A neurotoxic cryptic peptide arising from TDP-43-dependent cryptic splicing of PKN1

Abstract

Dysfunction of transactive response DNA-binding protein 43 (TDP-43) drives neurodegeneration in amyotrophic lateral sclerosis (ALS) and Alzheimer’s disease (AD), in part through inducing aberrant RNA splicing. However, whether such mis-splicing yields stable, pathogenic proteins remains unclear. Here, we identify a TDP-43–repressed cryptic exon in Protein kinase N1 (PKN1), designated PKN1-5a1, which is activated in ALS patient brains and introduces a premature termination codon. This aberrant transcript escapes nonsense-mediated decay and is translated into a truncated peptide, PKN1-N207 (PKN207), detectable in AD brains with TDP-43 pathology. In mice, PKN207 impairs cognition, memory, and synaptic plasticity. Our findings demonstrate that TDP-43 loss–induced cryptic splicing can generate stable neurotoxic polypeptides, revealing a peptide-mediated mechanism in TDP-43 proteinopathies.

Loss of TDP-43 induces a pathogenic cryptic splicing event in the PKN1 gene. The resulting stable, neurotoxic N207 peptide is detected in Alzheimer’s disease brains and drives cognitive deficits in mice, revealing a new mechanism in TDP-43 proteinopathy.

Introduction

Transactive response DNA-binding protein 43 (TDP-43) is a conserved RNA-binding protein that plays a key role in maintaining RNA splicing fidelity in the nervous system¹. By recognizing UG-rich sequences in pre-mRNA, TDP-43 effectively suppresses aberrant insertion of unannotated cryptic exons, thereby ensuring normal expression of key neuronal genes^2,3. Loss of TDP-43 function is widely recognized as a core pathological feature of amyotrophic lateral sclerosis (ALS)⁴, and is also commonly observed in frontotemporal dementia (FTD) and Alzheimer’s disease (AD). In FTD, TDP-43 pathology accounts for approximately 40–50% of cases, characterized by neuronal nuclear inclusions, cytoplasmic deposition, and concomitant gliopathy in the frontal and temporal lobes, and is closely associated with distinct clinical subtypes^5,6. In AD, abnormal TDP-43 deposition is detected in about 50–70% of patients (with proportions varying across cohorts)^1,4,7,8. Notably, limbic-predominant age-related TDP-43 encephalopathy (LATE), a special subtype of AD, is strongly linked to cognitive impairment in the elderly^8–10.

In recent years, studies have gradually elucidated how loss of TDP-43 function disrupts neuronal gene expression through cryptic splicing and thereby mediates dysfunction¹. For example, in STMN2 (Stathmin 2), TDP-43 loss activates an intronic aberrant event and introduces a premature polyadenylation site, causing transcription to terminate prematurely near exon 2 and preventing the production of functional STMN2 protein^11,12; in UNC13A (unc-13 homolog A), TDP-43 loss induces insertion of a cryptic exon carrying a premature termination codon (PTC), triggering nonsense-mediated mRNA decay (NMD) and markedly downregulating functional UNC13A protein^13,14. These mechanisms directly point to impairments in core processes such as axon growth and synaptic release, and are considered to constitute an important molecular basis of TDP-43 proteinopathies^11,12,15,16. In addition, because of their specificity in ALS and early activation, cryptic splicing events are becoming potential biomarkers for disease subtyping and diagnosis^17–21.

Notably, TDP-43–dependent cryptic splicing does not end solely in “loss of function”. Some events maintain open reading frame (ORF) continuity or retain key domains, thereby generating de novo proteins that participate in pathological processes. In the reported HDGFL2 (hepatoma-derived growth factor-related protein 2) event, the corresponding product can be detected in brain and cerebrospinal fluid (CSF) of ALS/FTD patients²²; cryptic splicing of MYO18A (myosin XVIIIA) likewise produces an unannotated protein that retains functional domains and is hypothesized to be aggregation-prone²³. These findings suggest that certain cryptic splicing events may exacerbate the progression of TDP-43 proteinopathies through gain-of-toxicity pathways.

Protein kinase N1 (PKN1) is a serine/threonine kinase involved in multiple processes, including cytoskeletal dynamics, axonal transport, intracellular signal transduction, and synaptic plasticity^24,25. PKN1 can bind to neurofilament light chain (NEFL) and phosphorylate its head domain, thereby regulating neurofilament assembly and axonal structural stability²⁶. Its activity is regulated by Rho-family small GTPase Rac1 and phosphoinositide-dependent protein kinase 1 (PDK1), and undergoes translocation or activation under hyperosmotic/stress conditions²⁷. Under apoptotic stimulation, caspase-3 cleaves PKN1 to form a truncated species that is thought to have kinase activity distinct from the full-length conformation and potential toxic effects²⁸.

In this work, we show that TDP-43 loss induces a cryptic splicing event in the PKN1 gene, designated PKN1-5a1, which inserts a 127-bp exon and introduces PTCs. This aberrant transcript partially escapes NMD and is translated into a stable truncated peptide, PKN1-N207 (PKN207). Using a specifically developed antibody, we detect PKN207 in AD brains with TDP-43 pathology. In vivo studies demonstrate that PKN207 expression impairs cognitive function and synaptic plasticity in mice. Proteomic profiling further reveals distinct alterations in neuronal pathways induced by PKN207. Our findings indicate that TDP-43 dysfunction not only causes loss-of-function effects but can also generate stable, neurotoxic polypeptides, revealing a peptide-level pathological mechanism in TDP-43 proteinopathies.

Results

TDP-43 knockdown induces PKN1 cryptic splicing and disrupts its protein expression

TDP-43 plays a pivotal role in maintaining RNA splicing fidelity, primarily by suppressing the aberrant insertion of cryptic exons. Loss of TDP-43 function leads to abnormal cryptic splicing in multiple crucial neuronal genes, resulting in a marked reduction of functional protein expression. Among these, STMN2 and UNC13A represent the most extensively studied targets, and their splicing dysregulation has been established as a key molecular event in neurodegenerative diseases^1,11–14,29. Furthermore, other genes including PFKP (phosphofructokinase, Platelet), HDGFL2, and GAP43 (growth associated protein 43) have also been reported to be regulated through this mechanism. Notably, the aberrant splicing of STMN2 and GAP43 has been directly validated to impair axonal regeneration, linking these molecular defects to specific pathological phenotypes^{1,15,21,22,29}. To discover additional TDP-43–regulated cryptic splicing events, we knocked down TDP-43 in neuroblastoma M17 cells and performed RNA-seq to systematically identify potential targets. Integrative Genomics Viewer (IGV) visualization showed that several known cryptic exon events, such as STMN2, PFKP, and HDGFL2, were all repeatedly detected under TDP-43 knockdown^1,11,12,15, verifying the reliability of the sequencing data and the sensitivity of splicing-event detection (Supplementary Fig. 1a).

On this basis, we identified an aberrant splicing region between exons 4 and 5 of the PKN1 gene, in which two distinct splicing forms were detected: one involves insertion of a previously unannotated exon (127 bp), designated PKN1-5a1, with genomic coordinates hg38: chr19:14,450,088–14,450,214; the other is an alternative 3’ splice site (A3SS) of 230 bp with coordinates hg38: chr19:14,450,088–14,450,317, designated PKN1-A3SS (Fig. 1a). Among these, the PKN1-A3SS splicing form was previously reported in TDP-43–knockdown HeLa cells³⁰; here, we characterize the PKN1-5a1 exon, a previously unannotated cryptic exon. We define its precise splice boundaries and sequence features, and elucidate its biological significance. Further analysis showed that the two splice sites of PKN1-5a1 are the canonical splice donor “GU” and acceptor “AG” (Fig. 1a), consistent with classical alternative splicing sequence features³¹; meanwhile, PKN1-5a1 contains multiple stop codons (Supplementary Fig. 1b). We calculated the PSI (percent-spliced-in) value of this region based on RNA-seq data and found that TDP-43 knockdown primarily activates insertion of the 5a1 exon, whereas the A3SS insertion event is relatively infrequent (Fig. 1b), suggesting that PKN1-5a1 may be the major splicing abnormality driven by TDP-43 loss.

Fig. 1. TDP-43 loss induces PKN1 cryptic exon 5a1 inclusion and reduces PKN1 protein levels.

a IGV visualization of RNA-seq reads in shCTR and shTDP-43 M17 cells showing cryptic exon 5a1 (pink) and A3SS (blue) activation. b Splicing analysis of RNA-seq data showing PSI for 5a1 ≈ 0.75 and A3SS ≈ 0.11 in shTDP-43 M17 cells. c RT–PCR with F1/R1 primers in M17 cells showing TDP-43 knockdown induces two additional bands corresponding to A3SS (blue) and 5a1 (red), alongside the constitutive product (yellow). d RT-PCR with event-specific primers showing both the aberrant 5a1 (cf1-ceR1) and A3SS (cf1-ceR2) isoforms markedly increased in shTDP-43 M17 cells. e qRT–PCR showing upregulated 5a1 and A3SS in shTDP-43 M17 cells (n = 3, two-sided t-test). f Schematic of dCas13a-GFP/crRNA imaging system targeting PKN1-5a1. g Secondary structures of PKN1-5a1 and crRNA binding sites. h Immunofluorescence staining of PKN1-5a1 RNA (green) in HeLa cells; white arrow indicates nuclear puncta in shTDP-43 (n = 6). Scale bar, 10 μm. i Quantification of GFP-positive puncta (n = 30 cells/group, two-sided Wilcoxon test). j Western blots showing PKN1 downregulation in shTDP-43 M17 cells; quantification shown (n = 3, two-sided t-test). k Schematic of dual-fluorescence splicing reporter: 5a1 insertion abrogates mCherry via PTC. l Flow cytometry showing reduced mCherry-positive cells in shTDP-43 group. m PK-intron and PK-Δ5a1 reporter constructs and expected transcripts. n Western blots of FLAG-PKN1 from reporters: shTDP-43 reduces signal from PK-intron but not PK-Δ5a1; quantification shown (n = 3, two-sided t-test). Source data are provided as a Source Data file.

We knocked down TDP-43 in multiple human cell types—including M17, HEK-293T, HeLa, and iNeurons (iPSC-derived cortical neurons)—and analyzed PKN1 splicing patterns. The iPSC-derived cortical neurons were obtained by differentiating human induced pluripotent stem cells (iPSC) and validated by MAP2 and NeuN immunostaining (Supplementary Fig. 1j), indicating successful differentiation into mature neurons. In M17 cells, we performed RT-PCR to detect splicing products in this region and observed that, in addition to the normally spliced product, two aberrant bands appeared after TDP-43 knockdown (Fig. 1c), corresponding to insertion of the PKN1-5a1 and A3SS splicing events (Fig. 1d), respectively, indicating that at least two forms of alternative splicing exist in this region. Subsequent qRT-PCR analysis showed that TDP-43 loss significantly upregulated insertion of PKN1-5a1 in M17, HEK-293T, HeLa, and iNeurons, whereas the A3SS event was detectable only in M17 cells (Fig. 1e and Supplementary Fig. 1c, d) and was expressed at a much lower level than 5a1. This suggests that TDP-43’s suppression of aberrant splicing of PKN1-5a1 is more conserved.

