With great interest, we read the article by Lou et al,1 who reported deviated gut microbiota in children with autism spectrum disorders (ASD). Recently, increasing studies have revealed abnormalities in extraneural tissues, particularly the gastrointestinal (GI) tract.2 Besides, the important role of lipids in the gut-brain axis of ASD is highlighted by their involvement in neurodevelopmental processes disrupted in ASD3 and their susceptibility to remodelling through intestinal intervention.4

In this study, we explored disrupted lipid profiles using lipidomics combined with single-cell transcriptomics, faecal metagenomics and cytokine profiling in multiple tissues of Chd8-mutant ASD mice (HET), examining how skewed lipidomes were related to ASD-relevant phenotypes. We analysed the lipidome of five tissues—brain, gut, liver, serum and GI contents in HET and wild type (WT) mice, and performed inflammatory cytokine assays and three behavioural tests in the same mice (figure 1a). We identified 726 lipid species in five tissues (online supplemental figure 1a), categorising them into five groups. Analysis of the relative abundance of these lipid species revealed significantly shifted lipidomes and differentially abundant lipid species in all five tissues of HET compared with WT (figure 1b,c). We further calculated the numbers of lipid species with increased or decreased abundance in total (figure 1d) and each lipid category (figure 1e), and observed that glycerophospholipids showed the highest number of changes among lipid categories. Besides, we observed altered levels of gut bacteria-specific lipid species in the GI contents, with increased levels of hydroxy and oxo fatty acids produced by Lactobacillus 5 and glycerophospholipids incorporating these fatty acids in the GI contents of HET. In contrast, PE-ceramides, produced by several bacteria, including Bacteroides,6 exhibited reduced levels (figure 1f and online supplemental table 1).

Disrupted lipid profiles in multiple tissues of Chd8-mutant autism spectrum disorders mice. (a) Schematic diagram of the study design. The mice used for lipidomic and biochemical analysis were the same ones on which behavioural tests were performed. The mice used for metagenomics and single-cell transcriptomics were different from each other and from those used in behavioural tests. For lipidomic analysis (b–e), the sample sizes were 13 and 10 mice for WT and HET, respectively. For single-cell transcriptomic analysis (f and g), n=4 mice per group. For faecal metagenomic analysis (i), n=12 and 11 mice for WT and HET, respectively. (b) Principal components analysis plots based on the relative abundance of lipid species (permutational multivariate analysis of variance). (c) Volcano plots of lipid species, with differences between WT and HET defined by adjusted p values <0.05. (d and e) Pie plots showing the number of lipid species in total (d) and with each lipid category (e), with increased or decreased abundance in HET tissues. χ2 tests were used to determine the differences between the numbers of increased and decreased lipid species. (f) The relative abundance of bacteria-specific lipid species in the GI contents: hydroxy and oxo fatty acids (the first and second panels), glycerophospholipids incorporating hydroxy and oxo fatty acids (the third to sixth panels) and PE-ceramides (the seventh panel). Significance indicated by adjusted p values. (g and h) Left panel: UMAP plot visualising the scores of lipid metabolism-related genes from single-cell transcriptomic profiles in the brain (g) and the gut (h). Right panel: violin plots displaying the expression of lipid metabolism-related genes in selected cell clusters in the brain (g) and in the gut (h). Significance indicated by p values (two-sided Wilcoxon rank-sum tests). (i) Box plots showing the relative abundance of bacteria from metagenomic data of faecal samples. Significance indicated by p values (two-sided Wilcoxon rank-sum tests). (j) Box plots showing the gene expression of KOs related to lipid metabolism within GO pathways from metagenomic data of faecal samples. K00059 and K16363 are involved in fatty acid biosynthesis, and K03621, K05879 and K19005 are involved in glycerolipid metabolism. Significance indicated by p values (two-sided Wilcoxon rank-sum tests). ASC, astrocytes; Ependy, ependymocytes; GI, gastrointestinal; MG, microglia; OLG, oligodendrocytes; Schwann, Schwann cells; WGCNA, Weight Gene Co-Expression Network Analysis.

Next, we investigated how the aberrant expression of lipid metabolism-related genes influenced the skewed lipidome in the brain and the gut at the single-cell transcriptomic level. We observed that in the cell clusters with high expression of lipid metabolism-related genes, the expression of these genes decreased in the brain and increased in the gut of HET mice (figure 1g,h, online supplemental figure 1b–j and online supplemental table 2), which may partially explain the altered lipidome of HET. Faecal metagenomic analysis reveals that the abundance of Lactobacillus johnsonii and L. plantarum increased and Bacteroides decreased in HET (figure 1i), corresponding with the altered levels of bacteria-specific lipids (figure 1f). Additionally, functional analysis of faecal metagenomic data showed that KOs involved in fatty acid biosynthesis decreased in HET, whereas those involved in glycerolipid metabolism elevated (figure 1j and online supplemental table 3).

