies in animal models have shown that alterations in gut microbiota can influence neurogenesis, synaptic plasticity, and behavior [6, 7]. In humans, there is growing evidence that the diversity and specific composition of the gut microbiota are associated with cognitive functions. For example, specific bacterial genera have been linked to cognitive performance in children, suggesting that gut microbiota may contribute to the variability in IQ [8–10].

Beyond the composition of the microbiota, gut-derived metabolites are now recognized as key mediators in the MGBA. Short-chain fatty acids (SCFAs), such as acetate, propionate, and butyrate, are among the most studied metabolites. These SCFAs can influence brain function by modulating the integrity of the blood-brain barrier, affecting neurotransmitter synthesis, and altering immune responses [11, 12]. Additionally, other metabolites such as bile acids, tryptophan derivatives, and secondary metabolites produced by gut microbiota have been implicated in cognitive processes [13, 14]. Despite these insights, the specific relationship between fecal metabolites and IQ in children remains largely unexplored [15, 16].

The preschool years represent a critical window for both gut microbiota establishment and cognitive development [17, 18]. During this period, the gut microbiota undergoes significant maturation, which coincides with rapid brain development [18]. Disruptions in the gut microbiota during this time may have lasting effects on cognitive outcomes [18]. Studies have begun to explore these associations, but the evidence is still preliminary, and the mechanisms remain unclear [19, 20].

Although the potential links between gut microbiota, fecal metabolites, and cognitive outcomes have been proposed, there are significant gaps in the current literature. Most studies to date have focused on specific neurodevelopmental disorders or general cognitive functions rather than directly examining IQ in healthy children [2, 20, 21]. Furthermore, the role of gut metabolites as mediators of microbiota-brain interactions in relation to IQ has not been fully addressed [22, 23]. This study aims to fill these gaps by investigating the associations between gut microbiota composition, fecal metabolites, and IQ in preschoolers. By doing so, it seeks to contribute to a deeper understanding of how early-life gut microbiota may influence cognitive development and to identify potential biomarkers for cognitive outcomes.

Materials and methods

Participants

The study participants were from an ongoing longitudinal study, Guangxi Zhuang Birth Cohort (GZBC), Baise, China [24], including 150 mother-child pairs with comprehensive covariate information. This study focused on children between 3 and 6 years old who had IQ scores available and provided fecal specimens for analysis. Children who were given probiotics within the first 6 months of life; had undergone surgery or been hospitalized for intestinal disease; had respiratory disease, diarrhea, rash, or hand, foot and mouth disease in the past 6 months; used antibiotics in the last month; gestational weeks ≤ 28; had a birth weight of less than 1500 g; twins; or had any congenital disabilities or other medical conditions were excluded. Approval was obtained from the Guangxi Medical University Ethics and Research Committees (Approval No.20140305-001). A written informed consent was obtained from all participants.

The covariates identified from previous literature encompassed the following variables: maternal pre-pregnancy body mass index (BMI, kg/m²), passive smoking, and drinking obtained through a structured self-administered questionnaire; maternal age at delivery, delivery mode, and gestational age extracted from the electronic medical records; child’s age, gender, BMI, and antibiotic medication usage in the past month; household income and maternal education obtained prior to cognitive ability testing in July–August 2021. Cognitive function was assessed in children by using the Chinese version of the Wechsler Preschool and Primary Scale of Intelligence, Fourth Edition (WPPSI-IV CN). The WPPSI-IV comprises five primary subscales: the verbal comprehension index (VCI), visual spatial index (VSI), working memory index (WMI), fluid reasoning index (FRI), and processing speed index (PSI). The scores were standardized and presented as standardized scores, with a mean value of 100 and a standard deviation (SD) normalized to 15 [25]. Standardized scores were produced by comparing the raw scores of five subtests and standardizing them against a specific reference population, and then full-scale IQ (FSIQ) was derived from the five domain subscales. Subjects were included in the analysis if they completed all tests, leading to the exclusion of 15 children. Quality control was implemented through professional training and certification of all examiners, environmental control during testing, double-blind data entry and verification of raw data, accurate calculation of standard scores, data protection, and ensuring the accuracy of result interpretation.

Fecal specimen collection

Fecal specimens from preschoolers were collected by parents on the day of the cognitive function assessed or within 1–3 days after the assessment. The specimens were collected using a standard fecal specimen kit (CY-F002-35, HCY Technology Co., Ltd.), which involved taking a small portion from fresh stool (within 5 min of defecation) and placing it into a sampling tube containing microbial preservative fluid. The preservative fluid was composed of potassium chloride, magnesium chloride, Tris buffer, anticoagulants, and ethanol. The specimens were transported to the project hospital at room temperature or sent to the laboratory via mail service. The time interval between defecation and storage varied depending on the mode of transport, but all samples were kept frozen at − 80 °C until further handling. An additional eight samples were excluded due to improper sample collection, such as insufficient sample volume and loss of preservative fluid.

