LINE-1 retrotransposons drive human neuronal transcriptome complexity and functional diversification

Abstract

The genetic mechanisms underlying the expansion in size and complexity of the human brain remain poorly understood. Long interspersed nuclear element-1 (L1) retrotransposons are a source of divergent genetic information in hominoid genomes, but their importance in physiological functions and their contribution to human brain evolution are largely unknown. Using multiomics profiling, we here demonstrate that L1 promoters are dynamically active in the developing and the adult human brain. L1s generate hundreds of developmentally regulated and cell type-specific transcripts, many that are co-opted as chimeric transcripts or regulatory RNAs. One L1-derived long noncoding RNA, LINC01876, is a human-specific transcript expressed exclusively during brain development. CRISPR interference silencing of LINC01876 results in reduced size of cerebral organoids and premature differentiation of neural progenitors, implicating L1s in human-specific developmental processes. In summary, our results demonstrate that L1-derived transcripts provide a previously undescribed layer of primate- and human-specific transcriptome complexity that contributes to the functional diversification of the human brain.

Fig. 1.. L1-derived transcripts are abundant in the adult human brain.

(A) Schematic illustrating sample collection, sequencing strategy, and bioinformatics approach. (B) Left: Phylogenetic tree showing the evolutionary age of young L1 subfamilies. Right: Structure of a L1 element with a zoom-in to its 5′UTR. Arrows indicate promoters in sense (blue) and antisense (red). YY1 binding sites indicated in purple boxes (sense on top and antisense on bottom). (C) Expression of primate-specific L1 subfamilies compared to ancient L1 subfamilies and selected housekeeping (HK) genes as reference. Row annotation showing average length (AL), average percentage of divergence from consensus (AD), and the total number of elements (TNE) (information extracted from RepeatMasker open-4.0.5). (D) Expression [reads per kilobase per million mapped reads (RPKM)] over full-length (>6 kbp) L1HS, L1PA2, L1PA3, and L1PA4 plus 6-kbp flanking regions. (E) Percentage of expressed full-length (>6 kbp) elements (mean normalized counts, >10; see Methods) among young L1 subfamilies (n = number of expressed elements; T = total number of full-length elements). (F) Read counts in sense (light teal) and antisense (dark teal) per sample. First four showing full-length elements in young L1 subfamilies and last four showing ancient L1 subfamilies with a comparable number of copies. (G) PacBio Iso-Seq schematic and mapping approach. (H) Coverage of PacBio Iso-Seq library mapped to L1HS and L1PA2 consensus sequence. (I) Genome browser tracks showing PacBio Iso-Seq reads over the promoter region of a full-length L1HS.

Fig. 2.. L1 expression in neurons in the adult human brain.

(A) Schematic of sample collection, sequencing approach, and analytical bioinformatics pipeline for TE expression in single-nucleus data. (B) snRNA-seq: Uniform Manifold Approximation and Projection (UMAP) colored by defined clusters. (C) Expression of selected markers for different cell types. (D) UMAP colored by characterized cell types. (E) Pseudo-bulk cluster expression of young L1 subfamilies on UMAP. OPC, oligodendrocyte precursor cell. (F) Comparison of glia versus neuronal clusters per L1 family (each data point corresponds to a particular cluster expression value in a sample) (P value as per Wilcoxon test). (G) Schematic of NeuN⁺ H3K4me3 CUT&RUN in adult human brain samples and bioinformatics approach. (H) H3K4me3 peaks (left heatmap) over full-length L1 subfamilies (L1HS to L1PA4) and their RNA-seq signal (right heatmap). Profile plots showing sum of signal. CTX, cortex. (I) Genome browser tracks showing the expression of a full-length L1HS with an H3K4me3 peak on its promoter and RNA-seq signal (RPKM) split by direction of transcription (blue, forward; red, reverse).

Fig. 3.. L1s are expressed in human brain development.

