Developmental dynamics of RNA translation in the human brain

Abstract

The precise regulation of gene expression is fundamental to neurodevelopment, plasticity and cognitive function. Although several studies have profiled transcription in the developing human brain, there is a gap in understanding of accompanying translational regulation. In this study, we performed ribosome profiling on 73 human prenatal and adult cortex samples. We characterized the translational regulation of annotated open reading frames (ORFs) and identified thousands of previously unknown translation events, including small ORFs that give rise to human-specific and/or brain-specific microproteins, many of which we independently verified using proteomics. Ribosome profiling in stem-cell-derived human neuronal cultures corroborated these findings and revealed that several neuronal activity-induced non-coding RNAs encode previously undescribed microproteins. Physicochemical analysis of brain microproteins identified a class of proteins that contain arginine-glycine-glycine (RGG) repeats and, thus, may be regulators of RNA metabolism. This resource expands the known translational landscape of the human brain and illuminates previously unknown brain-specific protein products.

Figure 1:. Ribosome profiling captures active translation in the human adult and prenatal brain.

(a) Overview of experimental design. (b) Histogram depiction of patient samples included in this study. (c) Bar plot displaying P-sites derived from offset-corrected Ribo-seq reads in the first 100 nt of annotated ORFs (left) and the percentage of footprints in each reading frame (right). Data are shown as mean ± SD, n=73 biologically independent tissues. (d) Schematic overview of ORF types detected by RibORF. (e) Number of ORFs of each type identified in human adult and/or prenatal brain. (f) Stacked bar plot of start codon usage by ORF type. (g) Stacked bar plot of numbers and percentages of translated non-coding RNAs separated by transcript biotype. dORF, downstream open reading frame; miscRNA, miscellaneous RNA; o-uORF, overlapping upstream open reading frame; pcw, post-conception weeks; y, years.

Figure 2:. Transcriptional and translational regulation across human brain development.

(a) Classification of genes based on RNA-seq, Ribo-seq, and ribosome density measurements. (b) Scatterplot of fold-changes between adult and prenatal brain for all canonical ORFs in Ribo-seq data and the corresponding gene in RNA-seq data. Positive values indicate enrichment in the adult brain, whereas negative values indicate enrichment in the prenatal brain. Transcriptionally regulated genes (blue; change in transcription with no change in ribosome density), translationally regulated genes (red; change in ribosome density with no change in transcription), buffered genes (light purple; change in ribosome density that counterbalances the change in mRNA transcription), and intensified genes (dark purple; change in ribosome density that amplifies the change in mRNA) are highlighted. (c) Heatmap of genes associated with the top GO term in each regulatory category identified in A. Black outlines indicate DESeq2 p_adj < 0.05, gene names in red indicate inclusion in a given regulatory category.

Figure 3:. Microprotein expression and validation across brain development.

(a) Number of sORFs of each type identified in human adult and/or prenatal brain. (b) Stacked bar plot of numbers and percentages of translated non-coding RNAs containing at least one sORF, separated by transcript biotype. (c) Scatterplot of fold-changes between adult and prenatal brain for all sORFs in Ribo-seq data and the corresponding gene in RNA-seq data. Positive values indicate enrichment in the adult brain, whereas negative values indicate enrichment in the prenatal brain. Genes regulated by transcription (blue), translation (red), buffered (light purple), and intensified mechanisms (dark purple) are highlighted. (d) Number and type of ORFs identified by size-selection proteomics in the adult and prenatal brain, or by Johnson et al. (e) Genomic locus of GLUD1. Tracks represent merged and depth-normalized reads across all adult vs. prenatal samples for RNA-seq, Ribo-seq, as well as P-site positions. The sORF identified by RibORF is shown in gold, and the TISU sequence is indicated with an arrow. dORF, downstream open reading frame; miscRNA, miscellaneous RNA; o-uORF, overlapping upstream open reading frame; QC, quality control.

Figure 4:. Activity-dependent translation in hESC-derived neurons.

(a) Schematic of Ribo-seq and RNA-seq from NGN2-derived hESCs following 6 h membrane depolarization. (b) Breakdown of translated ORFs of each type identified in NGN2-derived neurons. (c) Stacked bar plot of numbers and percentages of translated non-coding RNAs separated by transcript biotype. (d) PCA analysis based on RNA-seq and Ribo-seq reads mapping to annotated genes in primary adult and prenatal brain tissue and NGN2 neurons. (e) Volcano plot of -log₁₀(p_adj) versus log₂(fold-change) in RNA-seq expression between membrane-depolarized and unstimulated NGN2 neurons. Black indicates DEseq2 p_adj < 0.05, purple indicates activity-dependent non-coding RNAs with no evidence of translation in human brain or NGN2 neurons, red indicates activity-dependent non-coding RNAs with evidence of translation in human brain and/or NGN2 neurons. (f) Genomic locus of LINC00473 in NGN2 neurons. Tracks represent merged and depth-normalized reads across 3 biological replicates of membrane-depolarized (6 h KCl) and unstimulated neurons for RNA-seq, Ribo-seq, as well as P-site positions for Ribo-seq and harringtonine-treated Ribo-seq. The sORF identified by RibORF is shown in gold. (g) High resolution depiction of genomic locus of the ORF encoded by LINC00473 in NGN2 neurons. Tracks represent merged and depth-normalized reads across 3 biological replicates of membrane-depolarized (6 h KCl) and unstimulated neurons for RNA-seq, Ribo-seq, as well as P-site positions for Ribo-seq and harringtonine-treated Ribo-seq. The sORF identified by RibORF is shown in gold. (h) Bar plot displaying P-sites derived from offset-corrected Ribo-seq reads from NGN2 neurons treated with vehicle control. The first 50 nt (left) and last 50 nt (right) of annotated ORFs are shown. Data are shown as mean ± SD, n=6 independent cell differentiations examined over two independent experiments. (i) Bar plot displaying P-sites derived from offset-corrected Ribo-seq reads from NGN2 neurons treated with harringtonine. The first 50 nt (left) and last 50 nt (right) of annotated ORFs are shown. Data are shown as mean ± SD, n=3 independent cell differentiations. (j) Number of ORFs of each type identified in NGN2 neurons treated with harringtonine. Absolute number (n) and percentage of overlap with ORFs identified from NGN2 neurons treated with cycloheximide alone are noted in parentheses. dORF, downstream open reading frame; h, hours; miscRNA, miscellaneous RNA; o-uORF, overlapping upstream open reading frame; PC, principal component.