DNA extraction and 16 S rRNA sequencing

The CTAB (Hexadecyl trimethyl ammonium Bromide) method was used to extract the total genome DNA from the samples. Afterwards, the DNA samples were diluted to 1 ng/µL by using sterile water. The amplified region is 16 S V3-4, and the length of the amplicon ranges from 450 bp to 550 bp. The diluted genomic DNA served as the template for PCR amplification of bacterial 16 S rRNA genes by using the primers 515 F-806R. The reliability and precision of amplification were guaranteed using New England Biolabs’ Phusion High-Fidelity PCR Master Mix. The PCR product mixture was purified using the Qiagen Gel Extraction Kit. Subsequently, sequencing libraries were generated according to the manufacturer’s guidelines using the TruSeq DNA PCR-Free Sample Preparation Kit, with index codes integrated into the process. Finally, the library underwent sequencing on the Illumina NovaSeq platform.

16 S rRNA sequence analysis

Sample data were demultiplexed using barcode sequences and PCR primer sequences. After removing barcode and primer sequences, FLASH (V1.2.7) [26] was used to merge paired-end reads, resulting in Raw Tags. These Raw Tags underwent stringent filtering [27] to generate high-quality Clean Tags. Following Qiime’s (V1.9.1) [28] quality control protocol, the steps included: (a) Tag trimming: Raw Tags were truncated at the first low-quality base (defined as a quality score < = 19) after a sequence of three low-quality bases; (b) Length filtering: Tags with a sequence length less than 75% of the original after trimming were removed. After these steps, chimeric sequences were identified and removed by aligning the Tags with a species annotation database using the tool Vsearch (https://github.com/torognes/vsearch/) [29], yielding Effective Tags [30]. Clustering of all Effective Tags across samples was performed using the Uparse algorithm (Uparse v7.0.1001) [31] at a 97% similarity threshold to generate Operational Taxonomic Units (OTUs). The most frequent sequence within each OTU was selected as its representative. Taxonomic annotation of OTUs was conducted using the Mothur method, aligned with the SSUrRNA database of SILVA138.1 [32, 33], with a confidence threshold of 0.8 ~ 1. This provided taxonomic information across levels: kingdom, phylum, class, order, family, genus, and species. MUSCLE (V3.8.31) [34] was used for multiple sequence alignment, establishing the phylogenetic relationships among OTU representative sequences. Relative abundance of gut microbiota was calculated by dividing the number of sequences assigned to each OTU by the total number of sequences in the sample, expressed as a percentage. For diversity analyses, data were rarefied to a uniform sequence depth (the lowest number of sequences is 54,270). Alpha and Beta diversity analyses were subsequently conducted based on the rarefied data, with sequence counts proportionally subsampled for each OTU.

Metabolomic profiling

Fecal metabolites underwent targeted metabolomic analysis using the ABSciex QTRAP 6500 + LC-MS/MS platform by Metware Biotechnology Co., Ltd (Wuhan, China). The process began by thawing the samples initially kept at − 80 °C on ice. Subsequently, a 400 µL solution (composed of methanol and water in a ratio of 7:3) with internal standards was introduced to 20 mg of the specimen and vortexed for 3 min. Following this, the specimen underwent 10 min of sonication in an ice bath, followed by a 1-min vortexing, and placed in a − 20 °C environment for 30 min. Following centrifugation at 12,000 rpm and 4 °C for 10 min, 200 µL of the supernatant was extracted for LC–MS analysis. The sample extracts were analyzed using an LC-ESI-MS/MS system (UPLC, ExionLC ADhttps://sciex.com.cn/; MS, QTRAP^® System, https://sciex.com/). The analytical conditions were as follows, UPLC: column, Waters ACQUITY UPLC HSS T3 C18 (1.8 μm, 2.1 mm*100 mm); column temperature, 40 °C; flow rate, 0.4 mL/min; injection volume, 2 µL; solvent system, water (0.1% formic acid): acetonitrile (0.1% formic acid); gradient program, 95:5 V/V at 0 min, 10:90 V/V at 11.0 min, 10:90 V/V at 12.0 min, 95:5 V/V at 12.1 min, 95:5 V/V at 14.0 min. LIT and triple quadrupole (QQQ) scans were acquired on a triple quadrupole-linear ion trap mass spectrometer (QTRAP), QTRAP^® LC-MS/MS System, equipped with an ESI Turboon-Spray interface, operating in positive and negative ion mode and controlled by Analyst 1.6.3 software (Sciex). The ESI source operation parameters were as follows: source temperature 500 °C; ion spray voltage (IS) 5500 V (positive), -4500 V (negative); ion source gas I (GSI), gas II (GSII), curtain gas (CUR) were set at 55, 60, and 25.0 psi, respectively; the collision gas (CAD) was high. Instrument tuning and mass calibration were performed with 10 and 100 µmol/L polypropylene glycol solutions in QQQ and LIT modes, respectively. A specific set of MRM transitions were monitored for each period according to the metabolites eluted within this period.