To directly visualize the expression of PKN1-5a1 splicing product at the cellular level, we introduced a dCas13a-GFP (green fluorescent protein) system for detection^32,33. This system is based on a Cas13a variant from Leptotrichia wadei (dead LwaCas13a, dCas13a) whose nuclease activity is inactivated but which can still recognize target RNA under the guidance of a specific crRNA and form a fluorescent signal with GFP, thereby enabling spatial localization of target RNA molecules (Fig. 1f). We designed two crRNAs (designated crRNA-1 and crRNA-2) that recognize different regions of PKN1-5a1, with crRNA-1 targeting the junction between 5a1 and exon 4 and crRNA-2 targeting the junction between 5a1 and exon 5 (Fig. 1g). To ensure accessibility for binding, we used MFold³⁴ to predict the secondary structure of the RNA sequence of the PKN1-5a1 splicing product. The results showed that the sequences targeted by crRNA-1 and crRNA-2 were both located in single-stranded loop/bulge regions exposed on the RNA surface, displaying good accessibility suitable for dCas13a binding (Fig. 1g). We then co-transfected the dCas13a-GFP system with either crRNA-1 or crRNA-2 into TDP-43–knockdown HEK-293T cells. After cell fixation, TDP-43 was detected by immunostaining using a TDP-43 antibody to assess knockdown efficiency, and the dCas13a-GFP signal was visualized with a GFP antibody. We found that magenta fluorescence was markedly reduced in the TDP-43–knockdown group, indicating a clear decrease in its expression. Meanwhile, distinct green fluorescent puncta were observed predominantly in the nuclei of TDP-43–deficient cells, whereas no obvious signal was detected in the control group (Fig. 1h and Supplementary Fig. 1e). Quantification of green puncta per cell further showed a significant increase in the number of dCas13a-GFP puncta in the shTDP-43 group (Fig. 1i and Supplementary Fig. 1f), suggesting that this cryptic splicing product is activated under TDP-43 deficiency and can be specifically recognized and visualized.

More importantly, we found that TDP-43 knockdown significantly reduced PKN1 protein level (Fig. 1j and Supplementary Fig. 1g-i). We therefore constructed a dual-fluorescence, splicing-dependent translation reporter system to verify the impact of PKN1-5a1 insertion on translation. In this system, EGFP expression serves as an internal control, whereas mCherry expression depends on whether splicing proceeds normally. Once the PKN1-5a1 exon is inserted, the introduction of premature termination codons blocks downstream mCherry translation (Fig. 1k). After co-transfection of this system with shTDP-43 or shCTR into HEK-293T cells, Western blots showed that mCherry protein was completely absent in the TDP-43 knockdown group (Supplementary Fig. 1k); flow cytometry further showed that before knockdown, double-positive (mCherry and EGFP) cells accounted for 57.75%, whereas after knockdown, this proportion was decreased to 11.37% (Fig. 1l). These results suggest that insertion of PKN1-5a1 interrupts downstream translation, thereby aborting PKN1 protein synthesis.

To further verify whether deletion of cryptic exon 5a1 could rescue the decline in PKN1 protein levels under TDP-43 deficiency, we constructed two reporter plasmids: PK-intron (containing exons 4 and 5, and the full intronic interval between them, including 5a1) and PK-Δ5a1 (in the same background but with the 5a1 region specifically removed) (Fig. 1m). After co-transfecting these two plasmids with shTDP-43 into HEK-293T cells and performing Western blots analysis, we found that under TDP-43 knockdown, PKN1 protein expression was markedly reduced only in the PK-intron group, whereas remained essentially unchanged in the PK-Δ5a1 group (Fig. 1n). Interestingly, a 25–33 kDa FLAG band was appeared under TDP-43 knockdown in the PK-intron group, but no such band in the PK-Δ5a1 group (Fig. 1n). Together, these results indicate that insertion of PKN1-5a1 is the direct cause of the downregulation of PKN1 expression in the context of TDP-43 loss.

TDP-43 knockdown reduces PKN1 protein expression in mice but does not activate the PKN1-5a1 cryptic splicing

Given that the PKN1-5a1 splicing event is markedly activated under TDP-43 deficiency, we further investigated the evolutionary conservation of this exon and its expression characteristics in mouse models. The results showed that PKN1-5a1 is highly conserved in primates but lacks a corresponding conserved sequence in mice (Supplementary Fig. 2a, b). This feature is similar to previously reported cryptic splicing events in STMN2 and UNC13A^13,15, which likewise are conserved in humans or primates but absent or not spliced in mice, suggesting that such TDP-43–dependent cryptic splicing regulation exhibits pronounced species specificity during evolution.

Nevertheless, by immunofluorescence staining and Western blots, we observed a significant decrease in PKN1 protein levels in the hippocampus of TDP-43–knockdown mice (Supplementary Fig. 2c, d). However, IGV analysis and RT-PCR in TDP-43–knockdown N2a cells did not reveal obvious changes in PKN1 splicing patterns or mRNA levels. (Supplementary Fig. 2e). We speculate that TDP-43 knockdown may affect PKN1 protein stability in mouse cells. To test this hypothesis, after knocking down TDP-43 in N2a cells, we used cycloheximide (CHX) to inhibit protein translation and collected samples at 0, 4, 8, and 12 hours, assessing PKN1 expression by Western blots. We found that PKN1 protein was already decreased at 0 hours in shTDP-43 N2a cells, and its degradation rate accelerated significantly over time (Supplementary Fig. 2f). Although the rat genome harbors splice sites in this region that conform to the GU–AG rule, the overall sequence is not conserved relative to the human PKN1-5a1 region. Using non–exon-spanning primers for RT-PCR in TDP-43–knockdown rat B104 cells, we likewise did not observe bands other than the constitutively spliced product (Supplementary Fig. 2g). These results suggest that the presence of GU–AG splice signals alone is insufficient to drive cryptic exon splicing; the occurrence of the event may depend on higher-order cis-regulatory information.

In summary, although PKN1-5a1 lacks splicing activity in rodents, TDP-43 knockdown still induces a reduction of PKN1 protein expression in mouse brain tissue; this regulatory mechanism may involve protein stability rather than changes at the transcriptional or splicing level. These results suggest that TDP-43 regulation of PKN1 is markedly species-specific, and the physiological and pathological significance of its cryptic splicing remains to be further validated in primate models.

PKN1 cryptic exon in ALS neurons

In recent years, cryptic splicing events have been increasingly recognized as potentially disease-specific and important molecular markers in neurodegenerative disorders^{17–19,21,22,29}. Accordingly, we analyzed the publicly available ALS brain RNA-seq dataset (GSE153960) from the New York Genome Center (NYGC), and investigated the activation of PKN1 cryptic exon 5a1 to assess its clinical relevance as a candidate biomarker of ALS.

First, using RNA-seq reads covering all known PKN1 exons, we estimated full-length transcript expression with TPM (transcripts per million). Given that PKN1 has multiple splicing isoforms and sequencing depth varies across samples, TPM—being length-normalized—reflects within-sample abundance and is suitable for cross-sample comparisons, making it preferable for full-length transcript–level analyses. PKN1 expression was reduced not only in regions classically affected by TDP-43 pathology^4,35,36, e.g., frontal cortex, temporal cortex, motor cortex, and cervical/lumbar spinal cord; but, interestingly, also showed a similar trend in the cerebellum, which is typically considered less impacted by TDP-43 (Fig. 2a). This suggests that downregulation may not be solely attributable to loss of TDP-43 function. Given that PKN1 is a phosphor-protein involved in neuronal cytoskeletal regulation^24,25, we speculated that reduced PKN1 expression in the cerebellum may reflect broader neurodegenerative processes such as neuronal loss. We also noted varying degrees of PKN1 reduction in some “Other” group samples, although statistical differences were comparatively modest.

Fig. 2. PKN1 expression and cryptic exon 5a1 inclusion across brain regions in ALS.

a PKN1 (chr19:14,433,305–14,471,895, GRCh38) transcript expression across six brain regions showing a general decrease in ALS and ALS+ groups compared to neurologically normal controls (NC). P values are adjusted for multiple testing (two-sided Wilcoxon test). b Abundance of the PKN1-5a1 cryptic splicing event is significantly elevated in ALS and ALS+ samples in all regions except cerebellum. FDR-adjusted P values are shown (two-sided Wilcoxon test). Source data are provided as a Source Data file.

To delineate the regional distribution of PKN1 splicing abnormalities, we quantified PKN1-5a1 expression. Specifically, for the hg38 interval chr19:14,450,088–14,450,214, we calculated RPM (reads per million) as a relative measure of the cryptic exon. Compared with length-dependent metrics (e.g., TPM/FPKM), RPM is more appropriate for fixed loci such as cryptic exons, effectively controlling for between-sample depth differences and widely used for exon- or event-level quantification. Preliminary analyses detected reads over this interval in a subset of healthy controls (a read was counted toward PKN1-5a1 if ≥75% of its aligned length overlapped the 127-bp 5a1 region). However, junction-level visualization revealed that these signals largely arose from a low-abundance A3SS event (chr19:14,450,088–14,450,317) that shares the same 5′ donor with 5a1 (Fig. 1a), causing some junction reads to be misassigned to PKN1-5a1 and generating false positives in RPM analysis. To avoid A3SS contamination, we computed an endpoint-strict CE-pair RPM from RegTools/LeafCutter counts—counting only Exon4→CE and CE→Exon5 and discarding A3SS-exclusive junctions—which strengthened specificity and the reliability of between-region comparisons (see ‘Methods’).

After correction, PKN1-5a1 was significantly upregulated across multiple CNS regions in ALS, with the most pronounced changes in canonical disease sites—frontal cortex, temporal cortex, motor cortex, and cervical and lumbar spinal cord—whereas no clear changes were observed in the cerebellum (Fig. 2b). By contrast, the cryptic exon was virtually absent in healthy controls, supporting strong disease specificity. In the “Other” group, a subset of individuals also showed elevated PKN1-5a1, particularly in cortical regions, with levels intermediate between ALS and normal control (NC). This may indicate TDP-43 dysfunction in certain atypical cases, but this hypothesis requires further histological or molecular-pathological validation. Overall, PKN1-5a1 expression is concentrated in cortex and spinal cord—regions heavily affected by TDP-43 pathology^4,35,36—and remains low in the typically less affected cerebellum. This spatial pattern aligns with the regional selectivity of TDP-43 pathology in ALS, and further supports PKN1-5a1 as a candidate biomarker of TDP-43 proteinopathies–associated splicing abnormalities.

TDP-43 suppresses cryptic splicing of PKN1 through direct binding to conserved UG-rich elements in intron 5

As an aberrant splicing event abundantly detected in ALS patients (Fig. 2b), PKN1-5a1 has potential value as a target for intervention. Therefore, we further focused on the regulatory role of TDP-43 in splice selection of PKN1, exploring how it finely regulates the generation of this cryptic exon through binding to pre-mRNA, with the aim of providing a theoretical basis for mechanistic understanding and the development of subsequent therapeutic strategies.

To test this hypothesis, we analyzed TDP-43 CLIP-seq data in the POSTAR3 database³⁷. The results showed multiple high-confidence TDP-43 binding sites in the intronic region between exons 4 and 5 of PKN1, these sites are enriched for UG-repeat sequences and were ranked from high to low according to the crosslinking-induced truncation site (CITS) score (Fig. 3a). We further performed RNA immunoprecipitation (RIP) in M17 cells. Compared with the IgG control, TDP-43 specifically enriched PKN1 transcripts containing the 5a1 region (Fig. 3b), supporting the existence of direct binding between TDP-43 and PKN1 pre-mRNA in cells.

Fig. 3. TDP-43 binds UG-rich motifs near PKN1 exon 5a1 to suppress its inclusion.

a Schematic of six candidate TDP-43 binding sites (site1–site6) between exons 4 and 5 of PKN1, ranked by CLIP-seq CITS scores. b RIP-qPCR (n = 3) and Western blots showing TDP-43 binding to the 5a1-containing RNA region. HC: heavy chain of IgG. c Minigene PKmini recapitulates TDP-43-dependent suppression of 5a1 inclusion; shTDP-43 induces 5a1 insertion (n = 3). d, e Deletion of site1–5 (PKmini^Δ1–5, d) increases 5a1 inclusion (e) (n = 3). f,g Site6 partially mediates residual TDP-43 repression upon site1–5 deletion. In the PKmini^∆1–5 reporter, shTDP-43 (f) enhances, overexpression (g) suppresses, 5a1 cryptic exon inclusion (n = 3). h,i Site6 is essential to prevent aberrant A3SS usage, and PKmini^Δ6 (h) shows exclusive selection of the A3SS (i) (n = 3). j,k TDP-43 knockdown (j) enhances A3SS usage in PKmini^Δ6, while TDP-43 overexpression (k) suppresses A3SS and restores constitutive splicing (n = 3). l Schematic of combinatorial site-deletion constructs. m A3SS PSI values across deletion backgrounds showing site3 and site5 cooperate with site6 to repress aberrant splicing (n = 3). n,o EMSA (o) showing direct TDP-43 binding to probes (n) covering site1–3, site5, and site6 (n = 3). p Working model: TDP-43 binding to multiple UG-rich elements suppresses 5a1 insertion and A3SS usage, loss of TDP-43 relieves this repression. Source data are provided as a Source Data file.

