Sensitivity to TDP-43 loss and degradation resistance determine cryptic exon biomarker potential

Abstract

Cryptic splicing caused by TDP-43 proteinopathy is a hallmark of the neurodegenerative diseases amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD). However, which cryptic splicing events (CEs) are the most sensitive to TDP-43 depletion, where CEs localise within cells, and how specific CEs are in human tissues is poorly defined. Analyses of in vitro TDP-43 knockdowns and postmortem RNA-seq datasets revealed that a small subset out of thousands of CEs are specific markers for TDP-43 proteinopathy in vivo. Nonsense-mediated decay (NMD) masked a portion of CEs, influencing their subcellular localization and detectability in tissue. Dose-dependent TDP-43 depletion identified "early-responsive" CEs, which possess stronger splice sites and denser, more canonical TDP 43 binding motifs. Finally, we developed a composite cryptic burden score that effectively captured TDP-43 pathology across heterogeneous tissues and correlated with regional vulnerability and genetic background. Our work identifies robust biomarkers and offers new insights into TDP-43-mediated splicing dysregulation in neurodegeneration.

Extended Data Figure 1.. Sequencing features of the postmortem cohorts.

A. Age at death (top left), RIN (top right), postmortem interval (bottom left), and library depth (bottom right) in the different cohort analysed (NYGC ALS motor cortex, NYGC ALS spinal cord, NYGC FTLD frontal/temporal cortex, RiMOD FTLD frontal cortex). Wilcoxon test. B. Correlation of the fraction of samples with TDP-43 pathology where a CE is detected (y-axis) and the mean gene expression of that gene in tissues with TDP-43 pathology (x-axis) in the four cohorts (NYGC ALS motor cortex, NYGC ALS spinal cord, NYGC FTLD frontal/temporal cortex, RiMOD FTLD frontal cortex). C. Ranking of genes with motor cortex specific CEs based on their expression in the spinal cord. D. Ranking of genes with spinal cord specific CEs based on their expression in the motor cortex. E. Proportion of TDP-43-proteinopathy specific CE junctions derived from the bulk RNA-seq from NYGC ALS motor cortex, NYGC ALS spinal cord, NYGC FTLD frontal/temporal cortex, and RiMOD FTLD frontal cortex upregulated in TDP-43 negative FACS-sorted neuronal nuclei from FTLD frontal cortex. Box plots: boundaries 25–75th percentiles; midline, median; whiskers, Tukey style. Significance levels reported as * (p < 0.05) ** (p < 0.01) *** (p < 0.001) **** (p < 0.0001).

Extended Data Figure 2.. TDP-43 loss does not alter NMD efficiency, but NMD-sensitivity determines CE detection in postmortem tissue.

A. Counts of significantly changed poison exon junctions by PSI (3,3% bins) in the different NMD experiments. B. Volcano plot depicting PSI change of poison exon junctions detected in the NMD experiments between NMD inhibition and control condition. C. Volcano plot depicting PSI change of poison exon junctions detected in TDP-43 knockdown experiments. D. Differential gene expression of the 50 most upregulated genes between UPF1 KD and control condition, between UPF1 KD and control condition (left) and between TARDBP KD and control condition (right). E. Number of dark CE, including total number, number undetected in other TDP-43 knockdown, number observed in postmortem tissue, and number selective for TDP-43 pathology, divided by the three NMD inhibition experiments.

Extended Data Figure 3.. Cryptic splicing sensitivity to TDP-43 levels is reliable across two different neuronal lines and is not driven by sensitivity to NMD.

A. TARDBP normalised counts from RNA-seq experiment in SH-SY5Y (left) and SK-N-BE(2) (right) cells. B. Library depth, expressed as uniquely mapped reads, for each doxycycline level in SH-SY5Y (left) and SK-N-BE(2) (right) cells. C. Proportion of genes that are either significantly upregulated (left) or downregulated (right)(adj. p-value < 0.05), split by responsiveness category, in SH-SY5Y (top) and SK-N-BE(2) (bottom) cells. Genes not significant differential not plotted. D. Spaghetti plot showing the response of early (left) and late-responsive (right) last exon cryptic splicing events to TDP-43 loss of function in the SH-SY5Y (top) and SK-N-BE(2) (bottom) cells. E. Scatter plot displaying the PSI at the strongest TDP-43 knockdown level in SH-SY5Y (x-axis) and SK-N-BE(2) cells (y-axis). F. Number of CE junctions detected in each dose-response category split by NMD- (blue) and non-NMD (red) rescue in i3-derived cortical-like neurons with UPF1 KD (top) and SH-SY5Y cells treated with SMGi-j11 (bottom).