The metabolomic data underwent unit variance scaling and was then subjected to unsupervised principal component analysis using the prcomp function within R. Hierarchical cluster analysis of metabolites was conducted using ComplexHeatmap, visually represented as heatmaps where normalized signal intensities (unit variance scaling) are displayed in a color spectrum. Correlation coefficients between samples were caculated by the cor function in R and presented as only heatmaps. Differential metabolites were identified on the basis of VIP scores and P values (VIP > 1 and P < 0.05). Significantly regulated metabolites between groups were determined by VIP > = 1 and absolute Log₂FC (fold change) > = 1. VIP values were extracted from orthogonal partial least square discriminant analysis (OPLS-DA) result, which were generated using the MetaboAnalystR package in R. The data was log transform (log₂) and mean centering before OPLS-DA. In order to avoid overfitting, a permutation test (200 permutations) was performed. The identified metabolites were annotated by referencing the KEGG compound database (http://www.kegg.jp/kegg/compound/) and subsequently aligned with the KEGG pathway database (http://www.kegg.jp/kegg/pathway.html). Their significance was assessed using hypergeometric test P values, providing insights into the metabolic pathways associated with the identified metabolites.

Statistical analyses

Data analyses were conducted by R (version 4.2.2) and IBM SPSS (version 25). Participant characteristics were described as means (standard deviations) or medians (interquartile range) for continuous variables and numbers (frequencies) for categorical variables. Chi-square test was used to analyze differences between categorical variables, while independent sample t-test was employed for normally distributed continuous variables. For non-normally distributed continuous variables, Wilcoxon rank sum tests were utilized to assess the differences. This study utilized the Gut Microbiome Health Index (GMHI), a biologically interpretable mathematical model, as a comprehensive indicator to assess the health status of the gut microbiota [35]. Additionally, to evaluate the degree of microbial dysbiosis between the G1 and G2 groups, the Microbial Dysbiosis Index (MDI) was calculated according to the method established by Gevers et al. [36]. Spearman analysis was used to assess the association between the top 30 abundant genera and fecal differential metabolites. Additionally, the Benjamini and Hochberg’s false discovery rate was used to correct the correlated P values.

Generalized linear models (GLMs) were employed to investigate the associations between WPPSI-IV scores and various measures, including the relative abundance of genera, alpha diversity, and metabolite abundance. The results are presented as standardized betas (β) along with their 95% confidence intervals (CIs). Due to the skewed distribution of gut microbiome and fecal metabolome abundances, these variables were log₂-transformed to treat as continuous variables in the GLMs. The random forest model employs the Bootstrap sampling method to create multiple distinct training sets, with each decision tree independently trained on these bootstrapped datasets. Model performance is evaluated using out-of-bag (OOB) error. This model is a classification model that incorporates metabolite and microbiome data along with confounding factors in the analysis. The analysis was conducted using the varSelRF function from the R package varSelRF. Initially, a random forest analysis was performed using all variables, followed by the sequential removal of less important variables. This process was repeated until the model with the lowest OOB error was obtained, and the variables from this model were selected as the optimal set. Cross-validation was then performed, and the receiver operating characteristic (ROC) curve was plotted. In the model evaluation, variable importance was assessed using two metrics: MeanDecreaseAccuracy and MeanDecreaseGini. MeanDecreaseAccuracy represents the decrease in model prediction accuracy when a variable is removed, with higher values indicating greater importance. MeanDecreaseGini evaluates the impact of a variable on the heterogeneity of the observations at each node of the classification tree, with larger values indicating higher importance. FSIQ scores were categorized into two categories to assess the potential clinical significance of the gut microbiota and fecal metabolites: G1, indicating the need for monitoring or further assessment (scores below the borderline), and G2, indicating normal development (average or above, Table S1). Inter-group analysis considered microbiome diversity, the abundances of the top 30 genera, and the differential fecal metabolites. A two-sided P < 0.05 was considered statistically significant.

Results

Summary of participant characteristics

As shown in Table 1, the mean (SD) age of children in the study was 4.67 (0.68) years, with gestational age of 38.54 (1.34) weeks. Out of the 150 children, 80 (53.33%) were boys, 112 (74.67%) were spontaneous labor, and 92 (61.33%) were breastfed for a minimum of 6 months in the early life. The mean of maternal pre-pregnancy BMI and age at delivery were 19.82 (2.92) kg/m² and 29.39 (5.14) years, respectively. A total of 98 (65.33%) mothers received high school education or higher. Approximately 84.00% of families reported an annual household income of less than 150,000 yuan per year. The means for the WPPSI-IV scores were as follows: VCI [85.52 (11.19)], VSI [85.52 (11.19)], WMI [90.36 (12.87)], FRI [99.94 (12.38)], PSI [100.38 (9.24)], and FSIQ [87.54 (11.62)]. Most children in this analysis met the average of FSIQ. However, 45 (30.00%) of the 150 children had FSIQ scores below the borderline, suggesting the need for monitoring or further assessment. The G1 and G2 groups were different in terms of maternal education, pre-pregnancy BMI, household income, mode of delivery, child age, and child IQ (all P < 0.05).

Table 1.