To clarify the functional significance of these TDP-43 binding sites in regulating PKN1 cryptic splicing, we constructed a minigene containing exon 4, intron 5, and exon 5 of PKN1 (PKmini), and co-transfected with shTDP-43 or shCTR into HEK-293T cells. RT-PCR analysis showed that under shTDP-43 condition, in addition to the normal splicing product, the cryptic splicing product PKN1-5a1 was detected, whereas only normal splice form was produced in the shCTR group (Fig. 3c). This result not only validated the repressive role of TDP-43 in this cryptic splicing event, but also indicated that the PKmini system can be served as an important tool for analyzing downstream splicing regulatory mechanisms.

To further characterize TDP-43 regulatory sites, we predicted six conserved binding sites based on the CLIP-seq results (site1–site6), among which sites 1–5 are located upstream of the cryptic exon PKN1-5a1 and site6 downstream (Fig. 3a). Based on this structural feature, we prioritized constructing mutant minigenes lacking site1–5 (PKmini^D1–5) (Fig. 3d) or lacking site6 (PKmini^D6) (Fig. 3h). RT-PCR results showed that PKmini^D1–5 markedly induced the cryptic splicing product PKN1-5a1 (Fig. 3e), whereas PKmini^D6 primarily produced an A3SS product (Fig. 3i). Sanger sequencing further validated the splice junctions of these aberrant events (Supplementary Fig. 3a), indicating that deletion of distinct TDP-43 binding sites can regulate PKN1 splice selection through specific mechanisms. To further verify the functional role of TDP-43 in the above splicing regulation, we performed TDP-43 overexpression and knockdown on the background of PKmini^D1–5 or PKmini^D6, respectively. The results showed that TDP-43 overexpesssion significantly suppressed aberrant splicing in both mutants and enhanced normal splicing (Fig. 3g, k); conversely, TDP-43 knockdown further increased aberrant splicing (Fig. 3f, j). Notably, when querying human SNP databases (e.g., dbSNP), we found that although these variants are not associated with specific diseases, multiple SNPs annotated as nonpathogenic are located in the site6 region (Supplementary Fig. 3b). To further assess their functional impact, we constructed mutant minigenes containing the above SNPs, and these variants likewise induced aberrant splicing events in the minigene (Supplementary Fig. 3c), further supporting site6 as one of the key cis-acting elements regulating PKN1 splicing.

To clarify the cis-acting mechanism of the site1–6 regions in TDP-43–dependent splicing regulation, we constructed a mutant simultaneously lacking all six sites (PKmini^D1–6) (Fig. 3l). In this construct, the A3SS event could still occur, but the responsiveness to TDP-43 expression level was completely lost (Fig. 3m), i.e., neither overexpression nor knockdown of TDP-43 significantly altered its splicing pattern. By contrast, on the PKmini^D6 background lacking only site6, TDP-43 expression still effectively regulated A3SS splicing levels (Fig. 3j, k). Given that TDP-43 could still regulate A3SS splicing when only site6 was deleted, whereas this regulation was completely lost when sites 1–6 were deleted, these results suggest that TDP-43 regulation of A3SS splicing may depend on cis-acting elements in the site1–5 region, whereas site6 is more likely involved in selection of the A3SS acceptor site, thereby influencing the final splicing outcome. To further pinpoint key cis elements, we generated multiple combinatorial deletion mutants (e.g., PKmini^D1 + 6, PKmini^D2 + 6, etc.) (Fig. 3l) and systematically evaluated their effects on the A3SS event. PSI analysis showed that among all the combinations, deletion of site3 or site5 led to a more pronounced reduction in TDP-43’s ability to suppress aberrant splicing (Fig. 3m), suggesting that these two sites may be key cis elements mediating TDP-43 splicing regulation.

Finally, to determine whether TDP-43 directly binds specific sites on the pre-mRNA of PKN1, we designed biotin-labeled RNA probes covering the site1–site6 regions (Fig. 3n) and performed electrophoretic mobility shift assay (EMSA). We incubated the probes with recombinant TDP-43 protein. In native polyacrylamide gels, we observed distinct nucleic acid–protein complex bands for TDP-43 with site1, site2, site3, site5, and site6, indicating binding capability at these sites (Fig. 3o), whereas site4 showed no complex formation, suggesting no obvious affinity for this region. In addition, to further verify the binding specificity and affinity between TDP-43 and the candidate cis elements, we performed EMSA assay for each probe under different TDP-43 concentrations. We found that as TDP-43 concentration increased, the intensity of complex bands formed with the probes of site1, site2, site3, site5, and site6 were increased significantly in a typical dose-dependent manner (Supplementary Fig. 3d), indicating stable and specific binding. Conversely, site4 showed no complex band even at high TDP-43 concentrations, further indicating a lack of binding to TDP-43. Together, these results indicate that TDP-43 precisely regulates PKN1 splicing by directly binding multiple conserved sites on the pre-mRNA of PKN1, thereby suppressing production of cryptic exon 5a1 and maintaining the correct splicing of PKN1 mRNA (Fig. 3p).

The RRM1 domain and dimerization capacity jointly maintain the fidelity of TDP-43 regulation on PKN1 splicing

To delineate the structural basis by which TDP-43 regulates PKN1 cryptic splicing, we systematically evaluated the functional roles of its dimerization capacity and RNA-recognition domains in this process. Although previous studies have indicated that the dimeric state of TDP-43 influences its RNA splicing function, the mechanistic underpinnings in specific cryptic splicing events remain to be clarified^29,38. To test whether this structural basis controls PKN1 splicing, we overexpressed dimerization-impaired mutant, TDP-43^90–414 (90-414), and TDP-43^N6A (N6A), a dimerization-defective N-terminal mutant carrying six alanine substitutions^38–40 in HEK-293T cells. Cells were then treated with the cell-permeable dimer crosslinker disuccinimidyl glutarate (DSG) and analyzed by Western blots. We found that full-length TDP-43 (TDP-43^FL) formed stable dimers and higher-ordered oligomers, whereas 90–414 and N6A existed predominantly as monomers, with virtually no crosslinked bands (Fig. 4a), confirming markedly impaired dimerization for both mutants. Functionally, when 90–414 or N6A was co-transfected with the splicing reporter PKmini into HEK-293T cells without endogenous TDP-43 knockdown, neither mutant induced insertion of PKN1-5a1 (Fig. 4b). However, upon TDP-43 knockdown, both mutants failed to effectively suppress insertion of the cryptic exon PKN1-5a1 (Fig. 4c), whereas TDP-43^FL significantly suppressed this cryptic splicing event under the same conditions, indicating that loss of dimerization weakens the splicing regulatory function of TDP-43.

Fig. 4. TDP-43^ΔRRM1 impairs splicing repression by forming heterodimers with endogenous TDP-43.

a Western blots after DSG crosslinking showing TDP-43^FL forms dimers, while N6A and 90–414 mutants reduce dimerization; dimer/monomer ratio quantified (n = 3, two-sided t-test). b Co-expression of TDP-43 mutants with PKmini showing TDP-43^F147/149 L and ΔRRM1 induce 5a1 insertion; PSI quantified (n = 3, one-way ANOVA). c Under shTDP-43, TDP-43^FL, F227/229 L and ΔRRM2 restore 5a1 suppression, while F147/149 L, ΔRRM1, N6A and 90–414 fail; PSI shown (n = 3, one-way ANOVA). d Endogenous PKN1-5a1 splicing is activated by ΔRRM1, but not FL or ΔRRM2; 5a1/ACTB ratio quantified (n = 3, one-way ANOVA). e ΔRRM1 induces 5a1 insertion in PKmini, but dimerization-deficient ΔRRM1^N6A does not (n = 3, one-way ANOVA). f BiFC imaging shows that prominent nuclear puncta are formed when TDP-43-VN173 and ΔRRM1-VC155 are co-expressed together with ΔRRM1, whereas no such puncta are observed when TDP-43-VN173 and TDP-43-VC155 are co-expressed with ΔRRM1. Scale bar, 10 µm. g Quantification of BiFC puncta per nucleus (n = 20 cells/group; two-sided Wilcoxon rank-sum test). h Working schematic: ΔRRM1 preferentially recruits TDP-43/ΔRRM1 heterodimers rather than TDP-43/TDP-43 homodimers. Source data are provided as a Source Data file.

TDP-43 contains two conserved RNA recognition motifs (RRMs), among which RRM1 is considered central to RNA binding and splicing regulation⁴¹. To validate the role of RRMs in PKN1 splicing, we co-transfected TDP-43^FL, loss-of-function mutant of RRM1 (F147/149 L) or RRM2 (F227/229 L), deletion mutant of RRM1 (ΔRRM1) or RRM2 (ΔRRM2) with PKmini into HEK-293T cells, and analyzed by RT-PCR. Interestingly, we found that F147/149 L and ΔRRM1 both markedly induced aberrant insertion of PKN1-5a1, with the effects resembling TDP-43 knockdown, particularly pronounced for ΔRRM1 (Fig. 4b), suggesting a dominant-negative effect of RRM1 deletion. Similarly, upon a TDP-43 knockdown background (Supplementary Fig. 4a), co-transfection of these mutants with PKmini into HEK-293T cells revealed that TF147/149 L and ΔRRM1 failed to suppress PKN1-5a1 insertion, consistent with the loss-of-function phenotype (Fig. 4c). Moreover, ΔRRM1 not only failed to suppress the cryptic event but further enhanced PKN1-5a1 insertion. Notably, even without PKmini transfection, overexpression of ΔRRM1 in HEK-293T cells significantly induced insertion of 5a1 in the endogenous PKN1 transcript (Fig. 4d), further indicating that the ΔRRM1 mutant interferes with normal regulation of endogenous splice sites through a dominant-negative mechanism. In addition, we observed that F147/149 L and ΔRRM1 formed aberrant nuclear condensates and colocalized with TDP-43^FL in HEK-293T cells, as previously describe^42–44, whereas other constructs (e.g., TDP-43^FL, F227/229 L, or ΔRRM2) did not form nuclear condensates (Supplementary Fig. 4b).

Previous studies have shown that dominant-negative mutants can heterodimerize with the wild-type counterparts, thereby impairing normal function⁴⁵. Together with the prior finding that the N terminus of TDP-43 has dimerization capacity^38–40, we further hypothesized that ΔRRM1 might exert dominant-negative effects by forming heterodimers with endogenous wild-type TDP-43, thereby disrupting recognition and binding of RNA targets. To test this hypothesis, we generated a double mutant lacking both dimerization capacity and RRM1 (ΔRRM1^N6A), and co-transfected with PKmini into HEK-293T cells, then analyzed by RT-PCR. We found that, unlike ΔRRM1—which markedly induced PKN1-5a1 insertion—ΔRRM1^N6A did not elicit obvious cryptic exon insertion, indicating a lack of dominant-negative activity (Fig. 4e). To further determine whether these mutants retain the ability to associate with wild-type TDP-43, we employed the bimolecular fluorescence complementation (BiFC) system. TDP-43 and each mutant were fused to Venus N-terminal fragment 173 (VN173) or Venus C-terminal fragment 155 (VC155) fragments of the BiFC system, and co-expressed in HEK-293T cells. If dimerization occurs, the VN and VC fragments will produce a GFP signal. We found that TDP-43-VN173 and TDP-43-VC155 generated strong nuclear GFP signals, confirming homodimerization; ΔRRM1-VC155 also formed heterodimers with TDP-43-VN173; by contrast, N6A-VC155 and ΔRRM1^N6A-VC155 failed to induce GFP signals, indicating complete loss of dimerization capacity and further indicating that RRM1 forms heterodimers with wild-type TDP-43 via the N terminus (Supplementary Fig. 4c).

