mRNA and circRNA mislocalization to synapses are key features of Alzheimer's disease

Abstract

Proper transport of RNAs to synapses is essential for localized translation of proteins in response to synaptic signals and synaptic plasticity. Alzheimer's disease (AD) is a neurodegenerative disease characterized by accumulation of amyloid aggregates and hyperphosphorylated tau neurofibrillary tangles followed by widespread synapse loss. To understand whether RNA synaptic localization is impacted in AD, we performed RNA sequencing on synaptosomes and brain homogenates from AD patients and cognitively healthy controls. This resulted in the discovery of hundreds of mislocalized mRNAs in AD among frontal and temporal brain regions. Similar observations were found in an APPswe/PSEN1dE9 mouse model. Furthermore, major differences were observed among circular RNAs (circRNAs) localized to synapses in AD including two overlapping isoforms of circGSK3β, one upregulated, and one downregulated. Expression of these distinct isoforms affected tau phosphorylation in neuronal cells substantiating the importance of circRNAs in the brain and pointing to a new class of therapeutic targets.

Fig 1. RNA sequencing of synaptosomes is an effective strategy for identifying localized RNA transcripts in the human frontal lobe.

A. Schematic of synaptosome isolation, which are particles formed during the shearing forces of brain tissue homogenization whereby synaptic terminals pinch-off and reseal to form intact bipartite particles consisting of pre- and post-synaptic compartments. Based on their density, synaptosomes can be purified by sequential centrifugation steps concluding with gradient ultracentrifugation. After purification, the RNA contents of synaptosomes can be isolated and sequenced. Figure created using a biorender.com license. B. Western blot of a synaptic marker, synaptophysin, a nuclear marker, NeuN, and a microglial marker, IBA1, with Actin as a loading control for the different fractionation steps during synaptosome preparation in control samples. C. Quantification of the blot in (B). N = 4 per group. Significance calculated via T-test. *: p < 0.05, **: p < 0.01, ****: p < 0.0001. D. Multidimensional scaling analysis demonstrating separation between mRNA counts of homogenate and synaptosome fractions as well as variability between sporadic AD samples and controls. E. Volcano plot comparing synaptosome vs homogenate mRNA enrichment in only control samples. A distinct minority of mRNAs are significantly enriched at synaptic terminals (p < 0.01, gold box). F. Correlation between the fold change/enrichment of synaptosome vs homogenate between AD and control (R² = 0.9266). G. Venn diagram demonstrating overlap of synaptosome enriched genes between AD and control samples. H. GO biological process analysis of synaptosome-enriched genes.

Fig 2. Comparison between AD and control in synaptosomes vs homogenate reveals mislocalized transcripts in the human frontal lobe.

A. Volcano plot comparing expression differences between AD and control in the bulk, unfractionated homogenate. B. Volcano plot comparing expression differences between AD and control just within the synaptosome particles. C. Venn diagram comparing the significant expression differences in homogenate (A) and synaptosome (B). D. Comparison of the fold changes of AD vs control between synaptosomes and homogenate (R² = 0.2327). Genes that have a T-statistic most distant from a y = x trendline are significantly mislocalized, independent of global expression changes (FDR < 0.001, see methods). E. GO biological process analysis of upregulated genes in (D). F. GO biological process analysis of downregulated genes in (D).

Fig 3. Synaptosome sequencing in the human temporal lobe reveals similarities among synaptosome-enriched transcripts and differences in mislocalized transcripts.

A. Multidimensional scaling analysis among the frontal lobe and temporal lobe samples. B. Comparison of synaptosome-enriched transcripts between frontal and temporal lobe samples (R² = 0.886). C. Venn diagram showing overlap of synaptosome-enriched transcripts between frontal and temporal lobe samples (p < 0.01). D. Comparison of the fold changes of AD vs control between synaptosomes and homogenate in temporal lobe samples (R² = 0.3282). Genes that have a T-statistic most distant from a y = x trendline are significantly mislocalized, independent of global expression changes (FDR < 0.01, see methods). E. Venn diagram showing minimal overlap of mislocalized transcripts between frontal and temporal lobe samples. F. GO biological process analysis of upregulated genes in (D). G. GO biological process analysis of downregulated genes in (D).

