Research Article

Transcriptome-wide isoform-level dysregulation in ASD, schizophrenia, and bipolar disorder

See allHide authors and affiliations

Science  14 Dec 2018:
Vol. 362, Issue 6420, eaat8127
DOI: 10.1126/science.aat8127
  • The PsychENCODE cross-disorder transcriptomic resource.

    Human brain RNA-seq was integrated with genotypes across individuals with ASD, SCZ, BD, and controls, identifying pervasive dysregulation, including protein-coding, noncoding, splicing, and isoform-level changes. Systems-level and integrative genomic analyses prioritize previously unknown neurogenetic mechanisms and provide insight into the molecular neuropathology of these disorders.

  • Fig. 1 Gene and isoform expression dysregulation in brain samples from individuals with psychiatric disorders.

    (A) DE effect size (|log2FC|) histograms are shown for protein-coding, lncRNA, and pseudogene biotypes up- or down-regulated (FDR < 0.05) in disease. Isoform-level changes (DTE; blue) show larger effect sizes than at the gene level (DGE; red), particularly for protein-coding biotypes in ASD and SCZ. (B) A literature-based comparison shows that the number of DE genes detected is dependent on study sample size for each disorder. (C) Venn diagrams depict overlap among up- or down-regulated genes and isoforms across disorders. (D) Gene ontology enrichments are shown for differentially expressed genes or isoforms. The top five pathways are shown for each disorder. (E) Heatmap depicting cell type specificity of enrichment signals. Differentially expressed features show substantial enrichment for known CNS cell type markers, defined at the gene level from single-cell RNA-seq. (F) Annotation of 944 ncRNAs DE in at least one disorder. From left to right: Sequence-based characterization of ncRNAs for measures of human selective constraint; brain developmental expression trajectories are similar across each disorder (colored lines represent mean trajectory across disorders); tissue specificity; and CNS cell type expression patterns.

  • Fig. 2 Aberrant local splicing and isoform usage in ASD, SCZ, and BD.

    (A) Venn diagram showing cross-disorder overlap for 472 genes with significant differentially spliced (DS) intron clusters (FDR < 10%) identified by LeafCutter. P values for hypergeometric tests of pairwise overlaps between each disorder are shown at the bottom. (B) Scatter plots comparing PSI changes for all 1287 introns in 515 significant DS clusters in at least one disorder, for significant disease pairs SCZ versus ASD and SCZ versus BD (Spearman’s ⍴ = 0.52 and 0.59, respectively). Principal component regression lines are shown in red, with regression slopes for ASD and BD ΔPSI compared to SCZ in the top-left corner. (C) Top 10 gene ontology (GO) enrichments for DS genes in each disorder (see also fig. S8C). (D) Significant enrichment for neuronal and astrocyte markers (ASD and SCZ), as well as oligodendrocyte and microglia (SCZ) cell type markers in DS genes. The odds ratio (*OR) is given only for FDR < 5% and OR > 1. Oligo, oligodendrocytes; OPC, oligodendrocyte progenitor cells. (E) A significant DS intron cluster in GRIN1 (clu_35560; chr9:140,040,354-140,043,461) showing increased exon 4 (E4) skipping in both ASD and SCZ. Increased or decreased intron usage in ASD and SCZ cases compared to controls is highlighted in red and blue, respectively. Protein domains are annotated as ANF_receptor, extracellular receptor family ligand binding domain; Lig_chan, ionotropic glutamate receptor; Lig_chan-Glu_bd, ligated ion channel l-glutamate- and glycine-binding site; CaM_bdg_C0, calmodulin-binding domain C0 of NMDA receptor NR1 subunit. Visualization of splicing events in cluster clu_35560 with the change in PSI (ΔPSI) for ASD (left) and SCZ (right) group comparisons. FDR-corrected P values (q) are indicated for each comparison. Covariate-adjusted average PSI levels in ASD or SCZ (red) versus CTL (blue) are indicated at each intron. (F) Violin plots with the distribution of covariate-adjusted PSI per sample for the intron skipping E4 are shown for each disease group comparison. (G) DGE for GRIN1 in each disorder (*FDR < 5%). (H) Whole-gene view of NRXN1 highlighting (dashed lines) the intron cluster with significant DS in ASD (clu_28264; chr2:50,847,321-50,850,452), as well as transcripts NRXN1-004 and NRXN1-012 that show significant DTU in SCZ and/or BD. Protein domain mappings are shown in purple. DM, protein domains; Tx, transcripts; ConA-like_dom_sf, concanavalin A–like lectin/glucanase domain; EGF-like, epidermal growth factor-like domain; laminin_G, laminin G domain; neurexin-like, neurexin/syndecan/glycophorin C domain. (I) (Left) Close-up of exons and protein domains mapped onto the DS cluster and FDR-corrected P value (q). (Right) Visualization of introns in cluster clu_28264 with their change in percent spliced in (ΔPSI). Covariate-adjusted average PSI levels in ASD (red) versus CTL (blue) are indicated for each intron. (J) Violin plots with the distribution of covariate-adjusted PSI per sample for the largest intron skipping exon 8 (E8). (K) Bar plots for changes in gene expression and transcript usage for NRXN1-004 and NRXN1-012 (*FDR < 5%).