Extended Data Figure 4.. Cryptic splicing burden varies by region and disease mutation.

A. Cryptic splicing burden in cervical, thoracic, and lumbar spinal cord from NYCG ALS-TDP samples. B. Correlation between cryptic splicing burden of lumbar spinal cord (x-axis) and motor cortex (y-axis) in the same patients in ALS-TDP. C. Cryptic splicing burden between different genetic types of FTLD (unknown, C9orf72, and GRN) in NYGC FTLD frontal cortex (left) and NYGC FTLD temporal cortex (right), considering only FTLD type A cases.

Figure 1.. Identification of TDP-43 proteinopathy specific cryptic splicing from in vitro TDP-43 knockdowns.

A. Heatmap of overlapping cryptic splice events (CE) across TDP-43 knockdown datasets, clustered by overlap. Bars (right) show the number of CEs (control PSI <5%, knockdown >10%). The left colour scale indicates the log₂ fold change of TARDBP expression in knockdown versus control. B. Detection of common CEs across TDP-43 knockdown datasets. CEs present in at least 6 of 10 datasets were manually classified as: (1) significant by splicing software (orange), (2) lowly expressed but not called (≥5 junction reads) (yellow), (3) lacking evidence of expression (blue), or (4) absent due to no expression of the parent gene (white). C. False positive rate of CEs in the NYGG and RiMOD datasets based on the number of in vitro datasets where a CE was found. D. Upset plot of genes with a disease-specific CE (FPR ≤ 10%, TPR ≥10 % and detected in at least five distinct TDP-43 proteinopathy tissue samples). Highlighted genes are those with disease specific CEs present across all datasets and those only in cortical samples.

Figure 2.. Nonsense-mediated decay affects cryptic splicing detectability, localisation, and selectivity for TDP-43 proteinopathies.

A-C) Percent spliced in (PSI) of cryptic splicing events (CE) in (A) SH-SY5Y cells after nonsense-mediated decay (NMD) inhibition by CHX treatment, (B) i3-derived cortical-like neurons after NMD inhibition by UPF1 knockdown, (C) SH-SY5Y cells after NMD inhibition by SMGi-jll, split by genes that are rescued by more than 5% (left, blue) or less (right, red). Highlighted genes include exemplar genes where NMD-sensitivity has been described before and example dark cryptic genes. D. Nuclear to cytoplasmic ratio NMD- (blue) and non-NMD (red) sensitive CEs. E. Gene expression changes after TDP-43 knockdown in genes carrying CEs, divided by NMD category, gene expression, and subcellular fraction. Box plots: boundaries 25–75th percentiles; midline, median; whiskers, Tukey style. Wilcoxon test. F. Scatter plots relating the nuclear to cytoplasmic ratio of annotated junctions (x-axis) and the nuclear to cytoplasmic ratio of cryptic junctions (y-axis) in baseline condition, split between NMD rescued and non-NMD rescued CEs. G. Scatter plots relating the nuclear to cytoplasmic ratio of annotated junctions (x-axis) and the nuclear to cytoplasmic ratio of cryptic junctions (y-axis) after NMD inhibition, split between NMD rescued and non-NMD rescued CEs. H. Detection rate of NMD- (blue) and non-NMD-rescued (red) genes based on the three NMD inhibition experiments in the NYGC ALS motor cortex, NYGC ALS spinal cord, NYGC FTLD frontal/temporal cortex, RiMOD FTLD frontal cortex. Fisher’s exact test. I. Selectivity rate of NMD- (blue) and non-NMD-rescued (red) genes based on the three NMD inhibition experiments in the NYGC ALS motor cortex, NYGC ALS spinal cord, NYGC FTLD frontal/temporal cortex, RiMOD FTLD frontal cortex. Fisher’s exact test. J. Number of dark CEs across three NMD-inhibition datasets, including total number, number undetected in other TDP-43 knockdowns, number observed in postmortem tissue (≥2 junction reads), and number disease selective for TDP-43 proteinopathy. Significance levels reported as * (p < 0.05) ** (p < 0.01) *** (p < 0.001) **** (p < 0.0001).

Figure 3.. Cryptic splicing shows differential responsiveness to TDP-43 loss of function.