We had previously observed that ΔRRM1 formed pronounced nuclear condensates and colocalized with wild-type TDP-43 (WT), and, together with the ability of ΔRRM1 to form heterodimers with WT, we hypothesized that ΔRRM1 might recruit WT into the aberrant nuclear condensates formed via dimerization. To test this, we designed two key conditions: (1) whether ΔRRM1 condensates recruit TDP-43-VN173/ΔRRM1-VC155 heterodimers efficiently; and (2) whether these condensates recruit TDP-43-VN173/TDP-43-VC155 homodimers. We co-transfected TDP-43-VN173 with ΔRRM1-VC155 or TDP-43-VC155, and ΔRRM1 mutant into HEK-293T cells. Fluorescence imaging of living cells showed robust BiFC punctate signals in the TDP-43-VN173/ΔRRM1-VC155 and ΔRRM1 group, whereas only weak BiFC punctate signals were observed in the TDP-43-VN173/TDP-43-VC155 and ΔRRM1 group (Fig. 4f). Given that the fluorescences in BiFC system are arised from stabilized protein–protein interactions, these results suggest that ΔRRM1 condensates preferentially recruit TDP-43/ΔRRM1 heterodimers rather than TDP-43 homodimers (Fig. 4g). To determine whether ΔRRM1 is specific for this BiFC punctate signals, we performed parallel control experiments by co-expressing wild-type TDP-43 and TDP-43-VN173 with either ΔRRM1-VC155 or TDP-43-VC155; no similar nuclear puncta were observed (Supplementary Fig. 4d), further supporting ΔRRM1 as the key driver of the aberrant aggregation of TDP-43. Whether such condensates functionally “sequester” wild-type TDP-43 and thereby impair its binding to PKN1 pre-mRNA remains to be further validated using the MS2–MCP system⁴⁶.

Taken together, these results support our hypothesis that TDP-43^ΔRRM1 can form heterodimers with endogenous TDP-43 via its retained N-terminal region, thereby disrupting normal recognition and binding to RNA targets and inducing aberrant splicing through a dominant-negative mechanism, whereas N-terminal dimerization–deficient mutations effectively block this process (Fig. 4h).

TDP-43 knockdown generates and accumulates the cryptic peptide PKN1-N207

Previously, we knocked down TDP-43 in HEK-293T cells expressing PK-intron, and observed a ~ 25–33 kDa FLAG band (Fig. 1n), suggesting that an aberrant transcript arising from a cryptic splicing event might be translated into a truncated cryptic peptide. Interestingly, this splicing product contains a PTC (Supplementary Fig. 1a) and should be subject to NMD⁴⁷. However, the stable presence of this band implies that it may partially escape NMD.

To test this hypothesis, we treated TDP-43–knockdown HEK-293T cells with the translation inhibitor CHX (2 μg/mL) to indirectly inhibit NMD activity and performed to qRT-PCR analysis. We found that PKN1-5a1 expression was increased ~1.4-fold in the CHX group, compared with the DMSO group (Supplementary Fig. 5a). Furthermore, we knocked down up-frameshift protein 1 (UPF1), a key effector of NMD⁴⁸ (Fig.5a), to specifically block this pathway. We found that UPF1 knockdown alone did not induce aberrant splicing of PKN1-5a1, but in the context of TDP-43 deficiency, combined UPF1 knockdown further upregulated its expression by ~1.6-fold (Fig. 5a), supporting that PKN1-5a1 transcript can be partially recognized and degraded by NMD. Notably, either by CHX treatment or UPF1 knockdown, the magnitude of PKN1-5a1 upregulation was limited; together with our previous observation of the “FLAG band”. These suggest that PKN1-5a1 may not be a typical NMD target and may escape degradation by being translated into a truncated peptide. This, of course, required validation at the endogenous level. Therefore, we generated stable TDP-43–knockdown HEK-293T and M17 cell lines. Specifically, cells were infected with lenti/shTDP-43, treated with puromycin (2 μg/mL) for 4 days, and then cultured without antibiotics for an additional 6 days. It should be noted that TDP-43 knockdown markedly suppressed the proliferative capacity of HEK-293T and M17 cells, making long-term stable passaging difficult. To assess proliferation, we performed an EdU incorporation assay on HEK-293T cells at day 10 post-infection. We found a clear decrease in cell density and a significant reduction in EdU-positive cells in the shTDP-43 group (Supplementary Fig. 5b), indicating impaired proliferation, consistent with previous studies⁴⁹. We then collected the selected HEK-293T and M17 cells for Western blots analysis, using a specific antibody against the N terminus of PKN1 to monitor the expression of full-length PKN1 and the truncated peptide. We found that TDP-43 knockdown led to a marked decrease in full-length PKN1, accompanied by a pronounced increase in a truncated peptide within the 25 ~ 33 kDa range (Fig. 5b and Supplementary Fig. 5c). Based on the splice junctions and reading frame, this band corresponds to the first 207 amino acids of PKN1; we therefore designated the cryptic peptide PKN207. To further analyze its accumulation at different time points, we set a time course of TDP-43 knockdown at 48, 72, 96, and 120 hours, and performed to Western blots. We found that full-length PKN1 was gradually decreased from 48 hours onward, but the PKN207 band was clearly detected only after 120 hours of sustained TDP-43 deficiency (Fig. 5c), suggesting that this truncated peptide must gradually accumulate under continuous loss of TDP-43 to reach a detectable level.

Fig. 5. TDP-43 knockdown leads to production of a stable cryptic peptide, PKN207.

a TDP-43 and UPF1 knockdown in HEK-293T cells; TDP-43 depletion upregulates PKN1-5a1, enhanced by combined UPF1 knockdown (n = 3, one-way ANOVA). b Western blots of stable shCTR and shTDP-43 cell lines showing reduced PKFL and a ~ 33 kDa band (red arrow) matching predicted PKN207 (n = 3). c Western blots shows PKN207 appears from 120 h post shTDP-43 transfection (n = 3, one-way ANOVA). d CHX chase assay showing PKN207 more stable than PKFL after tetracycline induction (n = 3; slope comparison by ANCOVA). e shPKN1 abolishes the ~33 kDa band in shTDP-43 cells, confirming its origin from PKN1, quantification shown (n = 3, one-way ANOVA). f Four PKN207 antibodies specifically recognize FLAG-PKN1^N207 without cross-reacting with PKN1^FL (n = 3). g All four antibodies detect endogenous ~33 kDa PKN207 in shTDP-43 cells; E29426 selected for subsequent use (n = 3). h Western blots showing PKN207 in the hippocampus of AD with pTDP-43 (AD1, AD5). i PKFL protein levels do not differ between Normal and AD groups (n = 5, two-sided t-test). j Paired-point plot showing elevated PKN207 in individuals with pTDP-43 pathology. k RT-PCR showing increased PKN1-5a1/PKN1 ratio in AD hippocampus (Normal n = 5, AD n = 6; two-sided Wilcoxon test). Source data are provided as a Source Data file.

To further evaluate the stability of this cryptic peptide PKN207 in HEK-293T cells, we constructed a Tet-on inducible expression system to drive expression of full-length PKN1 (Tet-PKFL) and the N207 cryptic peptide (Tet-PKN207), respectively (Fig. 5d). After induction with doxycycline (Dox, 1 μg/mL) for 24 hours, the medium was replaced with Dox-free medium, then treated with CHX (2 μg/mL); cells were collected at 0, 4, 8, and 12 hours for immunoblotting. We found that with prolonged CHX treatment, full-length PKN1 levels were gradually declined, while PKN207 levels remained stable throughout (Fig. 5d), indicating that PKN207 cryptic peptide has high intracellular stability and may continuously accumulate in HEK-293T cells (Fig. 4g). Furthermore, we simultaneously knocked down TDP-43 and PKN1 in HEK-293T cells to examine changes in PKN207 expression. We observed that TDP-43 knockdown markedly enhanced PKN207 expression, whereas concurrent PKN1 knockdown clearly reduced the truncated peptide (Fig. 5e), further demonstrating that it originates from aberrant splicing of the PKN1 gene. These results indicate that the TDP-43–knockdown–induced cryptic splicing of PKN1 can, after prolonged accumulation, generate and express the cryptic peptide PKN207, and that this peptide exhibits high stability, suggesting that it may persistently exert functions in cells.

A specific antibody reveals that the PKN207 cryptic peptide is detectable in AD brain with TDP-43 pathology

Given that the PKN1-5a1 cryptic exon has been validated in RNA-seq data of ALS cohort from NYGC, we further hypothesized that whether this cryptic splicing event truly occurs in patients, its translation product—the PKN207 cryptic peptide—might also be detectable in patients with TDP-43 pathology. To test this hypothesis, we attempted to generate antibodies capable of specifically recognizing PKN207.

Sequence alignment showed that the C terminus of PKN207 protein contains a unique amino acid sequence “LAL”, which replaces the “SPD” sequence present in full-length PKN1, resulting in a distinct C-terminal structure (Supplementary Fig. 5d). To maximize antibody specificity, we selected an eight–amino acid peptide, “DDTQGLAL”, spanning this differential region, as the immunogen. Structural predictions indicated that this region is surface-exposed and exhibits favorable antigenicity, hydrophilicity, and immunogenicity (Supplementary Fig. 5e). On this basis, we immunized rabbits intraperitoneally with the peptide and prepared four polyclonal anti-PKN207 antibodies (E29426–E29429) (Supplementary Fig. 5f).

To validate the specificity of the antibodies for PKN207, we overexpressed either full-length PKN1 (PKFL) or the PKN207 cryptic peptide in HEK-293T cells and performed to Western blots analysis. All four antibodies showed a specific band of approximately 27–33 kDa in the PKN207 group, while no signal in the PKFL group, indicating that these antibodies do not cross-react with full-length PKN1 (Fig. 5f). We then further validated their recognition in TDP-43–knockdown or control HEK-293T cells. All four antibodies showed a band of approximately 27 kDa in the shTDP-43 group, no obvious signal in the control group (Fig. 5g); among them, E29426 exhibited the lowest background and best specificity (Fig. 5g). These results indicate that this antibody series specifically recognize PKN207 but not PKFL. To further assess the cellular localization of PKN207, we overexpressed PKN1^FL or PKN1^N207 in HEK-293T cells, and performed to nuclear–cytoplasmic fractionation and Western blots. We found that PKN1 localized mainly in the cytoplasm with a portion in the nucleus, whereas PKN207 localized exclusively in the cytoplasm (Supplementary Fig. 5g). We next evaluated the applicability of these antibodies in immunofluorescence staining. We transfected PKN207-EGFP into HEK-293T cells and performed immunostaining with the four antibodies (red fluorescence). We observed strong diffuse cytoplasmic red signals regardless of EGFP transfection (Supplementary Fig. 5h–k), indicating notable nonspecific background of these polyclonal antibodies in immunofluorescence.

We ultimately selected the E29426 antibody to determine whether it could be used to detect PKN207 expression in patients with TDP-43 pathology. Given that ALS brain tissue was unavailable and TDP-43 pathology is present in ~20–50% of AD patients^7,50,51, we examined hippocampal tissue from TDP-43-positive AD cases. First, we assessed the presence of phosphorylated TDP-43 (pTDP-43) in the RIPA-insoluble fraction of AD samples by Western blots. We found clear pTDP-43 signals in AD1 and AD5, indicating typical TDP-43 proteinopathy (Fig. 5h). We then examined PKN207 expression in the AD samples. Although there was no significant difference in PKFL between the AD and control groups (Fig. 5i), we detected a distinct ~25–33 kDa band in AD1 and AD5 using the E29426 antibody, whereas no such band was observed in the control group or in TDP-43-negative AD cases (Fig. 5h, j). Further RT-PCR analysis also showed that the PKN1-5a1 cryptic exon was detected in AD1, AD2, and AD5 (Fig. 5k).