  • Fig. 3 Overlap and genetic enrichment among dysregulated transcriptomic features.

    (A) Scatterplots demonstrate overlap among dysregulated transcriptomic features, summarized by their first principal component across subjects (R2 values; *P < 0.05). PRS show greatest association with differential transcript signal in SCZ. (B) SNP heritability in SCZ is enriched among multiple differentially expressed transcriptomic features, with down-regulated isoforms showing the most substantial association via stratified LD-score regression. (C) Several individual genes and isoforms exhibit genome-wide significant associations with disease PRS. Plots are split by direction of association with increasing PRS. In ASD, most associations localize to the 17q21.31 locus, harboring a common inversion polymorphism.

  • Fig. 4 Transcriptome-wide association.

    Results from a TWAS prioritize genes whose cis-regulated expression in brain is associated with disease. Plots show conditionally-independent TWAS prioritized genes, with lighter shades depicting marginal associations. The sign of TWAS z-scores indicates predicted direction of effect. Genes significantly up- or down-regulated in diseased brain are shown with arrows, indicating directionality. (A) In SCZ, 193 genes (164 outside of MHC) are prioritized at Bonferroni-corrected P < 0.05, including 107 genes with conditionally independent signals. Of these, 23 are also differentially expressed in SCZ brains with 11 in the same direction as predicted. (B) Seventeen genes are prioritized in BD, of which 15 are conditionally independent. (C) In ASD, a TWAS prioritizes 12 genes, of which 5 are conditionally independent.

  • Fig. 5 Gene and isoform coexpression networks capture shared and disease-specific cellular processes and interactions.

    (A) Coexpression networks demonstrate pervasive dysregulation across psychiatric disorders. Hierarchical clustering shows that separate gene- and isoform-based networks are highly overlapping, with greater specificity conferred at the isoform level. Disease associations are shown for each module (linear regression β value, *FDR < 0.05, –P < 0.05). Module enrichments (*FDR < 0.05) are shown for major CNS cell types. Enrichments are shown for GWAS results from SCZ (59), BD (97), and ASD (38), using stratified LD score regression (*FDR < 0.05, –P < 0.05). (B) Coexpression modules capture specific cellular identities and biological pathways. Colored circles represent module DE effect size in disease, with red outlines representing GWAS enrichment in that disorder. Modules are organized and labeled based on CNS cell type and top gene ontology enrichments. (C) Examples of specific modules dysregulated across disorders, with the top 25 hub genes shown. Edges represent coexpression (Pearson correlation > 0.5) and known protein-protein interactions. Nodes are colored to represent disorders in which that gene is differentially expressed (*FDR < 0.05).

  • Fig. 6 Two RBFOX1 isoform modules capture distinct biological and disease associations.

    (A) Previous studies have identified RBFOX1 as a critical hub of neuronal and synaptic modules down-regulated across multiple psychiatric disorders (13, 15). We identified two pairs of modules with distinct RBFOX1 isoforms as hub genes. Plots show the top 25 hub genes of modules isoM2 and isoM17, following the same coloring scheme as in Fig. 5C. (B) Distinct module-eigengene trait associations are observed for isoM2 (down-regulated in ASD only) compared with isoM17, which is down-regulated in ASD and SCZ. (C) Modules show distinct enrichments for nuclear and cytoplasmic RBFOX1 targets, defined experimentally in mouse (32). (D) Genes harboring DS events observed in ASD and SCZ show greater overlap with isoM17, consistent with its association with nuclear RBFOX1 targets. (E) Modules show distinct patterns of genetic association. isoM2 exhibits broad enrichment for GWAS signal in SCZ, BD, and MDD, as well as for epilepsy risk genes, whereas isoM17 shows no apparent genetic enrichment (21).

  • Fig. 7 Distinct neural-immune trajectories in disease.