Characteristics of the mother-child pairs [Mean (SD)/Number (%)]

Characteristics	Total (n = 150)	G1 (n = 45)	G2 (n = 105)	P
Maternal age at delivery (years)	29.39 (5.14)	29.02 (5.68)	29.55 (4.91)	0.564
Pre-pregnancy BMI (kg/m2)	19.82 (2.92)	19.03 (2.82)	20.16 (2.92)	0.029
Maternal education				0.004
Less than high school	52 (34.67)	20 (44.44)	32 (30.48)
High school	38 (25.33)	16 (35.56)	22 (20.95)
College graduate or higher	60 (40.00)	9 (20.00)	51 (48.57)
Household income (yuan/year)				0.006
< 60,000	64 (42.67)	28 (62.22)	36 (34.29)
60,000-150,000	62 (41.33)	13 (28.89)	49 (46.67)
≥ 150,000	24 (16.00)	4 (8.89)	20 (19.04)
Mode delivery				0.002
Spontaneous labor	112 (74.67)	41 (91.11)	71 (67.62)
Ceasarean delivery	38 (25.33)	4 (8.89)	34 (32.38)
Sex				1.000
Boys	80 (53.33)	24 (53.33)	80 (53.33)
Girls	70 (46.67)	21 (46.67)	70 (46.67)
Gestational age (weeks)	38.54 (1.34)	38.60 (1.44)	38.51 (1.30)	0.721
Child age (years)	4.67 (0.68)	4.93 (0.52)	4.56 (0.71)	0.002
Breastfeeding duration (months)				0.558
< 6	58 (38.67)	19 (42.22)	58 (38.67)
≥ 6	92 (61.33)	26 (57.78)	92 (61.33)
Child IQ
VCI	85.52 (11.19)	76.36 (4.68)	89.45 (10.88)	< 0.001
VSI	92.88 (11.55)	85.93 (7.57)	95.86 (11.71)	< 0.001
WMI	90.36 (12.87)	82.09 (8.356)	93.90 (12.87)	< 0.001
FRI	99.94 (12.38)	90.26 (10.61)	105.57 (9.55)	< 0.001
PSI	100.38 (9.24)	94.72 (7.28)	103.68 (8.68)	< 0.001
FSIQ	87.54 (11.62)	76.04 (2.65)	92.47 (10.43)	< 0.001

G1, FSIQ Score ≤ 79; G2, FSIQ Score ≥ 80 Abbreviations SD, Standard deviation; BMI, Body mass index; IQ, intelligence quotients; VCI, verbal comprehension index; VSI, the visual space index; FRI, the fluid reasoning index; WMI, the working memory index; PSI, processing speed index; FSIQ, full-scale intelligence quotient

Overview of gut microbiome

A total of 2137 OTUs were obtained following taxonomic assignment (Table S2). The unique OTUs for the G1 and G2 groups were 245 and 762, respectively (Figure S1). The relative proportions of the dominant taxa at phylum and genus levels were evaluated by the classification of microbial taxa (Fig. 1). The top 10 dominant phyla accounted for 75.55%, including Firmicutes, Bacteroidota, and Proteobacteria (Fig. 1A), and the Firmicutes/Bacteroidota ratio in G1 showed a decrease compared to that in G2, but this difference was not statistically significant. The top 30 dominant genera accounted for 78.73%, including Bacteroides, Blautia, and Bifidobacterium (Fig. 1B). The average of the alpha diversity indices is presented in Table S3. The observed species index was statistically significantly higher in G1 than those in G2 (P = 0.032, Table S3). The beta diversity analysis is presented in Fig. 2. The unweighted UniFrac distances showed significant differences between G1 and G2 (P < 0.001, Fig. 2B). Additionally, there were significant differences in GMHI (P < 0.001, Fig. 3A) and MDI (P = 0.001, Fig. 3B) between G1 and G2. According to metastats analysis, the abundance of 10 genera out of the top 30 genera, including Acinetobacter and Prevotella_9, was lower in G2 than in G1, without statistical significance (Table S4).

Fig. 1.

Histogram of the relative abundance at phylum (A) and genus (B) levels. G1, FSIQ score ≤ 79; G2, FSIQ score ≥ 80

Fig. 2.

Beta analysis of gut microbiome in preschoolers.between G1 and G2. A and C, PCoA plot; B and D, boxplots. G1, FSIQ score ≤ 79; G2, FSIQ score ≥ 80. ^**, P < 0.001

Fig. 3.

Comparative analysis of GMHI and MDI between G1 and G2. A, GMHI comparison between G1 and G2; B, MDI comparison between G1 and G2. G1, FSIQ score ≤ 79; G2, FSIQ score ≥ 80. ^*, P < 0.05; ^**, P < 0.001

Associations between gut microbiome and WPPSI-IV scores

The associations between gut microbiome and WPPSI-IV scores are presented in Tabel S5 (crude model), Table S6 (adjustment model) and Fig. 4. In the adjustment models, Prevotella_9, Dialister, and Holdemanella were associated with lower VCI scores of − 0.56 (95% CI: −1.08, − 0.03), − 0.81 (95% CI: −1.51, − 0.11), and − 0.48 (95% CI: −0.81, − 0.15), respectively. Similarly, Holdemanella was associated with a lower FSIQ score of − 0.38 (95% CI: −0.72, − 0.05). Faecalibacterium and Subdoligranulum were associated with lower VSI scores of − 2.02 (95% CI: −3.58, − 0.45) and − 1.63 (95% CI: −3.13, − 0.13), respectively. For PSI, a 1-unit increase in Blautia and Collinsella was associated with 2.24-unit increase (95% CI: 0.48, 4.00) and 1.17-unit increase (95% CI: 0.00, 2.34) in the WPPSI-IV score, respectively. Acinetobacter was associated with a 0.51-unit decrease (95% CI: −0.96, − 0.06) in FRI scores, and Methanobrevibacter was associated with a 0.69 -unit increase (95% CI: 0.04, 1.33) in FRI scores. Overall, after adjustment for confounders, the gut microbiota diversity indices were found to be not associated with WPPSI-IV scores (Table S6).