Figure 5:. Evolutionary origins of human brain sORFs.

(a) Percentage of canonical ORFs (top, all ORFs in human Ensembl database, ≥40 AA) and sORFs (bottom, ≥40 AA) grouped by evolutionary age. (b) Box and whisker plots of microprotein ORF length grouped by evolutionary age. (c) Box and whisker plots of the number of splice junctions per microprotein ORF (40-100 AA) grouped by evolutionary age. (d) Box and whisker plots of microprotein ORF ribosome density grouped by evolutionary age. (b-d) Data are shown as median ± IQR (whiskers = 1.5*IQR), notches indicate median +/− 1.58*IQR/sqrt(n). N = 2,488 (ancient), 1,396 (chordate), 1,859 (mammal), 604 (primate), 12,689 (human), 19,036 (all) sORFs. (e) Pie chart of the percentage of ORFs with a TE insertion at the start codon, grouped by ORF type or non-coding RNA biotype. (f) Pie chart of the distribution of TE types, grouped by ORF type or non-coding RNA biotype. Numbers indicate the number of ORFs in each category. IQR, interquartile range; ncRNA, non-coding RNA.

Figure 6:. Effects of uORF expression on downstream ORF translation.

(a) Beeswarm dot plot showing the Spearman’s r correlation between uORF translation (uORF normalized counts from ribosome profiling) and canonical ORF ribosome density (ccds normalized counts from ribosome profiling divided by normalized total RNA abundance) for individual genes across all 73 individuals. Red line represents the mean correlation across all genes. Red dots indicate developmentally regulated uORFs (described in Supplementary Figure 6a). (b-c) Scatterplot and Spearman’s r correlation between upstream ORF translation (uORF normalized counts from ribosome profiling) and canonical ORF ribosome density (ccds normalized counts from ribosome profiling divided by normalized total RNA abundance) for MAP2K1 (b) and PIK3C2B (c) across 73 individuals. Gray shading = 95% CI. (d) Box and whisker plot of RNA-seq reads (transcripts per million reads, TPM) from adult and prenatal samples over DLGAP1 exons 1-3 (p = 7.47*10⁻¹⁰), all exons except 1-3 (p = 4.20*10⁻⁴), and all exons (not significant). (e) Box and whisker plot of Ribo-seq P-sites (in TPM) from adult and prenatal samples over DLGAP1 uORF (p = 6.15*10⁻⁵) and ccds ORF (2.87*10⁻⁸). (d-e) **** p < 0.0001, *** p < 0.001 by two-sided Kolmogorov–Smirnov test. Data are shown as median ± IQR (whiskers = 1.5*IQR), notches indicate median +/− 1.58*IQR/sqrt(n), n = 43 (adult) and 30 (prenatal) biologically independent tissues. (f) Scatterplot and Spearman’s r correlation between upstream ORF translation (uORF normalized counts from ribosome profiling) and canonical ORF ribosome density (ccds normalized counts from ribosome profiling divided by normalized total RNA abundance) for DLGAP1 across 73 individuals. Gray shading = 95% CI. IQR, interquartile range; NS, not significant.

Figure 7:. Microprotein functional characterization.

(a) FoldIndex score distribution of proteins annotated in Uniprot (black), annotated proteins with intrinsically disordered regions (gray), and sORFs with and without a BlastP hit (red and blue, respectively). (b) Scatterplot of average enrichment per residue of sequence and physicochemical properties in sORFs with no BlastP homology versus annotated proteins (Uniprot). RGG repeats were the most highly enriched of the tested sequence and physicochemical properties in sORFs. (c) Heatmap and hierarchical clustering of z-scores for physicochemical parameters associated with the known disordered proteome (IDRs 21-100 AA in length) as well as sORFs with predicted IDRs that do not have a paralog and do not overlap annotated coding ORFs. For the purposes of this analysis, sORFs that overlapped with an annotated ccds ORF (e.g. internal, external, readthrough), were excluded from this analysis. Boxes to the right of the heatmap indicate clusters of IDRs with similar properties. Blue = clusters depleted for sORFs, yellow = clusters significantly enriched for sORFs. (d) Western blot of FLAG-HA-tagged unmodified and ATT-mutated LINC00473 and LOC606724 lincRNAs, which includes the endogenous 5’UTR of each transcript. Experiment was repeated twice with similar results. Unprocessed blots are provided in the source data. (e) Immunofluorescence of FLAG-HA-tagged sORFs (SLN, LNC-FANCM-8, LNC-KHDRBS2-14, LINC00473, NBEAL1, and LOC606724) containing the endogenous 5’UTR expressed in HEK293T cells. Experiment was repeated twice with similar results. WT, wild-type.