TDP-43’s disruption of RNA tail formation is a key problem in ALS and frontotemporal dementia

To identify APA events regulated by TDP-43 in neuronal cells, we employed the dynamic analysis of poly(A) from an RNA-Seq (DaPars) tool to previously published RNA-Seq datasets (). While a number of tools have been developed to quantify APA from bulk RNA-Seq data (), we selected DaPars (), which identifies and quantifies de novo poly(A) site usage by applying a linear regression model to localized changes in RNA-Seq read density within the 3′ UTR. Given the established role of TDP-43 in regulating the splicing of unannotated “cryptic” exons, we hypothesized that “cryptic,” unannotated poly(A) sites may be similarly utilized upon TDP-43 depletion or mutation. DaPars calculates the percentage of distal poly(A) site usage index (PDUI) for each transcript, with the results averaged across replicates to determine ΔPDUI in cells where TDP-43 was knocked down or mutated in comparison with control conditions. A positive ΔPDUI denotes increased relative expression of transcripts with a longer 3′ UTR, while a negative ΔPDUI signifies increased relative expression of transcripts with a shorter 3′ UTR. Figure 1A ^{23 25}

Using published RNA-Seq data from SH-SY5Y cells, SK-N-BE(2) cells, and iNeurons in which TDP-43 was depleted by shRNA knockdown (), we identified hundreds of genes in which TDP-43 knockdown resulted in significant APA (FDR adjusted [adj]< 0.05, |ΔPDUI| 0.1) (,,, and, and; supplemental material available online with this article;). APA genes exhibit both 3′ UTR lengthening and shortening upon TDP-43 knockdown; however, we observed increased use of distal poly(A) sites in 69.9% (460/658) of APA genes in SH-SY5Y cells, 59.6% (429/720) of APA genes in SK-N-BE(2) cells, and 78.1% (250/320) of APA genes in iNeurons. Nearly 20% of APA events were shared between the closely related SH-SY5Y and SK-N-BE(2) neuroblastoma cell lines (), including 4/10 of the top APA genes byvalue (and). Additionally, 16.3% (52/320) of APA genes in iNeurons overlapped with at least 1 of the immortalized neuronal cell lines (), indicating that, while APA is known to be highly cell-type specific (), many APA events regulated by TDP-43 are conserved across multiple neuronal cell types. 3 3 3 ¹² Figure 1, B–D Tables 1 2 3 Supplemental Tables 1–3 Figure 1E Tables 1 2 Figure 1E ¹⁷ P P https://doi.org/10.1172/JCI182088DS1↗

We also applied DaPars to previously published transcriptome data sets in which TDP-43 was depleted by siRNAs in SH-SY5Y cells () or in human embryonic stem cell–derived motor neurons (hESC-MNs) for 96 hours () (and). As expected, we observed fewer APA changes in cells in which TDP-43 was knocked down for a shorter duration or incompletely (~60% in hESC-MNs). Both datasets do, however, exhibit significantly decreased PDUI in, as previously shown for HEK293 cells upon TDP-43 siRNA knockdown (). ^{11 9} Supplemental Figure 1 Supplemental Tables 4 and 5 ¹⁴ SMC1A

Given that APA can affect cellular function by regulating mRNA stability and subcellular localization (), we performed Gene Ontology (GO) analysis of APA genes in each neuronal cell type (). Shared terms between datasets for biological pathways known to be highly relevant to ALS/FTD included “establishment of protein localization” (SH-SY5Y and iNeurons) and “cytoskeleton organization” (SH-SY5Y and SK-N-BE²), and shared terms for GO molecular functions were “kinase activity,” “protein kinase binding,” and “cytoskeletal protein binding” (SH-SY5Y and SK-N-BE²). These results indicate that TDP-43 regulation of APA impacts disease-relevant pathways and should be considered, along with differential expression and alternative splicing, as a key aspect of TDP-43 dysfunction. ²² Supplemental Figure 2 3