Fig. 4.

Association between the relative abundance of nine of the top 30 genera and children’s intelligence quotient. Abbreviations 95%CI, 95% credible interval; A, VCI, verbal comprehension index; B, VSI, the visual space index; C, FRI, the fluid reasoning index; D, WMI, the working memory index; E, PSI, processing speed index; F, FSIQ, full-scale intelligence quotient. The models were adjusted for maternal pre-pregnancy BMI, maternal age at delivery, maternal education, household income, delivery mode, gestational age, breastfeeding duration, child age, and child sex

Potential functional effect of gut microbiome at different cognitive levels

Tax4Fun analysis was applied using SILVA database sequence as reference to cluster OTUs to predict the possible effects of the altered gut microbiome at different cognitive levels in preschoolers. The results showed that 460 KEGG orthologs (KOs) were statistically significant between G1 and G2 (Table S7). Moreover, metabolic pathways were altered, such as glycolysis/gluconeogenesis, peptidoglycan biosynthesis, galactose metabolism, amino acid biosynthesis and metabolism (alanine, aspartate, glutamate, glycine, serine, cysteine, methionine and threonine), amino sugar and nucleotide sugar metabolism, amino acid-related enzymes, peptidases, and transporters (Figure S2). Indeed, alterations in the structure and composition of gut microbiomes may affect their metabolic function.

Overview of fecal metabolome

A widely targeted metabolomic approach was used to conduct metabolome analysis of fecal samples. A total of 1593 metabolites were successfully quantified (Figure S3), of which 87 were differential metabolites, mainly including amino acid and its metabolites, fatty acyl, and benzene and substituted derivatives (Table S8 and Figure S3). Compared with those in G1, the levels of 64 metabolites, including carnitine C20:1-OH, N,N-bis(2-hydroxyethyl)dodecanamide, and creatine in G2 increased significantly, whereas those of 23 metabolites, including 4-Hydroxydebrisoquine, 3-methyl-4-hydroxybenzaldehyde, and trans-cinnamaldehyde decreased significantly (Table S8). The identified differential metabolites were mapped to 72 distinct KEGG pathways, encompassing processes such as biosynthesis of amino acids, thermogenesis, phenylalanine metabolism, and purine metabolism (Table S9 and Fig. 5).

Fig. 5.

KEGG enrichment map of differential metabolites. The horizontal coordinate represents the rich factor corresponding to each pathway, the vertical coordinate is the pathway name (show the top 20 pathways ranked by P-value), the color of the points reflects the P-value size, and the more red indicates the more significant enrichment. The size of the dots represents the number of differentiated metabolites enriched

Associations between the differential metabolites of gut microbes and WPPSI-IV scores

The associations between differential metabolites and WPPSI-IV scores are detailed in Table S10 (crude model), Table S11 (adjustment model) and Fig. 6. In the adjustment models, specific associations were observed: a 1-unit increase in 4-hydroxydebrisoquine was linked to a 1.28-unit decrease (95% CI: −2.18, − 0.39) in VCI score, a 1.17-unit decrease (95% CI: −2.20, − 0.14) in VSI score, a 1.22-unit decrease (95% CI: −2.40, − 0.04) in FRI score, and a 1.22-unit decrease (95% CI: −2.11, − 0.33) in FSIQ score. Creatine showed associations with a 1.97-unit increase (95% CI: 0.02, 3.92) in VSI scores and a 2.19-unit increase (95% CI: 0.11, 4.28) in WMI scores. N, N-bis (2-hydroxyethyl)dodecanamide was linked to a 1.58-unit increase (95% CI: 0.31, 2.85) in VCI scores and a 1.36-unit increase (95% CI: 0.09, 2.63) in FSIQ scores. For VCI, each unit increase in carnitine C20:1-OH, pantothenol, FFA(20:5), zerumbone, and 3-methyl-4-hydroxybenzaldehyde was associated with a 0.77-unit increase (95% CI: 0.11, 1.43), a 2.16-unit increase (95% CI: 0.26, 4.07), a 2.38-unit increase (95% CI: 0.35, 4.41), a 2.35-unit increase (95% CI: 0.02, 4.69), and a 1.55-unit decrease (95% CI: −2.73, − 0.38) in WPPSI-IV score, respectively. For FRI, each unit increase in (R)-(-)-2-phenylpropionic acid, M-toluene acetic acid, trans-cinnamaldehyde, isonicotinic acid, and val-arg was associated with 1.47 (95% CI: −2.78, − 0.16), 1.51 (95% CI: −2.82, − 0.2), 1.39 (95% CI: −2.66, − 0.12), 2.69 (95% CI: −4.78, − 0.61), and 1.47 (95% CI: −2.9, − 0.04) decreases in WPPSI-IV score, respectively. Meanwhile, traumatin was associated with a 11.22 (95% CI: −20.88, − 1.56) decrease in PSI score.