Notably, PKN207 expression mirrored pTDP-43 pathology, being present in AD1 and AD5 (which showed marked aggregation) but absent in AD2 and controls (which lacked corresponding pathology) (Fig. 5j). In contrast, the PKN1-5a1 cryptic exon transcript was detected at low levels in AD2 as well (Fig. 5k), yet no corresponding PKN207 protein was detected, demonstrating that cryptic splicing does not always produce a stable protein. We speculate that even in the absence of classic aggregated pTDP-43, functional abnormalities of TDP-43 (e.g., mild nuclear loss or weakened splicing regulatory capacity) may still occur in AD2, thereby inducing activation of specific cryptic splicing. This splicing abnormality may represent as an early molecular feature of TDP-43 loss of function that has not yet progressed to visible aggregate pathology. In addition, the splicing product may not yet have accumulated to a detectable protein level in vivo, leading to failure to detect the corresponding PKN207 band. In summary, compared with the splicing event itself, the expression of PKN207 cryptic peptide may more directly reflect TDP-43 pathology, and the E29426 antibody holds promise as a more specific biomarker for TDP-43 proteinopathies.

Expression of the PKN207 cryptic peptide induces cognitive impairment in mice

We detected the expression of PKN207 cryptic peptide in TDP-43-positive AD brains, whereas PKFL showed no obvious change (Fig. 5h), suggesting that PKN207 may directly mediate neuronal dysfunction without altering full-length protein expression, and may even participate in the development of cognitive impairment. Given the limited number of AD samples in the immunoblot analysis, we further analyzed RNA-seq data from the parahippocampal gyrus (PHG) in the Mount Sinai Brain Bank (MSBB) AD transcriptome database to expand the sample size and validate the finding at the transcriptional level. We found that the PKN1-5a1 cryptic exon was stably detected in AD patients, and nearly absent in healthy controls (Fig. 6a), supporting its activation in disease and potential pathogenicity. Meanwhile, full-length PKN1 mRNA expression was slightly increased, which may explain the lack of a significant change at the protein level. Notably, activation of this cryptic splicing event was observed even at Braak stage 1, an early stage of AD, suggesting that it may be induced at the early stage of the disease process.

Fig. 6. The PKN207 cryptic peptide induces neurotoxicity and impairs synaptic function.

a Boxplot of human parahippocampal transcriptome data from MSBB showing activated PKN1-5a1 cryptic splicing in AD samples across Braak stages. Box plots show median (center line), the upper and lower quantiles (box), and the range of the data (whiskers). b Schematic of bilateral hippocampal AAV injection in C57BL/6J mice and subsequent behavioral testing. c Morris water maze showing prolonged escape latency in AAV-PKFL and AAV-PKN207 mice (AAV-CTR n = 9, AAV-PKFL n = 10, AAV-PKN207 n = 10, repeated-measures two-way ANOVA). d CSF NEFL levels are elevated in AAV-PKN207 mice (AAV-CTR n = 10, AAV-PKFL n = 20, AAV-PKN207 n = 20; one-way ANOVA). e Primary hippocampal neurons expressing PKFL or PKN207 showing increased LDH release (n = 4, one-way ANOVA). f KEGG enrichment of DEPs highlights pathways linked to synaptic function and neurodegeneration. g Heatmap showing downregulation of LTP-related genes in PKN207-expressing mice. h STRING network of NEFL-interacting proteins (red: upregulated; blue: downregulated). i Western blots of neurofilament proteins in N2a cells overexpressing PKFL or PKN207. j Hippocampal LTP is impaired in AAV-PKN207 mice; fEPSP slopes are shown (5 mice, 10 slices; one-way ANOVA). Source data are provided as a Source Data file.

To test the potential neurotoxicity of PKN207, we injected three neuron-specific promoter adeno-associated viruses (AAV-hSyn) into the hippocampus bilaterally of 2-month-old C57BL/6J mice: AAV-PKFL (encoding full-length PKN1), AAV-PKN207 (encoding the N207 cryptic peptide), and an empty control virus AAV-CTR (Fig. 6b). We also validated the expression of AAV-PKFL and AAV-PKN207 in the mouse hippocampus by Western blots, and confirmed robust expression of both PKFL and the PKN207 peptide (Supplementary Fig. 6e). Four months later, Morris water maze (MWM) testing showed significant cognitive deficits in both the AAV-PKFL and AAV-PKN207 groups (Fig. 6c), indicating that overexpression of either protein may adversely affect learning and memory.

Considering that cognitive impairment is often accompanied by structural damage to neuronal axons, we further measured NEFL level in CSF 4 months post virus injection. CSF-NEFL was significantly higher in the PKFL and N207 groups than in their respective control groups (Fig. 6d), suggesting compromised axonal integrity accompanied by increased neurofilament release. This change in a fluid biomarker reflects structural pathological alterations at the central nervous system level^52–55.

To further verify whether PKN207 is cytotoxic, we overexpressed PKFL or PKN207 in primary mouse hippocampal neurons and assessed toxicity by lactate dehydrogenase (LDH) release. Cytotoxicity was significantly induced in both groups, with PKFL showing slightly higher toxicity than PKN207 (Fig. 6e). Nevertheless, as a splicing product specific to TDP-43 loss, PKN207 has a unique pathological context, suggesting that it may possess disease-state–specific pathogenic mechanisms.

To systematically evaluate molecular effects of PKFL and PKN207, we performed quantitative proteomics on hippocampus of mice. Principal component analysis (PCA) and Pearson correlation analysis indicated that the protein expression profiles of PKFL and PKN207 were highly similar and clearly distinct from controls (Supplementary Fig. 6a, b). Differential expression analysis revealed only 214 significantly different proteins between the PKN207 and PKFL groups, indicating overall convergence and high concordance of their expression profiles (Supplementary Fig. 6c); this trend was also visualized in the volcano plot (Supplementary Fig. 6d). KEGG enrichment analysis showed that differential proteins in PKFL and PKN207 were significantly enriched in ALS- and AD-related pathways, as well as in pathways closely associated with synaptic function, such as LTP (Fig. 6f). Further examination of core factors in the LTP pathway revealed widespread downregulation of genes, such as Camk2d, Ppp1cc, Plcb4, Prkca, Rps6ka2, Itpr1, Rapgef3, and Kras, with only a few genes—such as Plcb1, Mapk3, and Prkcb—showing upregulation (Fig. 6g), reflecting an overall suppression of LTP signaling that may impair synaptic plasticity and learning/memory. These inhibitory effects were more pronounced in the PKN207 group, with more consistent within-group changes, suggesting a stronger suppressive effect on the LTP pathway.

Furthermore, given that PKN1 is a known regulator of neurofilament protein phosphorylation^25,26, we examined its impact on neuronal structural stability. Constructing a protein–protein interaction (PPI) network centered on NEFL, we found that multiple proteins related to axonal transport and neurofilament structure (e.g., Kif5a, Klc1/2/3, Sod1, Nefl, and Nefm) were downregulated in the contexts of PKN1 and PKN207 expression, suggesting a global perturbation of the neurofilament system. Meanwhile, the expression of NEFL and NEFM was significantly increased (Fig. 6h, i), potentially indicative of neurofilament aggregation or axonal transport defects. These changes were more pronounced in the PKN207 group, suggesting a stronger pathogenic potential in neuronal structural damage. Although PKN207 lacks the kinase domain, its expression still disrupted the neurofilament system, implying that it may interact with other proteins via retained N-terminal structures and affect cytoskeletal stability through a non–kinase-dependent mechanism. Together with the highly similar proteomic profiles observed for PKFL and PKN207 (Supplementary Fig. 6a-d), these findings further support that key functions of PKN1 may be primarily mediated by its N-terminal region.

We next recorded field excitatory postsynaptic potentials (fEPSPs) in the CA1 region of acute brain slices using ex vivo multichannel electrophysiology 4 months after virus injection to functionally verify changes in synaptic plasticity. The control group exhibited a typical LTP response after high-frequency stimulation (HFS), with a significant increase in fEPSP slope that was maintained after stimulation (Fig. 6j). By contrast, the PKFL and PKN207 groups showed markedly attenuated fEPSP responses with limited LTP potentiation, and the PKN207 group showed little sustained enhancement (Fig. 6j). Quantification at the endpoint (55–60 minutes post-stimulation) further confirmed a stronger inhibitory effect of PKN207 on LTP.

Although PKN207 and PKFL exhibited highly similar overall protein expression and functional phenotypes, as a cryptic peptide lacking PKN1 functional domains, PKN207 may still possess unique regulatory mechanisms. Accordingly, we focused on proteins differentially expressed specifically in PKN207 relative to PKFL for GO enrichment analysis. PKN207-specific proteins were enriched in pathways related to the cytoskeleton, regulation of ion homeostasis, and signal recognition, particularly involving intermediate filaments and the Wnt signaling module (Supplementary Fig. 6f–h), suggesting that it may perturb neuronal homeostasis through structural interference.

Discussion

The present study systematically identifies a TDP-43–loss–induced cryptic splicing event in PKN1 (PKN1-5a1). This cryptic exon is markedly upregulated in ALS brains and AD harboring TDP-43 pathology (Figs. 2a, b and 6a), and is translated into a stably expressed truncated cryptic peptide, PKN207 (Fig. 5b). Functional validation demonstrates that aberrant PKN207 expression elicits pronounced neurotoxicity—including cognitive impairment, deficits in synaptic plasticity, and disruption of the neurofilament system (Fig. 6b-j), suggesting that PKN207 may play an important role in the pathogenesis of TDP-43 proteinopathies.

Extensive prior works have shown that loss of TDP-43 function activates cryptic splicing in key neuronal genes (e.g., STMN2 and UNC13A)^{11–14,16,29,56}, leading to impaired axonal regeneration and synaptic dysfunction and forming part of the molecular pathology of ALS and related disorders. However, most reported TDP-43–dependent cryptic splicing events primarily introduce premature termination codons that trigger NMD^1,13, without clearly implicating stably expressed, functional peptide products. Our findings extend this field by showing that, after evading NMD, a cryptic splicing event can generate a stably expressed truncated toxic peptide, thereby proposing a distinct pathogenic mode in which loss of TDP-43 not only drives transcript degradation but can also, through cryptic splicing, produce aberrant peptides that perturb neuronal homeostasis.

Notably, PKN1 is a key serine/threonine kinase broadly involved in neurofilament phosphorylation and axonal structural stability^24–26. Prior studies indicate that, under cellular stress or apoptotic stimuli, PKN1 can be cleaved by caspase-3 to produce truncated forms with altered kinase activity that may influence neuronal fate and potentially exert toxicity²⁸. This context provides theoretical support for the neurotoxic potential of the PKN207 cryptic peptide identified here.

Unlike the de novo proteins reported to arise from cryptic splicing in HDGFL2 and MYO18A^22,23, the PKN1 cryptic splicing product PKN207 lacks the critical C-terminal kinase domain (Fig. 5d), implying functional properties that likely differ from canonical PKN1 kinase activity. Nevertheless, proteomic analyses show that, despite lacking the kinase domain, PKN207 elicits signaling pathway changes highly similar to those induced by full-length PKN1 and exerts even more pronounced effects on neurofilament architecture in some assays (Fig. 6b–i). These observations suggest that core biological functions of PKN1 may be primarily mediated by its N-terminal domains, and that loss of the kinase domain does not necessarily abolish function but may confer additional, kinase-independent toxicity that perturbs neuronal signaling networks. Such toxicity from truncated peptides could represent a common yet underappreciated mechanism in TDP-43 proteinopathies.

Of note, although the PKN1-5a1 cryptic exon is highly conserved in humans and other primates, the corresponding splicing activity is absent in mouse models (Supplementary Fig. 2a). Even so, TDP-43 knockdown in mice is associated with reduced PKN1 protein, an effect that may reflect accelerated protein degradation rather than direct splicing regulation (Supplementary Fig. 2c-f). These findings highlight marked species specificity in TDP-43–mediated regulation of PKN1 and underscore the value of humanized or primate models for validating cryptic splicing events.