By cross-referencing significant APA genes with TDP-43 enhanced crosslink and immunoprecipitation followed by sequencing (eCLIP-Seq) data from SH-SY5Y cells (), we found that TDP-43 directly binds either within the 3′ UTR or in another region of the transcript to 43.2% (284/658) of APA genes in SH-SY5Y cells, 47.4% (341/720) of APA genes in SK-N-BE(2) cells, and 51.3% (164/320) of APA genes in iNeurons (and). In neuronal cells chronically depleted of TDP-43, it is likely that cytotoxicity contributes to gene regulatory changes that do not necessarily reflect direct regulation by TDP-43. Indeed, we noted that in iNeurons, only 36.7% of differentially expressed genes are targets of TDP-43 binding by eCLIP-Seq, while a higher proportion of CE (60.3%) and APA (51.3%) events are observed in genes directly bound by TDP-43 (). In support of our overall finding that TDP-43 depletion preferentially increases use of a distal PAS, we found that APA genes exhibit increased PDUI upon TDP-43 loss — when TDP-43 binds within the 3′ UTR, with iNeurons displaying the largest trend (). ²⁶ Figure 2A Supplemental Figure 3 Table 4 Figure 2, B–D 3 3 3

To experimentally validate TDP-43 APA events, we first utilized the above SH-SY5Y cell model, in which a doxycycline-inducible shRNA against TDP-43 is stably integrated (). We selected 2 APA genes to validate based on their robust and consistent shift in PDUI upon TDP-43 knockdown, and because TDP-43 binds directly within each 3′ UTR (). The Canopy FGF signaling regulator 3 () gene showed among the 3 greatest PDUI increases in each of our tested neuronal cell lines ().exhibits extensive alternative splicing, with numerous annotated transcript variants. In examining the RNA-Seq tracks for, we found that TDP-43 knockdown significantly increases PAS usage within a specific transcript variant (, variant 2), defined by an alternative last exon without an upstream splice junction. In line with these data, we confirmed by quantitative reverse transcription PCR (qRT-PCR) that TDP-43 depletion results in a significant increase in the use of the variant two 3′ UTR relative to variant one () in SH-SY5Y cells (and). To demonstrate thatAPA is specifically regulated by TDP-43, we designed minigene expression constructs in which exon 3, intron 3, and exon 4 of thegene () were cloned downstream of nanoluciferase with either WT sequence or with the TDP-43–binding site mutated. We found that deletion of the TDP-43–binding motif was sufficient to recapitulate theisoform switch observed upon TDP-43 knockdown (). We further validated thatAPA occurs in iNeurons following TDP-43 depletion (and). Interestingly, we observed a cell-type–specific effect of TDP-43 knockdown onvariant 1, whereby iNeurons but not SH-SY5Y cells display a modest, but significant increase invariant 1 expression in addition to a substantial increase in variant 2 expression (). Because we found an overall increase in bothisoforms in iNeurons depleted of TDP-43, we evaluated CNPY3 protein levels. In contrast with increased expression ofat the RNA level, we found that CNPY3 protein expression is decreased upon TDP-43 knockdown (). This suggests that TDP-43 regulation ofAPA may be functionally relevant in human neurons. ¹² Supplemental Figure 4 Figure 2, B–D Figure 3A Supplemental Figure 5A Figure 3B Figure 3C Supplemental Figure 6A Supplemental Figure 5 Figure 3D CNPY3 CNPY3 CNPY3 CNPY3 CNPY3 CNPY3 CNPY3 CNPY3 CNPY3 CNPY3 CNPY3 CNPY3 NM_001318848.2 NM_006586.5 NM_006586.5 3 3 3