(A) Schematic of sequencing strategy of fetal human forebrain samples. (B) Expression of primate-specific L1 subfamilies compared to ancient L1 subfamilies and selected housekeeping and development-related genes as reference. Row annotation showing average length, average percentage of divergence from consensus, and the total number of elements (information extracted from RepeatMasker open-4.0.5). (C) Read count in sense (light teal) and antisense (dark teal) per sample. First four boxplots showing full-length elements in young L1 subfamilies and last four showing ancient L1 subfamilies with a comparable number of copies. (D) Expression (RPKM) over full-length (>6 kbp) L1HS, L1PA2, L1PA3, and L1PA4 plus 6-kbp flanking regions. (E) Percentage of expressed full-length (>6 kbp) elements (mean normalized counts, >10; see Methods) among young L1 subfamilies (n = number of expressed elements; T = total number of full-length elements). (F) Detected H3K4me3 peaks (left heatmap) over full-length L1 subfamilies (L1HS to L1PA4) and RNA-seq signal (right heatmap). Profile plots showing sum of signal. (G) Fetal human forebrain snRNA-seq UMAP colored by cluster. (H) UMAP colored by cell types. (I) Expression of selected biomarkers for different cell types. (J) UMAP colored by cell cycle state (based on CellCycleScoring from Seurat). (K) Velocity plot colored by cell type. (L) Pseudo-bulk cluster expression of young L1 subfamilies on UMAP.

Fig. 4.. L1s are dynamically expressed in the developing and the adult human brain.

(A) Left: Number of expressed L1HS-L1PA4 (>6 kbp) in fetal (red) and adult samples (blue) (mean normalized counts, >10; see Methods) and the number of elements found to be expressed in both datasets (intersection; purple). Right: Number of H3K4me3 peaks over L1HS-L1PA4 (>6 kbp) in fetal (red) and adult samples (blue) and the intersection between datasets (purple). (B) Genome browser track showing the expression of a development-specific full-length L1PA2 with an H3K4me3 peak at its promoter. (C) Number of intragenic (light) or intergenic (dark) L1HS to L1PA4 (>6 kbp) in fetal (red), adult (blue), or those expressed in both datasets (purple). (D) Log₂FoldChange (log₂FC) of the genes with an intragenic L1HS to L1PA4 (>6 kbp) in fetal (red), adult (blue), and the intersection (purple) [fetal versus adult (ref); DESeq2]. (E) L1s initiating antisense transcripts. Top: Schematic definition of L1 chimeras. Bottom: Total number of L1 chimeras expressed in fetal and adult samples. Number of the subset de novo annotated transcripts (not present in GENCODE hg38 version 38) in italics. (F and G) Genome browser tracks showing (from top to bottom): H3K4me3 CUT&RUN (samples overlayed in purple), short-read bulk RNA-seq (overlayed) split by strand (blue, forward; red, reverse), overlayed cluster expression (adult snRNA-seq) per cell type (or group of cell types), and PacBio Iso-Seq reads validating the presence of the transcript (supporting reads are highlighted in red). Annotation to the right showing data type and dataset (adult/fetal). (F) SYT1 with an antisense full-length L1PA2 at the beginning of one of its isoforms (L1 chimera). snRNA-seq tracks showing excitatory neurons (EN), inhibitory neurons (IN), and non-neuronal cell types overlayed [astrocytes, oligodendrocyte precursor cells (OPC), oligodendrocytes (Oligo), and microglia]. (G) ZNF638 with an antisense full-length L1HS as an alternative promoter (L1 chimera).

Fig. 5.. The L1-lncRNA LINC01876 is a human-specific transcript.