A. RT-qPCR in SH-SY5Y (left) and SK-N-BE(2) (right) cells showing levels of TARDBP, normalised to GAPDH, compared to the average in the untreated samples. B. Differential gene expression in the SH-SY5Y (left) and SK-N-BE(2) (right) experiment. Number of up and downregulated genes (adj. p-value < 0.05) called by DESeq2 shown at each level. UNC13A (salmon) and STMN2 (red) and gene expression highlighted. C. Spaghetti plot showing the response of cryptic splicing events (CE) to TDP-43 loss of function in the SH-SY5Y (left) and SK-N-BE(2) (right) cells. D. Alluvial plot showing the overlap of the early, intermediate, and late-responsive CEs in SH-SY5Y and SK-N-BE(2) cells. E. Number of CE junctions detected in each dose-response category split by NMD- (blue) and non-NMD (red) rescue in SH-SY5Y cells treated with CHX. No significant change in proportion of NMD-rescued vs non-rescued CE, pairwise fisher’s exact test. F. Heatmap visualisation of differential expression of CE-bearing genes categorised by NMD sensitivity in SH-SY5Y cells. Significantly differentially expressed (adj. p-value < 0.05) genes marked with *. G. Detection rate (left) and selectivity rate (right) in the NYGC ALS motor cortex, NYGC ALS spinal cord, NYGC FTLD frontal/temporal cortex, and RiMOD FTLD frontal cortex between early-responsive (pink) and late-responsive (blue) CEs. Numbers above graphs represent the number of CEs meeting either detection or selectivity criteria. Fisher’s exact test. Significance levels reported as * (p < 0.05) ** (p < 0.01) *** (p < 0.001) **** (p < 0.0001).

Figure 4.. Splice site strength and TDP-43 binding density determine cryptic splicing sensitivity.

A. Pangolin splice site probability for early-responsive (pink) and late-responsive (blue) cryptic splicing events (CE). Wilcoxon test. B. Fraction of TDP-43 bound junctions (based on POSTAR3 dataset) around CEs acceptor (left) and donor (right), for early-responsive (pink) and late-responsive (blue) CEs compared to 500 randomly sampled exons (grey). Region inside CEs highlighted in yellow. C. Percentage of TDP-43 iCLIP within 250 base pairs (bp) based on SH-SY5Y iCLIP data between early-responsive and late-responsive cryptic and annotated splicing. D. Whole-intron UG repeat density for early-responsive (pink) and late-responsive (blue) CEs. Wilcoxon test. E. UG fraction of TDP-43 binding sites near acceptor and donor splice sites between early-responsive (pink) and late-responsive (blue) CEs. Wilcoxon test. F. Sequence logo plot depicting probability of each nucleotide for 50 positions from the donor splice site between early-responsive (top) and late-responsive (bottom) CEs based on SH-SY5Y categories. Splice site highlighted in grey. G. Hexamer frequency of TDP-43 bound regions by TDP-43 binding category36 per kilobase in acceptor (left) and donor (right) splice site between early-responsive (pink) and late-responsive (blue) CEs. Fisher’s exact test. Box plots: boundaries 25–75th percentiles; midline, median; whiskers, Tukey style. Significance levels reported as * (p < 0.05) ** (p < 0.01) *** (p < 0.001) **** (p < 0.0001).

Figure 5.. A cryptic burden score reveals disease heterogeneity.

A. Receiving operator curves using cryptic splicing burden or top two individual CE predictors in NYGC ALS spinal cord (left) and NYGC ALS motor cortex (right). B. Scatter plot displaying frontal cortex (x-axis) and temporal cortex (y-axis) cryptic exon burden in NYGC FTLD cases split by pathological type (A, B, or C). C. Differential cryptic splicing burden between temporal and frontal cortex in the same patient for NYGC FTLD type A and type C. Wilcoxon test. D. Cryptic splicing burden between controls and different genetic types of FTLD (MAPT, C9orf72, and GRN) in the RiMOD dataset. Wilcoxon test. E. Cryptic splicing burden between different genetic types of FTLD (unknown, C9orf72, and GRN) in NYGC FTLD frontal cortex (left) and NYGC FTLD temporal cortex (right). Wilcoxon test. Box plots: boundaries 25–75th percentiles; midline, median; whiskers, Tukey style. Significance levels reported as * (p < 0.05) ** (p < 0.01) *** (p < 0.001) **** (p < 0.0001).