Given that CNPY3 was one of the most consistent APA events observed across cell models (), we next investigatedAPA in postmortem frontal cortex tissue from FTLD-TDP and ALS/FTD patients. Because even healthy controls exhibit low, but detectable levels of phosphorylated TDP-43 by ELISA, we used the presence or absence of theCE as a more sensitive proxy for TDP-43 dysfunction (and). In agreement with our bioinformatics analyses and our experimental validation ofAPA in cell models of TDP-43 depletion, we observed significantAPA, corresponding with increased expression ofisoform variant 2 in the frontal cortex of TDP-43 proteinopathy patients (). Altogether, these results demonstrate that TDP-43 regulation of poly(A) site selection occurs in the CNS of ALS/FTD and FTLD-TDP patients. Specifically,APA increases the expression of a distinct protein-coding CNPY3 isoform with unknown function. Figure 2, B–D Table 5 Supplemental Table 6 Figure 3E CNPY3 UNC13A CNPY3 CNPY3 CNPY3 CNPY3

In contrast to the majority of APA genes in which TDP-43 binds within the 3′ UTR, DaPars calculated a significant decrease in distal PAS usage for(), which we experimentally validated by qRT-PCR in SH-SY5Y cells (). As noted, APA ofwas previously characterized in HEK293 cells upon TDP-43 knockdown (), indicating that this is a highly sensitive APA event across diverse cell types. Given that SMC1A is a subunit of the cohesin complex and plays a critical role in chromatin organization, we sought to investigate SMC1A in neuronal cells. TDP-43 depletion in iNeurons resulted in significantly decreased expression of the long 3′ UTRisoform and, in contrast to SH-SY5Y cells, increased expression of the short 3′ UTRisoform (). Furthermore, we found thatAPA corresponded to a nearly 3-fold increase in SMC1A protein expression (). To determine whether TDP-43 regulation ofoccurs in other neuronal cell types affected by TDP-43 proteinopathy, we similarly evaluatedAPA and protein expression in DIV38 iPSC-derived motor neurons (iPSC-MNs) in which TDP-43 was knocked down for 10 days (). We confirmed that TDP-43 loss indeed reduces use of the distalPAS in human motor neurons (). We further found that APA ofcorresponds with a significant increase in SMC1A protein levels in iPSC-MNs (), suggesting that altered regulation ofAPA may contribute to impaired chromatin organization upon TDP-43 loss, as previously reported in postmortem neuronal nuclei (). SMC1A SMC1A SMC1A SMC1A SMC1A SMC1A SMC1A SMC1A SMC1A SMC1A Supplemental Figure 4B Figure 4, A and B ¹⁴ Figure 4, C and D Figure 4E Supplemental Figure 6B Figure 4, F and G Figure 4H ²⁷ 3

To explore APA dysregulation that reflects TDP-43 mutation in addition to TDP-43 knockdown, we next applied DaPars to RNA-Seq data generated from SH-SY5Y cells with homozygous mutation of TDP-43achieved via CRISPR/Cas9 genome editing (). Consistent with prior studies, 72% (59/82) of significant APA events corresponded with increased use of a distal PAS (,). Notably, GO analysis revealed enrichment of genes that function in the “response to oxidative stress” pathway (). Numerous studies have implicated oxidative stress in ALS/FTD pathogenesis, including recent evidence that TDP-43 aggregation induces the generation of ROS (). To determine whether TDP-43SH-SY5Y cells display an altered oxidative stress response, we measured ROS at 30 minutes and 120 minutes after treating WT or TDP-43SH-SY5Y cells with hydrogen peroxide. While there was no initial difference in ROS levels between control and TDP-43cells, ROS levels were significantly higher in TDP-43SH-SY5Y cells compared with WT cells after 120 minutes of exposure (). This provides proof of concept that TDP-43 APA genes function in cellular pathways implicated in neurodegenerative disease and that characterization of APA events can highlight disease-relevant phenotypes in cell models of ALS. N352S N352S N352S N352S N352S ¹¹ Figure 5A Supplemental Table 7 Figure 5B ²⁸ Figure 5C