But our study still has several limitations. First, the ALS RNA-seq datasets used here lack annotations of TDP-43 pathological status (Fig. 2a, b), precluding direct sample-level validation of the relationship between TDP-43 pathology and PKN1-5a1 cryptic splicing; we should integrate detailed clinicopathological information to strengthen this link in the future work. Second, the number of AD samples used to verify PKN207 expression is limited (Fig. 5h), constraining generalization of PKN207 as a broadly applicable biomarker; larger cohorts will be needed to improve robustness. Third, although the PKN207-specific antibody performs well by Western blots, minimal sequence divergence leads to suboptimal specificity in immunofluorescence staining, limiting precise cellular and subcellular localization (Fig. 5f–h and Supplementary Fig. 5h–k). Finally, while PKN207 and full-length PKN1 exhibit similar neurotoxic effects (Fig. 6c–j), the unique mechanisms of PKN207 or specific targets relative to PKN1 remain undefined; further work is required to delineate N207’s distinct molecular network and disease-relevant effectors.

In summary, our work defines a TDP-43–loss–induced cryptic splicing event and proposes a pathogenic model of “splicing dysregulation → aberrant translation → functional interference”, providing additional mechanistic insight into how toxicity mediated by a cryptic splicing product may contribute to the pathology of neurodegenerative diseases. These results expand current understandings of mechanisms underlying TDP-43 proteinopathies, and offer a conceptual basis and candidate molecular targets for diagnostic development and targeted therapeutic strategies.

Methods

Ethical statement

This study was conducted in accordance with the ethical principles of the Declaration of Helsinki. The use of human post-mortem brain tissue and associated clinicopathological data was approved by the Ethics Committee of Tongji Medical College, Huazhong University of Science and Technology (Approval No. [2019] IEC(S775)). All tissue samples were obtained from the National Health and Human Brain Tissue Resource Center (China). Informed consent for brain donation and the use of tissue for research purposes was obtained from all donors ante-mortem. The brain bank operates under strict ethical guidelines to ensure donor anonymity and confidentiality.

All animal procedures were approved by the Animal Ethics Committee of Tongji Medical College, Huazhong University of Science and Technology (IACUC 4696). The study was performed in accordance with the relevant guidelines and regulations.

Human brain tissue

Human brain autopsy specimens were obtained from the National Health and Disease Human Brain Tissue Resource Center (Table 1). Each specimen carried a unique Brain Bank identifier, with donor ID, sampled brain region, sex, age at death, brain wet weight, and Braak neuropathological stage recorded. Clinical and neuropathological diagnoses were made according to established criteria, referencing the NIA–AA guidelines for the neuropathologic assessment of AD⁵⁷, Braak staging followed the original method⁵⁸. Inclusion criteria were based on ante-mortem clinical presentation and postmortem histological evaluation. The primary clinical diagnosis among included cases was AD. Detailed donor information is provided in the accompanying table (including Brain Bank ID, sampled brain region, age at death, sex, Braak stage, and brain wet weight). Normal group refers to a healthy population with no history of neurological diseases.

Table 1.

Hippocampus of Alzheimer’s disease (AD) and control (N) cases used in this study

Diagnosis	Case #	Age at death (y)	Gender	Braak Stage*
Normal	1	83	M	0
Normal	2	74	M	0
Normal	3	70	F	0
Normal	4	98	F	0
Normal	5	72	M	0
		79.4 ± 11.5
AD	1	83	M	Ⅳ
AD	2	99	F	Ⅴ
AD	3	73	M	Ⅲ
AD	4	79	M	Ⅵ
AD	5	71	F	Ⅳ
AD	6	88	M	Ⅵ
		82.2 ± 10.4

^*Neurofibrillary pathology was staged according to Braak and Braak⁵⁸.

Plasmid Information and Construction

Detailed information of the plasmids used in this study and their construction procedures is provided in Supplementary Table 2.

Animals and stereotaxic injection

Two-month-old C57BL/6J mice were randomly assigned to three groups: AAV-Con (n = 15), AAV-PKN1 (n = 25), and AAV-PKN207 (n = 25). Operators were blinded to group allocation. AAV9 vectors (SYN promoter; working titer 1×10¹² vg/mL) were provided by Obio Co., Ltd (Shanghai). Mice were anesthetized with isoflurane (induction 3–4%, maintenance 1.5–2.0%; RWD R510-22-10) and positioned in a stereotaxic frame (RWD 69100). Using bregma as the reference, bilateral hippocampal injections were made at AP − 2.0 mm, ML ± 1.5 mm, and DV − 1.8 mm. A Hamilton syringe (LFGJ-80383) connected to a microinjection pump was used to deliver virus at 100 nL/min; 0.5 μL was infused per side ( ≈ 5 × 10⁸ vg), for a total dose of ≈1 × 10⁹ vg per mouse. The needle was left in place for 5 min after infusion and then withdrawn slowly to minimize reflux. A heating pad was used intraoperatively; postoperative analgesia was provided with meloxicam (2 mg/kg, subcutaneous, once daily for 2 days) and recovery was monitored. Mice were housed at 22 ± 2 °C with 40–60% humidity under a 12-h light/dark cycle, with food and water available ad libitum. Surgical and stereotaxic procedures followed established methods, and injection coordinates were referenced to Paxinos & Franklin, The Mouse Brain in Stereotaxic Coordinates, 4th ed.

Behavioral testing—Morris water maze (MWM)

At 4 months after viral injection, the MWM was used to assess spatial learning and memory. The apparatus consisted of a circular pool (diameter 120 cm) filled with opaque water maintained at 22 ± 1 °C. A hidden platform (diameter 10 cm) was fixed in the same quadrant and submerged ~1 cm below the water surface. Training lasted 4 days with 4 trials per day, with mice released from different quadrants; the maximum swim time per trial was 90 s. The escape latency was recorded upon reaching the platform; trials in which the platform was not reached were assigned 90 s. After each trial, mice were allowed to remain on the platform for 15 s. Behavior was automatically recorded and analyzed using the EthoVision XT video-tracking system (Noldus). Learning ability was evaluated by the mean escape latency during the training period.

Cisterna magna CSF collection and NEFL measurement

Mice received AAV injection at 2 months of age, and CSF was collected 4 months later for analysis. Groups were: AAV-CTR (n = 10), AAV-PKFL (n = 20), and AAV-PKN207 (n = 20). Procedures were performed under aseptic conditions with continuous isoflurane anesthesia (1–2%). With the mouse placed prone and the neck flexed to expose the cisterna magna, the area was disinfected with 75% ethanol; a midline skin incision was made, muscles were separated, and the transparent dura mater was exposed. Under a stereomicroscope, a pulled glass capillary (outer diameter ~0.5 mm) was inserted vertically into the cisterna magna, and CSF was slowly aspirated; approximately 5–15 μL of clear fluid was obtained per mouse. Blood contamination was strictly avoided; any blood-tinged sample was discarded. CSF was immediately transferred to RNase-free microcentrifuge tubes, kept on ice, and stored at −80 °C. NEFL levels were measured using a mouse NEFL ELISA kit (Jonln, JL51267) according to the manufacturer’s instructions.

LDH release assay in mouse hippocampal neurons

To evaluate the effects of differential PKN1 expression on cell membrane integrity, LDH levels in the culture supernatant of primary hippocampal neurons were measured using an LDH Cytotoxicity Assay Kit (Beyotime, C0017). Three groups were included: control virus (Null), PKN1 full-length lentivirus (PKFL), and N207 truncated lentivirus (PKN207). Cells were seeded at equal density in 24-well plates and, after stable adherence, subjected to the indicated treatments. Supernatants were collected 48 h later. According to the manufacturer’s instructions, supernatants were mixed with the working reagent and incubated at room temperature for 30 min protected from light; after addition of the stop solution, absorbance was read at 490 nm (OD₄₉₀).

Cytotoxicity was assessed by comparing LDH release in each experimental group relative to the control group. Four technical replicates were included per group, and background values from blank wells were subtracted prior to analysis.

fEPSP recordings

Acute hippocampal slice electrophysiology was performed on 6-month-old C57BL/6J mice at 4 months after AAV injection. Mice were anesthetized with isoflurane and rapidly decapitated; the brain was removed, rinsed with ice-cold artificial CSF (aCSF), and mounted on a vibratome. Coronal cortico-hippocampal slices (350 μm) were prepared in aCSF at 4 °C equilibrated with 95% O₂/5% CO₂, then transferred to room-temperature aCSF (20.5 ± 1 °C) to recover for ≥1 h. Recordings were completed within 8 h of tissue collection. The aCSF contained: 118 mM NaCl, 10 mM glucose, 20 mM sucrose, 25 mM NaHCO₃, 3.0 mM KCl, 1.5 mM CaCl₂, 1.0 mM NaH₂PO₄, and 1.0 mM MgCl₂; osmolality 305–315 mOsm, pH 7.42 ± 0.02.

Recordings were performed under continuous superfusion with aCSF (31.0 ± 2 °C, 95% O₂/5% CO₂). A bipolar stimulating electrode (CBARC75, FHC Inc.) was placed in the Schaffer collateral pathway, and a glass recording electrode (1–2 MΩ) was positioned in the CA1 stratum radiatum. Stimuli (100–400 μA, 0.1–0.2 ms) were delivered at the lowest intensity that evoked 50% of the maximal fEPSP amplitude. After establishing a stable baseline for ≥10 min, long-term potentiation (LTP) was induced with theta-burst stimulation: 4 trains, each comprising 10 bursts at 5 Hz, with 4 pulses per burst at 100 Hz. Post-tetanus responses were recorded for 60 min, and fEPSP slopes were normalized to baseline. Signals were acquired using an AM Systems differential amplifier.

RNA sequencing

M17 cells were infected with lentiviruses (Obio Co., Ltd., Shanghai, China): control (Lenti-shCTR) and a TARDBP–targeting knockdown virus (Lenti-shTDP-43). 96 h later, all samples were lysed in TRIzol (Takara, 9109), immediately stored at −80 °C, and subsequently submitted to BGI Genomics (Shenzhen, China) for RNA sequencing. Each group included three biological replicates. Libraries were prepared using poly(A)+ mRNA enrichment and sequenced with paired-end 150-bp reads (PE150), yielding approximately 10 Gb per sample. Reads were aligned to the NCBI human reference genome (hg38) and used for downstream gene-expression and splicing analyses. The raw sequencing data (FASTQ) have been deposited in the NCBI SRA database.

DIA mass spectrometry (Orbitrap Astral platform)

Hippocampus stored at −80 °C from AAV-CTR, AAV-PKN207, and AAV-PKFL mice, with three biological replicates per group, following protein extraction and trypsin digestion, peptides were combined in equal amounts and submitted to Shanghai Bioprofile Biotechnology Co., Ltd. for data-independent acquisition (DIA) using the Orbitrap Astral platform. Chromatographic separation was performed on a Vanquish Neo UHPLC (Thermo Scientific). Mobile phase A was 0.1% formic acid in water; mobile phase B was 0.1% formic acid in 80% acetonitrile. The column was pre-equilibrated with 96% A. Samples were enriched on a trap column (PepMap Neo 5 μm C18, 300 μm × 5 mm; Thermo Scientific), then subjected to gradient separation on a μPAC Neo High Throughput analytical column (Thermo Scientific), followed by DIA acquisition on an Orbitrap Astral mass spectrometer (Thermo Scientific). The total analysis time was 8 min.

IGV visualization

RNA-seq reads were aligned to the NCBI GRCh38 reference using STAR v2.7.10a to generate BAM files. To display expression intensity, BAM files were converted to BigWig format with bamCoverage from deepTools v3.5.1. Tracks were then visualized in Integrative Genomics Viewer (IGV).