Fig. 6.

Association between the relative abundance of 14 differential metabolites and children’s intelligence quotient. Abbreviations: 95%CI, 95% credible interval; A, VCI, verbal comprehension index; B, VSI, the visual space index; C, FRI, the fluid reasoning index; D, WMI, the working memory index; E, PSI, processing speed index; F, FSIQ, full-scale intelligence quotient. The models were adjusted for maternal pre-pregnancy BMI, maternal age at delivery, maternal education, household income, delivery mode, gestational age, breastfeeding duration, child age, and child sex

A

Integrated analysis of key microbes and differential metabolites

Spearman’s correlation analysis was performed to reveal metabolites associated with microbes by using the data from the relative abundance of the top 30 genera and the differential metabolites (Fig. 7A and Figure S4). As shown in Fig. 7A, the fecal metabolites that decreased in the G2 group, such as isonicotinic acid, trans-cinnamaldehyde, and 3-methyl-4-hydroxybenzaldehyde, exhibited a positive association with the relative abundance of Faecalibacterium and Subdoligranulum. Meanwhile, isonicotinic acid, FFA(20:5), and traumatin were inversely associated with the relative abundance of Blautia. The relative abundance of Acinetobacter was positively associated with FFA(20:5) and inversely associated with N, N-bis(2-hydroxyethyl)dodecanamide. A random forest regression model was constructed using a combination of meaningful nine genera of bacteria and 14 fecal metabolites in the adjusted GLMs to identify preschoolers with FSIQ in G1 and G2. The variable importance of the key gut microbiome and fecal metabolites is arranged in descending order, as shown in Fig. 7B. Notably, after adjusting for confounders, the combination analysis showed that Acinetobacter, isonicotinic acid and 3-methyl-4-hydroxybenzaldehyde had better identification ability (area under the curve, 0.718; 95% CI: 0.624, 0.812; Fig. 7C and D).

Fig. 7.

Comprehensive correlation analysis of gut microbiome and fecal metabolites. A, Heatmap of the Spearman correlation between 9 bacteria genera and 14 fecal differential metabolites (*FDR < 0.05), the red squares suggest a positive relationship, while the green squares signify a negative relationship. B, Variable importance of gut microbiome and fecal metabolites was identified from random forest classifiers, the variables are ranked in descending order based on their importance to the accuracy of the model.“MeanDecreaseAccuracy” and “MeanDecreaseGini” are represented by magenta and blue bars, respectively. C, Unicompounds bestmodel ROC, the red curve represents 3-methyl-4-hydroxybenzaldehyde, the green curve represents Acinetobacter, and the yellow curve represents Isonicotinic acid. D, Combined bestmodel ROC

Discussion

This study found significant differences in the observed species index, GMHI, and MDI between children in G1 and G2. The abundance of nine genera, including Blautia, Holdemanella, and Acinetobacter, was associated with various WPPSI-IV subscales (VCI, VSI, FRI, and PSI) and FSIQ in preschoolers. Furthermore, 14 fecal metabolites, such as creatine, isonicotinic acid, and 3-methyl-4-hydroxybenzaldehyde, were linked to WPPSI-IV scores. While these results suggest a potential role for Acinetobacter, isonicotinic acid, and 3-methyl-4-hydroxybenzaldehyde in distinguishing between G1 preschoolers, further research is required to confirm their utility as reliable biomarkers.

Emerging research suggests that gut microbiota may influence intelligence and neurodevelopment through various mechanisms. First, metabolites produced by gut microbes, such as SCFAs, are thought to regulate neurotransmitter synthesis and inflammation, potentially influencing cognitive functions and IQ [37]. Second, the gut microbiota indirectly affects central nervous system function by modulating gut permeability and the blood-brain barrier through the gut–brain axis (GBA), thereby influencing intellectual development [1]. Additionally, specific microbial metabolites, such as tryptophan derivatives, can cross the blood-brain barrier and participate in the production of neurotransmitters like serotonin, which plays a crucial role in emotional regulation and cognitive function, both closely linked to IQ [38]. Imbalances in the gut microbiota are also associated with neurodevelopmental disorders such as autism and attention deficit hyperactivity disorder, which can further impair cognitive function [39].