(A) Genome browser tracks showing RNA-seq and H3K4me3 signal (bottom) (in purple) over L1-lncRNA in fetal and adult samples. RNA-seq signal (RPKM) split by strand (blue, forward; red, reverse). Right: A zoom-in into the TSS (highlighted in yellow). (B) Experimental approach for fbNPCs human and chimpanzee comparison. (C) Genome browser tracks showing RNA-seq and H3K4me3 signal (bottom) (in purple) over L1-lncRNA in human and chimpanzee fbNPCs. RNA-seq signal (RPKM) split by strand (blue, forward; red, reverse). Right: A zoom-in into the TSS (highlighted in yellow). (D) LINC01876 (L1-lncRNA) expression [transcripts per million (TPM)] from bulk RNA-seq of human, chimpanzee, bonobo, gorilla, and macaque rhesus NPCs from Linker et al. (47). (E) Percentage of cells expressing LINC01876 (L1-lncRNA) in human, chimpanzee, and macaque rhesus cerebral organoids from Kanton et al. (48). (F) Multiple sequence alignment of the L1-lncRNA L1PA2 ORF0 (highlighted in purple) in different primates and their Kozak sequence (highlighted in yellow). The TSS of the L1-lncRNA is indicated in orange.

Fig. 6.. CRISPRi-silencing of the L1-lncRNA in human fbNPCs.

(A) CRISPRi construct and schematic of the L1-lncRNA CRISPRi in fbNPCs. (B) Genome browser tracks showing the expression over L1-lncRNA in control (LacZ) and L1-lncRNA CRISPRi. RNA-seq signal (RPKM) split by strand (blue, forward; red, reverse). (C) Immunohistochemistry of forebrain (red, FOXG1) and nuclear [blue, 4′,6-diamidino-2-phenylindole (DAPI)] markers. Enhanced green fluorescent protein (eGFP) showing transfected cells (green). Scale bars, 128 μm. (D) Volcano plot showing differential gene expression results (DESeq2). Significantly up-regulated and down-regulated genes are highlighted in red and blue, respectively (log2FoldChange > 1, P_adj < 0.05). (E) Log2FoldChange of the significantly up-regulated or down-regulated genes upon L1-lncRNA CRISPRi [as highlighted in (D) in the two datasets (L1-lncRNA CRISPRi versus control and human versus chimp). Genes up-regulated or down-regulated in both datasets are highlighted in red (first and third quadrants). (F) Normalized expression (median of ratios; DESeq2) of two example genes up-regulated in both datasets.

Fig. 7.. Silencing of L1-lncRNA in cerebral organoids indicates it has a role in developmental timing.

(A) Schematic of experimental design for organoid differentiation, L1-lncRNA CRISPRi, and sequencing. DEA, Differential Expression Analysis; KD, knock down. (B) Bright-field pictures of iPSC-derived cerebral organoids (top). Black scale bars, 200 μm. Immunohistochemistry of PAX6 (orange), ZO1 (red), and DAPI (blue) (bottom). White scale bars, 100 μm. (C) Quantification of organoid diameter between days 10 and 30 (n = 20 to 30 organoids per time point) (mixed-effects analysis and a Sidak correction for multiple comparisons). (D) snRNA-seq: UMAP colored by cluster. (E) UMAP colored by identified cell types. Neuronal-like clusters colored in two shades of green and uncharacterized clusters or progenitor-like cells colored in grey. VIL+ , Villin 1 positive cells. (F) Dot plot showing expression of neuronal and neuronal progenitor markers in the NPC and neuronal clusters. (G) UMAP showing expression of L1-lncRNA. (H) Selected examples of significantly up-regulated genes in L1-lncRNA CRISPRi NPCs (FindMarkers from Seurat; P_adj < 0.05). (I) Selected up-regulated terms of the gene set enrichment analysis (GSEA) over NPCs (gseGO; P_adj < 0.05). (J) Selected examples of significantly up-regulated genes in L1-lncRNA CRISPRi neurons (FindMarkers from Seurat; P_adj < 0.05). (K) Selected up-regulated terms of GSEA over neurons (gseGO; P_adj < 0.05).