We then examined how the disturbed lipid profiles were associated with ASD-relevant phenotypes and found that the numbers of lipid species significantly correlated with inflammatory, behavioural and anatomical phenotypes were highest in GI contents, brain and gut, respectively (figure 2a). The lipid species correlated with inflammatory or anatomical phenotypes shared more similarities, whereas those correlated with behavioural phenotypes shared few (figure 2b). This observation suggests similar lipid profiles modulate intestinal inflammation, but diverse profiles affect ASD-like behaviours. To explore the phenotype-associated lipid species, we grouped them into several modules using the Weight Gene Co-Expression Network Analysis (WGCNA) method based on lipid profile similarity.7 We focused on modules significantly correlated with various phenotypes (figure 2c and online supplemental figure 2a), black triangles), and observed that certain lipid species in selected modules were highly correlated with other lipid species (figure 2d and online supplemental figure 2b), indicating the core role of these lipids in modulating the relevant phenotypes.

Associations between lipid species and autism spectrum disorders (ASD)-relevant phenotypes in Chd8-mutant ASD mice. (a) Numbers of lipid species significantly associated with ASD-relevant phenotypes in five tissues of HET. Social refers to social deficits; Learning refers to learning and memory impairment; Brain (%) refers to the ratio of brain weight to body weight and B.W. refers to body weight. Significant correlations were defined by a p value of Spearman correlation less than 0.05. (b) Venn plots showing the intersection of lipid species in gastrointestinal (GI) contents of HET correlated with different inflammatory phenotypes (left panel), in the brain correlated with different behavioural phenotypes (middle panel) and in the gut correlated with different anatomical phenotypes (right panel). (c) Heatmaps showing the correlation of modules (X axis) with genotype and phenotypes (Y axis) analysed by Weight Gene Co-Expression Network Analysis. The colour represents correlation coefficients between modules and phenotypes/genotype. Modules selected for further investigation are marked by black triangles. *p<0.05; **p<0.01; ***p<0.001; ****p<0.0001. (d) Covariation of lipid species and lipid species distribution at the category level in the selected module. The line indicates the covariation of lipid species, and the dot size indicates the number of covariated lipid species. (e) Schematic diagram of the ketogenic diet administration. WT+CD: WT mice on the chow diet. HET+CD: HET mice on the chow diet. HET+KD: HET mice on the ketogenic diet. For behavioural tests, n=18, 15 and 10 mice for WT+CD, HET+CD and HET+KD, respectively. For lipidomics and cytokine profiling, n=13, 10 and 10 mice for WT+CD, HET+CD and HET+KD, respectively. (f) Percentages of time spent in the light box in the light/dark box test. (g) ELISA of IL-17A and IL-6 in the small intestine. (h) Principal component analysis plots based on the relative abundance of lipid species (permutational multivariate analysis of variance). (i and j) Upper panels: Venn plots displaying the intersection between lipid species recovered by the ketogenic diets and lipid species in selected module in the serum (i) and the gut (j). Lower panels: the relative abundance of shared lipid species shown in the Venn plots in the serum (i) and the gut (j). Statistical analysis was determined by one-way analysis of variance with two-tailed Tukey’s multiple comparison test (f and g). Significance was indicated by the p value (f and g).

We further investigated the phenotype-relevant lipid species by simultaneously influencing the lipidome and phenotypes via the ketogenic diet (KD).8 9 We treated a 8-week-old mice with KD for 4 weeks and measured the lipidome, intestinal inflammation and ASD-related behaviours (figure 2e). We found that KD improved anxiety and reduced intestinal inflammation in HET mice (figure 2f,g and online supplemental figure 2c–e), and changed the lipid composition (figure 2h and online supplemental figure 2f). Next, we focused on the relationship between KD-altered lipid species and the ASD-relevant phenotypes. We found that KD-restored lipid species overlapped with the lipid species in the selected WGCNA modules, especially the core lipid species with high covariation (figure 2i,j).

This study explored the lipidomic changes in different tissues of ASD from the multi-omics perspective, which provides a strong theoretical basis and practical approach for targeting the lipidome to alleviate ASD symptoms. Moreover, the lipid species in the serum, which showed a high correlation with ASD-related phenotypes and were prone to modification by external interventions, hold immense potential as complementary diagnostic markers in addition to gut microbes,1 marking a significant step forward in ASD research and diagnosis.

Not applicable.

All animal studies were performed in accordance with guidelines approved by the Animal Ethics Committee at the Institute of Zoology, Chinese Academy of Sciences.