RNA immunoprecipitation (RIP)

RIP was performed with RIP RNA-binding protein immunoprecipitation kit (GENE CREATE, JKR23003) according to the manufacturer’s instructions. Endogenous TDP-43 was immunoprecipitated from M17 cell lysates using the anti–TDP-43 antibody (ProteinTech, 10782-2-AP). RNA was extracted from the immunoprecipitates and reverse-transcribed to cDNA using a reverse transcription kit (ABclonal, RK20429). The resulting cDNA was used as template for RT-PCR to detect transcripts containing PKN1 cryptic exon 5a1. A rabbit IgG isotype control (GENE CREATE, JKR23003) served as the negative control. Primers: Rip-F, 5′-TTTCCCTAGGACTGGCCCTGT-3′; Rip-R, 5′-GACAAACCATCATGAAGATGCATCACG-3′. PCR cycling conditions: 95 °C for 5 min; 35 cycles of 95 °C for 30 s, 60 °C for 10 s, and 72 °C for 10 s; followed by 72 °C for 5 min. PCR products were resolved by agarose gel electrophoresis. PCR reactions were performed with OneTaq 2 × Master Mix (standard buffer) (Vazyme).

RNA electrophoretic mobility shift assay (RNA EMSA)

Experiments were performed with Chemiluminescent RNA EMSA Kit (Beyotime, GS606) according to the manufacturer’s instructions. 5′-biotin–labeled RNA probes were synthesized by Sangon Biotech and used at a final concentration of 150 pmol/L, adding 1 μL per reaction. Recombinant TDP-43 protein was purchased from MCE (HY-P71788); 2 μg was added to each binding reaction, and a concentration series (0.4, 0.8, 1.2, 1.6, 2.0 μg) was used to evaluate the affinity of different probes for TDP-43. Probe sequences (5' to 3') were: probe 1, CUGUGACUGUGUUUCUCUGCCC; probe 2, UGUAGUAUGCUGUGGUUGUCCA; probe 3, GAAUGAGGUGUGUCAACACGGG; probe 4, GACUUAGAGUCUCACUCUGUCA; probe 5, GUGUGCCUGUGUGUCAUUGCUC; probe 6, CUCACUGCCCUGUAUUCCACGC.

Samples were separated on 6% EMSA precast gels (Beyotime, GS306S) at constant voltage (100 V) for ~1.5 h in pre-chilled 0.5× TBE (Tris-Borate-EDTA buffer). Proteins were then wet-transferred to nylon membrane (Beyotime, FFN10) in pre-chilled 0.5× TBE at 360 mA for 1 h; membranes were pre-soaked in 0.5× TBE for 30 min prior to transfer. After transfer, membranes were UV-crosslinked for 10 min at ~8 cm from the light source. Following blocking, membranes were incubated with streptavidin–HRP for 20 min, washed according to the kit instructions, and developed with an ultrasensitive ECL substrate. Images were acquired using a ChemiScope 6100 imaging system (Clinx).

Cell culture and transfection

Human embryonic kidney 293T cell line (HEK-293T), human cervix epithelial cell line (HeLa), mouse  neuroblastoma Neuro-2a cell line (N2a), and rat neuroblastoma B104 cell line were cultured in high-glucose Dulbecco’s modified Eagle’s medium (DMEM) (Gibco, C11995500BT) supplemented with 15% fetal bovine serum (FBS) (Vazyme, F101-01), 100 U/ml penicillin and 100 μg/ml streptomycin (Gibco, 15140122); and human neuroblastoma M17  cells were cultured in DMEM/F12 (Gibco, C11330500BT) with 10% FBS, penicillin/streptomycin (Gibco), and incubated in a humidified atmosphere containing 5% CO₂ at 37 °C. Cells were seeded to culture plates (Sigma, 27964-99-4), and all transfections were performed with Lipofectamine^® 2000 (Invitrogen) according to the manufacturer’s instructions. Empty vectors were used as controls for the corresponding transfection.

Primary cortical neurons were prepared from E18.5 C57BL6/J embryos. After decapitation of the pregnant dam, embryos were placed in ice-cold Ca²⁺/Mg²⁺-free PBS. Under a stereomicroscope, the cerebral cortex was dissected and meninges removed. Cortex and hippocampus (with superficial vascular membranes removed) were separately incubated in 0.25% trypsin-EDTA (Gibco, 25200072) at 37 °C for 15 min, with gentle inversion every 5 min. Tissues were gently triturated using a 5 mL pipette tip to obtain single-cell suspensions, passed through a 40 μm cell strainer, and centrifuged at 500 g for 5 min. Pelleted cells were resuspended in high-glucose DMEM and plated onto poly-L-lysine–coated coverslips or culture plates (Sigma). After incubation at 37 °C for 4 h, the medium was replaced with Neurobasal (Gibco, 21103049) supplemented with B27 (Gibco), GlutaMAX (Gibco, 35050061), and penicillin/streptomycin (Gibco) for maintenance culture; half of the medium was replaced every 3 days. Plasmid transfection was performed with Lipofectamine^® 2000 (Invitrogen).

shRNA lentiviral treatment timelines: (1) Cell lines: Day 0, seeding; Day 2, addition of shRNA lentiviruses; Day 3, medium change; Day 6, harvest for RT-qPCR or immunoblotting. (2) Primary hippocampal neurons: Day 0, seeding; Day 6, infection; Day 7, medium change; Day 12, sampling for LDH cytotoxicity assay. (3) iPSC-derived cortical neurons: Day 0, seeding; Day 14, upon appearance of typical neuronal morphology, switch to maturation medium (Neurobasal + B27 + GlutaMAX) and add shRNA lentiviruses; Day 16, medium change; Day 22, harvest for RT-qPCR or immunoblotting.

iPSC source and cortical neuron differentiation

Human iPSCs were derived from normal human dermal fibroblasts and maintained on Matrigel-coated 6-well plates (Corning, 354234) in xeno-free mTeSR1 medium (STEMCELL Technologies, 85850) at 37 °C in 5% CO₂. Medium was changed daily, and cultures were passaged every 4–5 days using ReLeSR (STEMCELL, 100-0484). Typical colonies were round with smooth edges and high refractivity.

Cortical neuron differentiation followed a dual-SMAD inhibition strategy. On Day 0, healthy iPSCs were dissociated to single cells and plated on Matrigel; neural induction was performed in basal medium (DMEM/F12 + N2 + B27 minus vitamin A) supplemented with 10 μM SB431542 and 100 nM LDN-193189 for 7 days. Beginning on Day 7, inhibitors were gradually withdrawn, and 20 ng/mL FGF2 was added to expand neural progenitor cells (NPCs). Neuron-like morphology emerged by Day 14; cultures were then switched to a maturation medium (Neurobasal + B27 + GlutaMAX) supplemented with 10 ng/mL BDNF, 10 ng/mL GDNF, 200 μM ascorbic acid, and 1 μM cAMP to support differentiation and maturation. By approximately Day 30, cortical neurons stably expressing NeuN and MAP2 were obtained and, after confirmation by immunofluorescence, used for splicing analyses and functional assays.

Protein extraction and immunoblotting (Western blots)

Lysis and quantification: Cells were lysed in RIPA buffer (Beyotime, P0013B) supplemented with EDTA-free protease inhibitor (MCE, HY-K0010) and phosphatase inhibitor (MCE, HY-K0021). Lysates were centrifuged at 12,000 g for 15 min at 4 °C, and supernatants were collected. Protein concentrations were determined by the BCA assay (Beyotime, P0012). Equal amounts of protein were mixed with SDS loading buffer and denatured at 95 °C for 10 min.

Electrophoresis and transfer: Proteins were separated by 10% SDS–PAGE and transferred to PVDF membranes (Millipore Sigma, IPVH00010). Membranes were blocked in 5% nonfat milk in Tris-buffered saline (TBS) for 1 h at room temperature, incubated with primary antibodies (diluted in blocking buffer) overnight at 4 °C, washed in TBST (TBS + 0.1% Tween-20, 3 × 5 min), and then incubated with HRP-conjugated secondary antibodies (ABclonal, AS014 or AS003; 1:5,000) for 1 h at room temperature.

Detection and quantification: Signals were developed by enhanced chemiluminescence (ECL) and imaged using a ChemiScope 6100 system (Clinx). Band intensities were quantified with Fuji (ImageJ). Details of all primary antibodies are provided in Supplementary Table 1.

Extraction of RIPA-insoluble proteins from AD hippocampi

Approximately 2 μg of parahippocampal gyrus tissue from AD patients was homogenized on ice for 30 min in 300 μL prechilled RIPA buffer (50 mM Tris-HCl, pH 7.4; 150 mM NaCl; 1% Triton X-100; 0.1% SDS; 2 mM EDTA; supplemented with protease and phosphatase inhibitors). The homogenate was centrifuged at 500 g for 5 min at 4 °C to remove debris; the resulting supernatant was collected and further centrifuged at 16,000 g for 15 min at 4 °C. The final supernatant was designated the RIPA-soluble fraction, and the pellet the RIPA-insoluble fraction.

The pellet was washed once with prechilled RIPA buffer, resuspended in 10% SDS lysis buffer (10% SDS; 250 mM Tris-HCl, pH 6.8), and subjected to water-bath sonication until fully solubilized. Protein concentrations of both fractions were determined by the BCA assay (Beyotime). The RIPA-insoluble fraction was used for subsequent immunoblot analysis of pTDP-43 (S409/410).

Immunofluorescence staining

HEK-293T, M17, and HeLa cells were seeded on glass coverslips in 24-well plates and transfected/treated as described above. Cells were fixed at room temperature with 4% paraformaldehyde in PBS for 15 min, then permeabilized with 0.5% Triton X-100/PBS (Biosharp, BS084) for 15–30 min. Blocking was performed with 3% goat serum in PBST (0.1% Triton X-100/PBS) for 1 h at room temperature.

Primary and secondary antibodies were diluted in blocking buffer and incubated either overnight at 4 °C (up to 48 h) or for 1–2 h at room temperature, as required. After staining, samples were washed in PBST (3 × 5 min). Coverslips were mounted using an aqueous mounting medium containing DAPI (ZSGB-Bio, ZLI9557). Images were acquired and analyzed using a Zeiss LSM800 confocal microscope or an OLYMPUS IX81 inverted fluorescence microscope. Details of all primary antibodies are provided in Supplementary Table 1.

Protein crosslinking

To stabilize native intracellular interactions, cells were crosslinked with disuccinimidyl glutarate (DSG; MCE, HY-114697). Cells at ~80% confluence grown in 10-cm dishes were washed once with PBS and the medium was replaced with PBS containing an EDTA-free protease inhibitor (MCE, HY-K0010). DSG was added to a final concentration of 400 μM, and cells were gently rotated at room temperature for 30 min. The reaction was quenched with 20 mM Tris (pH 7.4) at room temperature for an additional 15 min. Cells were then collected by centrifugation at 1500 g for 5 min.

EdU cell proliferation assay

During the generation of a stable TDP-43 knockdown HEK-293T cell line, we observed markedly reduced proliferation. To verify this, DNA synthesis was assessed using an EdU-555 Cell Proliferation Kit (Beyotime, C0075S). Cells were infected with lentiviruses (shCTR and shTDP-43) and cultured for 10 days; on Day 10, cells were seeded at equal density onto sterile round coverslips placed in 6-well plates. After attachment and entry into logarithmic growth, EdU working solution was added (final concentration 10 μM) and cells were incubated for 2 h at 37 °C in 5% CO₂.

Following incubation, cells were washed twice with PBS and processed according to the manufacturer’s protocol: fixation (30 min), permeabilization, Click-reaction fluorescent labeling, and DAPI nuclear staining, all performed at room temperature protected from light. Images were acquired with an Olympus SLIDEVIEW VS200 fluorescence microscope. EdU⁺ cells exhibited red fluorescence, whereas DAPI stained nuclei blue to determine total cell counts. The fraction of EdU⁺ cells (EdU⁺/total cells) was quantified with ImageJ and used to evaluate DNA synthesis and proliferative capacity.

NMD blockade

To assess whether aberrantly spliced transcripts induced by TDP-43 knockdown are regulated by nonsense-mediated mRNA decay (NMD), NMD was inhibited in M17 cells by pharmacological and genetic approaches.

