Circular RNAs in the human brain are tailored to neuron identity and neuropsychiatric disease

Little is known about circular RNAs (circRNAs) in specific brain cells and human neuropsychiatric disease. Here, we systematically identify over 11,039 circRNAs expressed in vulnerable dopamine and pyramidal neurons laser-captured from 190 human brains and non-neuronal cells using ultra-deep, total RNA sequencing. 1526 and 3308 circRNAs are custom-tailored to the cell identity of dopamine and pyramidal neurons and enriched in synapse pathways. 29% of Parkinson’s and 12% of Alzheimer’s disease-associated genes produced validated circRNAs. circDNAJC6, which is transcribed from a juvenile-onset Parkinson’s gene, is already dysregulated during prodromal, onset stages of common Parkinson’s disease neuropathology. Globally, addiction-associated genes preferentially produce circRNAs in dopamine neurons, autism-associated genes in pyramidal neurons, and cancers in non-neuronal cells. This study shows that circular RNAs in the human brain are tailored to neuron identity and implicate circRNA-regulated synaptic specialization in neuropsychiatric diseases.

Supplementary Fig. 1 Supplementary Figure 1.RNA-seq sample filters and quality control, including outlier detection, sex concordance, and assay performance measures.a, Schematic of RNA-seq sample filtering.IDs of excluded samples are listed under each filtering step in gray text.Two QC tests were performed to identify outlier samples based on systematic abnormalities in overall expression (b,c).Moreover, we tested for sex concordance to identify potential sample mix-ups (d).b, Dendrogram visualizing pairwise Spearman correlations between gene expression levels of temporal cortex neuronal samples that are newly added after our previous study (Dong et al. Nature Neuroscience, 2018).c, Histogram of median pairwise k-mer distances for each of the 221 samples with all other samples.d, Concordance between clinical sex and sex-specific gene expression in neuronal and non-neuronal samples: normalized expression levels of the female-specific XIST transcript (x axis) and normalized expression levels of the Y-chromosome specific RPS4Y1 transcript (y axis) are shown.e, Scatterplot of two technical replicates based on lcRNAseq.n = 57,814; all annotated genes in GENCODE v19.f, Scatterplot of circRNA expression from two technical replicates (panel e) based on lcRNAseq.(intron-derived) enriched by RNase R treatment.It showed that RNase R treatment can significantly enrich exon-derived circRNAs, but not intron-derived ciRNAs.c, Size distribution of circular RNAs, color-coded by ciRNAs (orange) and circRNAs (red), before and after filters.It showed that ciRNAs are relatively shorter than circRNAs and after above enrichment filtering, ciRNAs are mostly filtered out.d, circRNAs expressed in the PICALM locus in different cell types.e, Genes with more exons are generally more likely to produce more circRNAs.Exons being circularized are significantly longer than average.Flanking introns of circRNAs have significantly less overlapping RBP binding activity peaks (using ENCODE eCLIP peaks 26 ) than control introns.f, Scatterplot of circular reads vs. linear reads per junction sites (left) and per host gene (right).g, DisGeNet disease class enriched in the host genes of all detected circRNAs in this study.h, Percentage of synaptic genes among all circRNA host genes are significantly decreased in PD (t-test, P = 0.036) and in AD (t-test, P = 0.012) comparing their corresponding controls.

Supplementary Fig. 3
Supplementary Figure 3. qPCR validation of circRNAs.a, Barplot of qPCR delta Ct values (comparing to UBC control) for 17 select circRNAs, grouped into substantia nigro dominant circRNAs (in red frame), temporal cortex dominant circRNAs (blue frame), brain dominant circRNAs (green), and non-brain dominant circRNA (brown).The tissue dominance is based on visual inspection.Orange bars are for RNase R treated samples while darkblue bars are for untreated samples.Note that the qPCR validation were performed in bulk tissue (e.g.SN: susbtantia nigra and TC: temporal cortex), instead of laser-captured neurons, due to the requirement of input RNAs.b, Schema of qPCR design for circRNA validation using circPSEN1-2 as an example.Paired primers targeted the back-splicing junction sites were listed in Supplementary Data 4. c, Cell-specificity score distribution of circRNAs and linear mRNAs from the same genetic loci in three major cell types.This is limited to those host linear mRNAs which only produce one circular RNA. ) overlapping the meta-exon and flanking intron (+/-200 bp) window of ubiquitous circRNAs and cell-type specific circRNAs from our study.Exonic region of a circRNA were concatenated and scaled into a meta-exon, in order to show the relative position of eCLIP peaks.The boxplot inset shows the eCLIP peak density (per basepair) on circular exon and its flanking intron.b, Number of eCLIP peaks overlapping the two putative types of cell type-specific circRNAs.c, Overlapped eCLIP binding peaks for 41 RBP were significantly positively (NIP7) or negatively enriched (all other RBPs) in circular exons of cell typespecific circRNAs for one cell type (e.g.circRNAs from locus producing one or more cell type-specific circRNAs specific to the same one cell type) compared to ubiquitous circRNAs with FDR < 0.05.Odds ratios from Fisher's exact test; n = 5,608 for circRNAs one cell type compared to n = 1,942 ubiquitous circRNAs.d.Overlapped eCLIP binding peaks for six RBPs were significantly less frequent at circular exons of cell type-specific circRNAs for multiple cell types (e.g., circRNAs from locus producing multiple cell type-specific circRNA for multiple cell types) compared to ubiquitous circRNAs.Supplementary Fig. 6 Supplementary Figure 6.Exploring associations between circRNA expression and neuropathology.