Previous research have suggested a connection between the gut microbiota and cognitive function in children [8, 10, 16, 17, 40, 41]. Similarly, the present our research revealed that the gut microbiota of preschoolers is predominantly dominated by Firmicutes and Bacteroidetes [18, 42]. Firmicutes, especially Blautia and Faecalibacterium, were related to cognition in many cases. Consistent with the results of the present study, the abundance of Blautia was inversely associated with Mental Developmental Index (MDI) and Psychomotor Developmental Index (PDI) among healthy 3-year-old children [10]. Furthermore, a higher relative abundance of Blautia was related to fewer internalizing difficulties in children aged 3 years [43], and depleted Blautia was associated with cognitive dysfunction in children with Down syndrome [44]. Lapidot et al. [16] observed a reduction in Faecalibacterium abundance corresponding to higher FSIQ scores among healthy school-aged children. Conversely, a positive association was found between the abundance of Faecalibacterium and better MDI and PDI scores in children aged 3 years [10]. Faecalibacterium and Blautia, known for producing SCFAs like butyrate and propionate, which are anti-inflammatory molecules [45], play a crucial role in preserving intestinal environmental balance. Butyrate may influence cognition by modulating gut hormone levels in addition to its role in regulating inflammation [46]. These findings suggest that Faecalibacterium and Blautia could be investigated as potential microbial markers for mental health, though further research is needed to validate their role across different populations and age groups. Studies have also reported that a higher abundance of Firmicutes, such as Holdemanella, was associated with improved fine motor skills in 18-month-old children [40], whereas this genus was less abundant in 3-year-old children with language impairment [47]. Dialister showed a positive correlation with FSIQ levels among healthy school-aged children [16]. Subdoligranulum may be linked to Kleefstra syndrome, which is characterized by intellectual disability and autistic behavior [48]. In comparison to the findings of this study, the mechanisms through which the gut microbiota affects cognition may vary depending on the specific psychopathology and chronological age. Furthermore, the decline in the abundance of Holdemanella, Dialister, and Subdoligranulum observed in children with FSIQ below the borderline could potentially result in reduced levels of microbiota-derived metabolite SCFAs, and prolonged deficiency in SCFAs may consequently become a risk factor for cognitive decline [49].

Prior studies have shown increased abundance of Methanobrevibacter and Collinsella in children with autism [50] or enhanced fine motor skills [40]. Acetate and lactate (intermediate products of SCFAs), are often converted into butyrate by symbiotic bacteria [51], and butyrate is widely acknowledged for its beneficial effects on the host [52]. The present study showed that Methanobrevibacter and Collinsella may positively influence cognitive function by producing acetic acid and propionic acid. In addition, Bacteroidete-Prevotella_9, which has been demonstrated positively related to externalizing behavior in chidren [53, 54], was inversely related to preschoolers’ VCI. These findings align with previous results, which indicated that co-abundance groups dominated by Bacteroidetes were associated with adverse neurodevelopmental outcomes in early childhood [55]. Additionally, the association between Prevotella and behavioral issues, as well as cognitive problems, may be potentially connected to chronic intestinal inflammation [56], suggesting a harmful effect on cognitive health. Notably, Acinetobacter, a genus from Pseudomonadota, was found to be inversely related to FRI. Pseudomonadota plays a role in facilitating the colonization of obligate anaerobes essential for healthy intestinal function, by consuming oxygen and reducing the redox potential in the intestinal environment [57]. Consequently, Acinetobacter may be crucial for maintaining a stable GBA.

The gut microbiota communicates through its metabolites along the GBA [18]. Certain gut microbiota can produce neuroactive metabolites, such as neurotransmitters, which can then enter the brain through the bloodstream [18]. In addition, the microbiota and tis metabolites can stimulate enteroendocrine cells to release hormones into the bloodstream, thereby regulating the immune system [58]. While the influence of SCFAs produced by the gut microbiota on cognitive function has been well-documented [12, 49], research on other fecal metabolites remains limited. In the present study, preschoolers with normal IQ development exhibited higher abundance of fecal metabolites, shedding light on the connection between cognitive function and the gut microbiota. The fecal metabolites significantly associated with the IQ of preschoolers predominantly comprised organic acid and its derivatives, benzene and substituted derivatives, amino acid and its metabolites, fatty acyls, ketones, and coenzyme and vitamins (Table S8). Moreover, the identified differential metabolites are primarily involved in metabolic pathways related to the biosynthesis and metabolism of amino acids and glucose, which are largely in line with the predicted outcomes from Tax4Fun.