Pharmacological inhibition: Cycloheximide (CHX; MCE, HY-12320) was added to shTDP-43 and shCTR cells at 200 μg/mL. After 6 h of treatment, samples were collected for RT–qPCR.

Genetic interference: shRNAs targeting UPF1 (shUPF1#1/#2) were constructed and co-infected with shTDP-43 into M17 cells. Total RNA was extracted after 96 h.

Regulation by NMD was inferred by comparing the expression of aberrant splicing transcripts (e.g., PKN1-5a1) before and after CHX treatment or UPF1 knockdown.

Protein stability assays

DOX-inducible assessment of PKN1 and PKN207 stability in HEK-293T cells: HEK-293T cells were transfected with Tet-PKFL or Tet-PKN207 constructs. After 24 h, doxycycline (DOX; 1 μg/mL; MCE, HY-N0565) was added for 24 h to induce expression, after which induction was terminated by replacing the medium with DOX-free high-glucose DMEM. CHX (200 μg/mL) was then added to inhibit de novo protein synthesis, and cells were harvested at 0, 4, 8, and 12 h. Total proteins were analyzed by immunoblotting, and band intensities across time points were compared to evaluate the degradation rates of PKN1 and PKN207.

PKN1 stability under TDP-43 knockdown in N2a cells: In N2a cells, a CHX chase was initiated on Day 5 after establishing TDP-43 knockdown. Cells were switched to fresh serum-free medium, then CHX (200 μg/mL) was added, and samples were collected at 0, 4, 8, and 12 h. Total proteins were subjected to immunoblotting, and degradation rates of PKN1 were compared between knockdown and control groups to assess the effect of TDP-43 on PKN1 stability.

RNA extraction and cryptic exon detection

Cultured cells and neurons were directly lysed in TRIzol (Takara, 9109) to extract total RNA; human brain tissue was mechanically homogenized and likewise extracted with TRIzol. cDNA was synthesized using a reverse transcription kit (ABclonal, RK20429). RT–PCR was performed with OneTaq 2 × Master Mix with Standard Buffer (Vazyme, P222-03-AA) under the following program: 95 °C for 5 min; 35 cycles of 95 °C for 15 s, 60 °C for 15 s, and 72 °C for 15 s; followed by 72 °C for 5 min.

Detection of the PKN1 cryptic exon 5a1 used primers: forward (ceF1), 5′-CAGCACCAAGGACCGGAAGCTG-3′; reverse (ceR1), 5′-TCACAGGGCCAGTCCTTGGGTGT-3′. Detection of the alternative 3′ splice site (A3SS) used primers: forward (ceF1, as above); reverse (ceR2), 5′-GTCAGGACTCCCTGTGAGCGTGGAAATA-3′. Amplification of total PKN1 transcripts (non–junction-spanning) used: F1, 5′-ATGTTGCAGGACAGTAAGACCAAGA-3′; R1, 5′-CGCAGTACGTTCTTGGCACCC-3′.

All PKN1 minigene constructs and their mutants were also assayed by RT–PCR with the exon-spanning primers F1/R1 to simultaneously detect the 5a1 and A3SS events. PCR products were resolved on 2–3% TAE (Tris-acetate-EDTA) agarose gel and imaged with the SH-510 bioimaging system (Hangzhou Shenhua Technology Co., Ltd.). Band intensities were quantified in ImageJ.

RT–qPCR

RNA extraction and cDNA synthesis were performed as described above; primers were identical to those used for RT–PCR. Reactions employed a SYBR Green kit (ABclonal, RK21203) on a real-time PCR system (Bio-Rad, CFX2000). The 20 μL reaction contained: 10 μL 2 × SYBR Mix, 0.5 μL forward primer (10 μM), 0.5 μL reverse primer (10 μM), 50 ng cDNA, and ddH₂O to volume. Cycling conditions were: 95 °C for 5 min; 35 cycles of 95 °C for 15 s and 60 °C for 10 s; followed by a melt curve to verify specificity. Relative expression was calculated using the ^ΔΔCt method.

RNA-Seq analysis of ALS patient samples

Data source and preprocessing. Public human brain RNA-seq data were obtained from GSE153960 (NYGC ALS Consortium)¹⁷. The project applied unified QC, including verification of tissue type, brain region, and sex using >250 marker genes. For samples included in this study, the median sequencing depth was 4.2 × 10⁷ read pairs (range: 1.6 × 10⁷–1.67 × 10⁸). Raw reads were processed with the RAPiD-nf (Nextflow) pipeline: adapter trimming with Trimmomatic v0.36; alignment to GRCh38 (hg38.primary_assembly) using STAR v2.7.2a with GENCODE v30 annotation; and gene-level quantification with RSEM v1.3.1. Unless otherwise specified, downstream analyses used only uniquely mapped reads (STAR criteria: NH = 1 and MAPQ ≥ 255). When library strandedness was not reported, it was inferred with RSeQC infer_experiment.py, and strand corrections were applied accordingly in adapter annotation and sequence motif checks.

PKN1 full-length expression. Based on the GENCODE v30 PKN1 locus (chr19:14,433,305–14,471,895; GRCh38), RSEM outputs were summarized and TPM was computed: transcript counts were length-normalized to RPK, then normalized by the sum of RPKs across all transcripts within each sample to enable cross-sample comparison.

Splice-junction extraction and clustering. To obtain splicing evidence across the PKN1 region, splice junctions were extracted from BAM files with RegTools v0.5.1 (minimum anchor length 8 bp; maximum intron size 500 kb) and clustered with LeafCutter v0.2.9 (min_reads_per_cluster = 30, min_cluster_fraction = 1e-4) to generate a sample-by-junction count matrix. Junctions mapping ambiguously, of low complexity, or overlapping genomic blacklist regions were excluded. For stability, only junctions with ≥5 uniquely mapped reads in at least one group were carried forward.

Definition, coordinates, and deconvolution of PKN1-5a1. The cryptic exon of interest, PKN1-5a1, lies within the intron between exons 4 and 5 at chr19:14,445,088–14,445,214 (127 bp, GRCh38). A3SS event is present in the same region (chr19:14,445,088–14,445,317). To prevent A3SS contamination in quantification, we applied endpoint-strict matching and event typing: (1) Paired cryptic-exon junctions (CE-pair): Exon4 → CE boundary (14,445,088) and CE boundary (14,445,214) → Exon5. Both junctions must be supported by uniquely mapped reads within the same sample called as “PKN1-5a1 detected”. (2) A3SS-exclusive junctions: Exon4 → A3SS site (14,445,317) or A3SS site → Exon5 were labeled as A3SS and not counted toward the CE-pair. (3) The CE segment had to reside entirely within a single annotated intron, not overlap any annotated exon, and, after strand correction, both splice sites had to conform to the GT–AG rule (CT–AC on the negative strand). (4) N_mapped denotes the total number of uniquely mapped reads in a sample (primary, unique-mapped; counted by read end), excluding unmapped, secondary, or supplementary alignments; for paired-end data, each read end was counted separately.

RPM metric and definition of “positive” samples

For each sample, the relative abundance of the CE-pair was quantified by RPM (reads per million mapped reads): ${\mathrm{RPM}}_{\mathrm{up}\to \mathrm{CE}}={\mathrm{C}}_{\mathrm{up}\to \mathrm{CE}}/{\mathrm{N}}_{\mathrm{mapped}}\times {10}^6,{\mathrm{RPM}}_{\mathrm{CE}\to \mathrm{down}}={\mathrm{C}}_{\mathrm{CE}\to \mathrm{down}}/{\mathrm{N}}_{\mathrm{mapped}}\times {10}^6,$ and the total CE-pair abundance was reported as: ${\mathrm{RPM}}_{\mathrm{pair}}={\mathrm{RPM}}_{\mathrm{up}\to \mathrm{CE}}+{\mathrm{RPM}}_{\mathrm{CE}\to \mathrm{down}}.$

To mitigate depth effects, N_mapped was the sample’s total uniquely mapped reads. A sample was called positive only if both up→CE and CE→down junctions each had ≥1 uniquely mapped read and their endpoints strictly matched the CE boundaries. For cross-sample summaries/plots, only samples in which at least one CE junction had ≥5 uniquely mapped reads in any group were displayed.

Between-group comparisons and statistics. Group differences (e.g., ALS vs. control or across brain regions) were evaluated using RPM_pair as the primary metric and the Wilcoxon rank-sum test, with Benjamini–Hochberg correction (FDR < 0.05 as the significance threshold). Robustness was additionally assessed by comparing positive-sample proportions with Fisher’s exact test (BH-FDR correction). To reduce false positives, we imposed a consistency filter: in cases, both CE-junction RPMs had to increase concordantly, and in controls the median RPM_pair was approximately 0.

Visualization and software. Data processing was performed in R v4.2.2. Plots were generated with ggplot2, ggpubr, and patchwork. Distributions of RPM_pair were shown as violin plots overlaid with boxplots and individual sample points.

Mount Sinai AD transcriptome analysis

Bulk RNA-seq data from the MSBB project (brain area BA36; AMP-AD; Synapse ID: syn3159438)¹⁹ were analyzed. Adapters were trimmed with Trimmomatic v0.36; reads were aligned to GRCh38 using STAR v2.7.2a with GENCODE v30 annotation; TPM was quantified with RSEM v1.3.1. Downstream analyses used only uniquely mapped reads (primary, unique-mapped; counted by read end).

Splice junctions were extracted with RegTools v0.5.1 (minimum anchor 8 bp; maximum intron 500 kb) and clustered with LeafCutter v0.2.9; only junctions with ≥5 uniquely mapped reads in at least one group were retained for comparison. PKN1-5a1 was defined as the concurrent presence of the paired junctions Exon4 → 14,445,088 and 14,445,214→Exon5 (endpoints strictly matched; GT–AG/CT–AC after strand correction). Junctions for the co-located A3SS event (14,445,317) were labeled separately and excluded from the CE count.

For each sample, RPM was computed as: RPM_up→CE= C_up→CE/N_mapped×10⁶, RPM_CE→down=C_CE→down/N_mapped×10⁶, and the total paired CE abundance was: RPM_pair = RPM_up→CE+RPM_CE→down, where N_mapped denotes the sample’s total uniquely mapped reads.

Group comparisons were performed between Braak 0 and Braak 1–6, using log₂(RPM + 1) and log₂(TPM + 1) as inputs; statistical testing used the Wilcoxon rank-sum test.

Statistical analysis

Statistical analyses were performed using GraphPad Prism 8.0 (GraphPad Software Inc., San Diego, CA, USA). For comparisons between two groups, two-sided unpaired Student’s t-test or two-sided Wilcoxon test was used as indicated in the figure legends. For comparisons among multiple groups, one-way ANOVA was applied where appropriate. For large-scale transcriptomic datasets, including ALS and AD patient cohorts, statistical comparisons were performed using a two-sided Wilcoxon rank-sum test. For analyses involving comparison of regression slopes, analysis of covariance (ANCOVA) was used. Correlation analyses were performed using Pearson correlation, and correlation matrices were visualized as Pearson correlation heatmaps. A P value ≤ 0.05 was considered statistically significant. Data are presented as mean ± SEM unless otherwise stated.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

M.Y., Q.W., R.Y., D.K., and L.Z. performed the experiments. W.L. and M.Y. analyzed the RNA-seq data from ALS and AD patients. M.Y. designed the experiments, analyzed and interpreted the data, and wrote the original draft of the manuscript. M.Y., R.L., J.G., L.W., and X.W. supervised the project and contributed to data analysis. M.Y. and J.G. revised and finalized the manuscript. All authors contributed to the research and approved the manuscript.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Data availability

The RNA-seq data generated from the M17 cell model in this study have been deposited in the GEO under accession code GSE260737 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE260737)⁵⁹. Publicly available ALS patient RNA-seq data analysed in this study were obtained from the GEO under accession code GSE153960. The proteomics data from mouse hippocampal neurons was obtained from the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD069368. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-68916-0.