a
All samples except outliers, controls etc.(

Supplementary
Figure 2. Circular RNA charateristics.a, Scatterplot of circRNA reads count between paired RNase R-treated vs. mock-treated RNA-seq samples.Blue dots are those enriched in RNase-R treatment based on (1) ratio between RNase treated vs. mock reads is greater than or equal to 2, and (2) at least 20 reads in the RNase treated RNA-seq.b, Number of circRNAs (exon-derived) and ciRNAs Supplementary Fig. 4. RNA Binding Protein (RBP) activity and cell type-specific circRNAs.a, Number of eCLIP peaks (ENCODE eCLIP data from HepG2 and K562 cell lines 26

be
Number of eCLIP peaksCell type-specific circRNAs for multiple cell types d Cell type-specific circRNAs for one cell type c Odds ratio (log scale) Odds ratio (log scale) Odds ratio (log scale) u it o u s c ir c R N A s C e ll ty p e -s p e c if ic c ir c R N A s fo r m u lt ip le c e ll ty p e s C e ll ty p e -s p e c if ic c ir c R N A s fo r o n e c e ll ty p e

Supplementary Fig. 5 circular
028 circRNAs for cell type-specific circRNAs for multiple cell types compared to n = 1,942 ubiquitous circRNAs.e. Overlapped eCLIP binding peaks for 54 RBP were significantly positively (n = 10) or negatively enriched (n = 44) in circular exons of cell type-specific circRNAs for one cell type compared to cell type-specific circRNAs for multiple cell types with FDR < 0.05.Odds ratios from Fisher's exact test; n = 5,608 for circRNAs for one cell type compared to n = 6,028 circRNAs for multiple cell types.RNAs expressed in non-neuronal cells (n = 1) circular RNAs expressed in pyramidal neurons (n = 6) circular RNAs expressed in dopamine neurons (n = 3) circular RNAs expressed in non-neuronal cells (n = 4) circular RNAs expressed in pyramidal neurons (n = 3) circular RNAs expressed in dopamine neurons (n = 3) circular RNAs expressed in non-neuronal cells (n = 0) circular RNAs expressed in pyramidal neurons (n = 10) circular RNAs expressed in dopamine neurons (n = 6) circular RNAs expressed in non-neuronal cells (n = 0) circular RNAs expressed in pyramidal neurons (n = 14) circular RNAs expressed in dopamine neurons (n = 10) circular RNAs expressed in non-neuronal cells (n = 3) circular RNAs expressed in pyramidal neurons (n = 11) circular RNAs expressed in dopamine neurons (n = 10) circular RNAs expressed in non-neuronal cells (n = 6) circular RNAs expressed in pyramidal neurons (n = 5) circular RNAs expressed in dopamine neurons (n = 4) seq combined tracks for the major cell groups BRAINcode RNA-seq combined tracks for the major cell groupsBRAINcode RNA-seq combined tracks for the major cell groups BRAINcode RNA-seq combined tracks for the major cell groups BRAINcode RNA-seq combined tracks for the major cell groups 2plots for example circRNAs and their host loci, including DNAJC6, VPS13C, RIMS1, RIMS2, PSEN1, and APP.Note that the cell-specific circRNAs are based on healthy samples only.
a, Dopamine neuron circRNAs associated with Lewy body stages.The abundance of 26 circRNAs was associated with Lewy body stage with nominal P values ≤ 0.05 using linear regression analysis adjusted for covariates of sex, age, RIN, and PMI; 95 (out of 104) dopamine neuron transcriptomes with available neuropathology staging information (e.g., the Unified Lewy Body Staging System scores of 0, I, IIa, III, and IV, see Supplementary Data S1).None achieved the multiple-testing-corrected significance threshold of FDR ≤ 0.05.The line graph shows mean and standard error of the normalized expression counts of positively and negatively associated circRNAs.
b, Effect sizes of circRNAs associated with Lewy body neuropathology were highly correlated with the effect sizes for association with clinical diagnosis for the 18 samples with a clinical diagnosis of PD compared to 59 healthy controls without Lewy body neuropathology (Pearson's r = 0.75, P ≤ 2.22×10 -16 ).Red dots, 32 suggestive circRNAs associated with Lewy body neuropathology or PD clinical diagnosis (e.g., P < 0.05 in either comparison); grey dots, circRNAs that were not associated with Lewy body neuropathology or PD clinical diagnosis or (e.g., P ≥ 0.05 in both comparisons).c, Pyramidal neuron circRNAs associated with AD neuropathology.The abundance of 51 circRNAs was associated with the neuropathological Braak stages of AD patients with nominal P values ≤ 0.05 using linear regression analysis adjusted for covariates of sex, age, RIN • • d, Effect sizes of circRNAs associated with neuropathological AD Braak stages were highly correlated with the effect sizes for association with clinical diagnosis of AD compared to healthy controls (Pearson's r = 0.91; P ≤ 2.22 × 10 -16 ; n = 43 AD and 40 control samples; Fig.1b).Red dots, 71 suggestive circRNAs associated with AD clinical diagnosis or AD neuropathology (e.g., P < 0.05 in either comparison); grey dots, circRNAs that are not associated with AD clinical diagnosis or AD neuropathology (e.g., P ≥ 0.05 in both comparisons).