Fecal differential metabolite 3-methyl-4-hydroxybenzaldehyde belongs to the class of benzene and substituted derivatives. 3-Methyl-4-hydroxybenzaldehyde could affect the normal function and development of the nervous system by mechanisms such as causing DNA damage, oxidative stress, neuroinflammation, apoptosis, disruption of neural transmission, and alteration of neurotransmitters [59, 60]. Isonicotinic acid, categorized as an organic acid and its derivatives, is known to participate in drug metabolism [61]. Previous research has demonstrated significant neurotoxicity associated with isonicotinic acid in zebrafish larvae in vivo [62]. This evidence supports the potential negative effect of isonicotinic acid on the IQ of preschoolers as observed in the present study. Creatine, zerumbone, and trans-cinnamaldehyde are mainly involved in metabolic pathways; glycine, serine, and threonine metabolism; arginine and proline metabolism; and the inflammatory mediator regulation of TRP channels (Table S9). While no research evidence specifically regarding preschoolers is available at present, studies involving adults or animals have shown that zerumbone can promote the proliferation of endogenous neural stem cells in vascular dementia through the modulation of Notch signaling [63]. Additionally, zerumbone has been shown to improve behavioral disorders and neuropathology in transgenic mice by inhibiting MAPK signaling [64]. Indeed, creatine has been confirmed to positively affect memory and cognitive function in healthy individuals [65, 66]. In animal experiments, trans-cinnamaldehydecan mitigate age-induced cognitive impairment in elderly rats by modulating mitochondrial function, inflammation, and apoptotic mediators in the prefrontal cortex [67]. Additionally, trans-cinnamaldehyde has been demonstrated to alleviate cognitive dysfunction by reducing oxidative stress [68, 69]. It also exhibits a reversal effect on behavioral and neurochemical deficits induced by diabetes [68, 69]. The significant associations observed between fecal metabolites and IQ in preschoolers suggest that changes in gut microbiota and its metabolites may potentially impact the endogenous metabolic function of the host. This study also revealed a consistent trend in the GMHI and MDI between G1 and G2. This suggests that while differences in microbial health exist between the two groups, these differences reflect a shared pattern of changes in microbial health and dysbiosis across populations. Such a trend may highlight the interconnectedness of the microbial ecosystem’s overall stability [35, 36], further supporting the potential link between gut microbiota health and intelligence. Additionally, correlation analysis revealed associations between various gut microbiota, including Faecalibacterium, Blautia, and Acinetobacter, and differential fecal metabolites in preschoolers. These findings may suggest an interactive relationship between the gut microbiota and its metabolites, and this interaction may affect the endogenous metabolic function of preschoolers, potentially contributing to nervous system function. The combination of Acinetobacter, 3-methyl-4-hydroxybenzaldehyde, and isonicotinic acid showed a potential trend towards better identification ability for distinguishing preschoolers with FSIQ scores below or above the borderline in the random forest analysis. This finding implies that gut microbes and their metabolites could be explored as potential markers for assessing the intelligence levels of preschoolers. Although none of the children in this study were clinically diagnosed with intellectual disability, the investigation into the relationship among gut microbes, fecal metabolites, and intelligence levels in preschoolers within an atypical developmental population implies that these cross-sectional associations may hold relevance in longitudinal studies or in children at a high risk of intellectual disability. Therefore, even with a moderate decline in the association between the WPPSI-IV scores and the gut microbiota and fecal metabolites, potential clinical and public health significance remains.

Some limitations of this study should be noted. First, the inclusion of a general population based in Baise, Guangxi suggests that the findings may not directly apply to clinical or high-risk populations for intellectual disability. Second, the cross-sectional study design implied that causal relationships between gut microbiota, fecal metabolites, and the IQ in preschoolers cannot be captured. Moreover, considering that the gut microbiota of children aged 3–6 years is considered to be stable [70], the assessments may have underestimated the true association. Longitudinal studies are necessary to explore whether a critical window exists for the gut microbiome and its metabolites in influencing the neurodevelopment of preschoolers. Third, the associations between gut microbiota and fecal metabolites may appear weak due to the complexity of the gut microbiome and the influence of environmental and individual factors, such as diet and lifestyle. The cross-sectional design also limits the ability to capture dynamic interactions over time. Future longitudinal studies with larger sample sizes are needed to better elucidate the strength and direction of these associations. Fourth, comprehensive data on dietary habits or exposure to pets or other animals were not gathered, which could act as confounding factors in the relationship among gut microbiota, fecal metabolites, and IQ. Lastly, IQ scores may not fully reflect the multidimensional nature of intelligence, such as creativity, emotional intelligence, and social skills. IQ tests may, to some extent, offer a narrow view of intelligence, overlooking broader cognitive development. Thus, while this study uses IQ scores as an indicator of cognitive development, future research should adopt more diverse assessments to better evaluate children’s cognitive abilities. Meanwhile, the strengths lie in the comprehensive analyses conducted, which included group comparisons, GLMs, and a random forest model. These approaches allowed for thorough investigation of the impact of the gut microbiota and fecal metabolites on preschoolers’ IQ. Furthermore, the potential correlation among gut microbiota, fecal metabolites, and IQ in preschoolers could be identified. This findings also suggest that the combination of gut microbiota and fecal metabolites may serve as potential markers for IQ levels in preschoolers.

Conclusion

This study revealed differences in gut microbiota composition and metabolite profiles among preschoolers with varying IQ levels, suggesting that the gut microbiota and its metabolites may play a role in influencing cognitive development. The combination of Acinetobacter, 3-methyl-4-hydroxybenzaldehyde, and isonicotinic acid shows potential as markers for distinguishing preschoolers with FSIQ scores below the borderline. However, these associations need to be confirmed in larger and more diverse cohorts before they can be considered reliable biomarkers.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

Jinghua Long performed the investigation, data curation, writing - original draft and review & editing the manuscript. Jiehua Chen, Huishen Huang and Jun Liang performed the investigation and writing - review & editing. Lixiang Pang, Kaiqi Yang and Junwang Gu organize the data. Huanni Wei and Qian Liao participated in the survey. Xiaoyun Zeng supervised the progress of the research. Dongping Huang performed the supervision and funding acquisition. Xiaoqiang Qiu performed the project administration, funding acquisition, and writing-review & editing. All authors read and approved the final manuscript.

Funding

This research was funded by a grant from the Guangxi Key Research Program (Grant number AB17195012) and the National Natural Science Foundation of China (Grant number 22366007).

Data availability

Sequence data that support the findings of this study have been deposited in the NCBI BioProject repository with the BioProject ID: PRJNA1124508.

Declarations

Informed consent statement

All the participants and, in the case of participants under the age of 16, their parents or legal guardians provided written informed consent prior to participation. This project was approved by the ethics committee of Guangxi Medical University (No. 20140305-001).

Competing interests

The authors declare no competing interests.