Fig 4. Synaptosome sequencing of frontal lobe samples from the APP^swe/PSEN1^dE9 AD and wild-type mouse models reveals substantial differences in transcript localization patterns from that of human samples.

A. Volcano plot comparing synaptosome vs homogenate in wild-type mouse samples (FDR < 0.001). B. Comparison of synaptosome-enriched transcripts between mouse and human frontal lobe samples (R² = 0.2164). C. Venn diagram showing overlap of synaptosome-enriched transcripts between human and mouse frontal lobe samples (p < 0.01). D. Comparison of the fold changes of AD vs control between synaptosomes and homogenate in temporal lobe samples (R² = 0.0729). Genes that have a T-statistic most distant from a y = x trendline are significantly mislocalized, independent of global expression changes (FDR < 0.01, see methods). E. Venn diagram showing minimal overlap of mislocalized transcripts between mouse and human and mouse frontal lobe samples. F. GO biological process analysis of upregulated genes in (D). G. GO biological process analysis of downregulated genes in (D).

Fig 5. Analysis of circular RNAs in synaptosomes shows their preferential localization to synapses and substantial differences in AD samples.

A. Schematic of circular RNAs (circRNAs) that are formed by back-splicing a later exon onto a preceding exon. Figure created using a biorender.com license. B. Workflow for analyzing circRNAs includes first mapping the sequencing data to the canonical transcriptome. Second, unmapped reads are extracted from the alignment file. Third, unmapped reads are aligned to a custom list of all possible back-splice junctions within genes including 150bp flanking each side. Finally, reads mapping to back splice junctions are counted and differential expression is evaluated using DESeq2. C. Volcano plot comparing circRNAs in synaptosome vs homogenate in control frontal lobe samples. The overwhelming majority of circRNAs are significantly enriched at synaptic terminals. D. Volcano plot comparing circRNAs in AD vs control in the unfractionated homogenate (p < 0.05). E. Volcano plot comparing circRNAs in AD vs control in the synaptosome fraction (p < 0.05). F. Venn diagram showing overlap of AD vs control differentially expressed circRNAs between the homogenate and synaptosome fractions. G. Heatmap of differentially expressed circRNAs between AD vs control in synaptosomes. Differential expression is both independent of, and substantially greater than, the bulk homogenate.

Fig 6. CircGSK3β isoforms show contrasting expression differences between AD and control and may play regulatory roles related to AD pathologies.

A. Two distinct isoforms of circGSK3β with contrasting expression differences in AD. An isoform where exon 9 is back-spliced onto exon 7 is upregulated in AD, while an exon 10 spliced onto exon 9 isoform is downregulated. Figure created with a biorender.com license. B. Confirmation of circGSK3β isoform back-splice junctions by Sanger sequencing using primers flanking the splice junction. C. BaseScope RNA hybridization probes show that circGSK3β isoforms (green) are more frequently distal to the nucleus than the linear GSK3β transcript (red) in human control frontal lobe brain slices. Tissue and nuclei were stained with hematoxylin. D. A specialized plasmid containing Alu elements coerces circular back-splicing of gene insert [59]. E. Western blot of neuron-differentiated SH-Sy5y cells transfected with either a control plasmid, a circGSK3β exon 9–7 plasmid, or a circGSK3β exon 10–9 plasmid. Cell lysates were probed for total Tau (Tau5), phospho-Tau (AT8), GSK3β, phospho-S9 GSK3β, and GAPDH. F. Quantification of the ratios between phospho-Tau and total Tau as well as between phospho-S9 GSK3β and total GSK3β in each of the transfection conditions in (E). N = 8 samples per group. Significance calculated by T-test. *: p < 0.05, **: p < 0.01, ***: p < 0.001.