    (A) Coexpression networks refine the neural-immune/inflammatory processes up-regulated in ASD, SCZ, and BD. Previous work has identified specific contributions to this signal from astrocyte and microglial populations (13, 19). Here, we identify additional contributions from distinct IFN-response and NFkB signaling modules. (B) Eigengene-disease associations are shown for each of four identified neural-immune module pairs. The astrocyte and IFN-response modules are up-regulated in ASD and SCZ. NFkB signaling is elevated across all three disorders. The microglial module is up-regulated in ASD and down-regulated in SCZ and BD. (C) Top hub genes for each module are shown, along with edges supported by coexpression (light gray; Pearson correlation > 0.5) and known protein-protein interactions (dark lines). Nodes follow the same coloring scheme as in Fig. 5C. Hubs in the astrocyte module (geneM3/isoM1) include several canonical, specific astrocyte markers, including SOX9, GJA1, SPON1, and NOTCH2. Microglial module hub genes include canonical, specific microglial markers, including AIF1, CSF1R, TYROBP, and TMEM119. The NFkB module includes many known downstream transcription factor targets (JAK3, STAT3, JUNB, and FOS) and upstream activators (IL1R1, nine TNF receptor superfamily members) of this pathway. (D) The top four GO enrichments are shown for each module. (E) Module enrichment for known cell type–specific marker genes, collated from sequencing studies of neural-immune cell types (98102). (F) Module eigengene expression across age demonstrates distinct and dynamic neural-immune trajectories for each disorder.

  • Fig. 8 LncRNA annotation, ANK2 isoform switching, and microexon enrichment.

    (A) FISH images demonstrate interneuron expression for two poorly annotated lincRNAs—LINC00643 and LINC01166—in area 9 of adult human prefrontal cortex. Sections were labeled with GAD1 probe (green) to indicate GABAergic neurons and lncRNA (magenta) probes for LINC00643 (left) or for LINC01166 (right). All sections were counterstained with DAPI (blue) to reveal cell nuclei. Lipofuscin autofluorescence is visible in both the green and red channels and appears orange. Scale bar, 10 μm. FISH was repeated at least twice on independent samples (table S9) (21), with similar results (see also fig. S16). (B) ANK2 isoforms ANK2-006 and ANK2-013 show significant DTU in SCZ and ASD, respectively (*FDR < 0.05). (C) Exon structure of ANK2 highlighting (dashed lines) the ANK2-006 and ANK2-013 isoforms. (Inset) These isoforms have different protein domains and carry different microexons. ANK2-006 is affected by multiple ASD DNMs, while ANK2-013 could be entirely eliminated by a de novo CNV deletion in ASD. (D) Disease-specific coexpressed PPI network. Both ANK2-006 and ANK2-013 interact with NRCAM. The ASD-associated isoform ANK2-013 has two additional interacting partners, SCN4B and TAF9. (E) As a class, switch isoforms are significantly enriched for microexon(s). In contrast, exons of average length are not enriched among switch isoforms. The y axis displays odds ratio on a log2 scale. P values are calculated using logistic regression and corrected for multiple comparisons. (F) Enrichment of 64 genes with switch isoforms for: ASD risk loci (81); CHD8 targets (103); FMRP targets (33); mutationally constraint genes (104); syndromic and highly ranked (1 and 2) genes from SFARI Gene database; vulnerable ASD genes (105); genes with probability of loss-of-function intolerance (pLI) > 0.99 as reported by the Exome Aggregation Consortium (106); genes with likely-gene-disruption (LGD) or LGD plus missense de novo mutations (DNMs) found in patients with neurodevelopmental disorders (21).

Supplementary Materials

  • Transcriptome-wide isoform-level dysregulation in ASD, schizophrenia, and bipolar disorder

    Michael J. Gandal, Pan Zhang, Evi Hadjimichael, Rebecca L. Walker, Chao Chen, Shuang Liu, Hyejung Won, Harm van Bakel, Merina Varghese, Yongjun Wang, Annie W. Shieh, Jillian Haney, Sepideh Parhami, Judson Belmont, Minsoo Kim, Patricia Moran Losada, Zenab Khan, Justyna Mleczko, Yan Xia, Rujia Dai, Daifeng Wang, Yucheng T. Yang, Min Xu, Kenneth Fish, Patrick R. Hof, Jonathan Warrell, Dominic Fitzgerald, Kevin White, Andrew E. Jaffe, PsychENCODE Consortium, Mette A. Peters, Mark Gerstein, Chunyu Liu, Lilia M. Iakoucheva, Dalila Pinto, Daniel H. Geschwind

    Materials/Methods, Supplementary Text, Tables, Figures, and/or References

    Download Supplement
    • Materials and Methods
    • Figs. S1 to S17
    • Captions for Tables S1 to S9
    • PsychENCODE Consortium Authors and Affiliations
    • References
    Table S1
    Differential gene and isoform expression summary statistics and DE enrichment analyses
    Table S2
    Annotation of neuropsychiatric ncRNAs ('NPncRNAs')
    Table S3
    Differential splicing summary statistics, annotation and disease overlaps
    Table S4
    TWAS and SMR summary statistics, PRS associations with gene and isoform expression
    Table S5
    Gene and isoform co-expression module annotation
    Table S6
    csuWGCNA network annotation and putative lncRNA-mRNA targets
    Table S7
    Switch isoform and microexon characterization
    Table S8
    Splicing and isoform validation primers and samples
    Table S9
    RNAscope - Tissue samples and RNA FISH probes

Navigate This Article