To further determine the significance of TDP-43 APA dysregulation in human patients, we considered APA events in neuron nuclei obtained from 7 postmortem ALS/FTD neocortex samples, where FACS sorting resulted in transcriptome data sets for neurons either containing or depleted of nuclear TDP-43 (). We identified 87 APA genes (|ΔPDUI| > 0.1,< 0.05) in neuronal nuclei lacking nuclear TDP-43 (,, and), but unlike in neuronal cell culture models, we observed a preference toward negative ΔPDUI in postmortem TDP-43 nuclear-depleted neurons, as 72.7% (63/87) of genes exhibited 3′ UTR shortening. By correlating the most significant APA events (|ΔPDUI| > 0.1, FDR< 0.05) with eCLIP-Seq of TDP-43, we noted that 57.7% of these APA genes are bound by TDP-43 (), suggesting that many APA events are directly regulated by TDP-43. We then performed GO analysis for APA events in coding mRNA with< 0.05 and [|ΔPDUI| ≥ 0.1] and found enrichment of genes functioning in the “histamine response,” “synapse assembly,” and “protein transport” pathways (). These pathways have been previously implicated in ALS disease models, again underscoring potential contributions of APA events to ALS pathobiology (–). ²⁷ Figure 6A Table 6 Supplemental Table 8 Table 7 Supplemental Figure 7 ^{29 31} P P P

The most substantial 3′ UTR lengthening in ALS/FTD neuronal nuclei (ΔPDUI = +0.363), was observed in the gene encoding(). Visualizing RNA-Seq tracks from this experiment with overlaid eCLIP-Seq generated by the ENCODE project (), we confirmed that TDP-43 bindsin its 3′ UTR immediately upstream of a canonical ATTAAA poly(A) signal (hg38: chr14:103503803-103503809) as well as in an upstream intron (). This suggests that nuclear TDP-43 normally represses use of a distalPAS, which then becomes preferentially utilized upon nuclear depletion of TDP-43. To test this hypothesis, we designed minigene expression constructs in which the3′ UTR was cloned downstream of nanoluciferase with either WT sequence or with the TDP-43 binding site mutated. We found that deletion of the TDP-43–binding motif was sufficient to drive increased distal PAS usage in the3′ UTR, supporting the hypothesis that TDP-43 represses use of the distal PAS (). Importantly, significant 3′ UTR lengthening inwas also observed in our APA analysis in SH-SY5Y and SK-N-BE(2) cells (), indicating that increased utilization of a distal poly(A) site in3 is a prominent effect in neuronal cells upon loss of nuclear TDP-43. Given the consistency of this result, we evaluated this phenotype in iPSC-MNs and confirmed that TDP-43 knockdown can induce markedly increased use of a distal PAS in thegene (). MARK3 MARK3 MARK3 MARK3 MARK3 MARK3 MARK MARK3 Table 6 ³² Figure 6B Figure 6C Figure 2, B and C Figure 6D

As MARK3 is a tau kinase associated with tau S262 phosphorylation in the early stages of AD pathogenesis (), and because mutations in closely related MARK4 can significantly increase AD risk, promote hyperphosphorylation of tau, and induce neuron toxicity (), we sought to investigate the functional consequences ofAPA in neurons. qRT-PCR analysis revealed that TDP-43 knockdown results in increased use of the distalPAS in iNeurons (), which corresponds with an increase in overallexpression (). To characterize the cis-regulatory elements within the3′ UTR that may affect its metabolism, we evaluated predicted miRNA-binding sites using miRBD () and TargetScan (), and we obtained predicted RBP-binding motifs using RBPmap () (). Both miRNA prediction tools highlighted a conserved binding motif for miR-142-3p within the3′ UTR (). Intriguingly, this miRNA was found to be upregulated in ALS mouse models and in sporadic ALS patients; indeed, serum levels of miR-142-3p have been negatively correlated with ALS clinical outcomes (). We also identified 2 conserved RBP-binding motifs within the3′ UTR, which are only present when poly(A) occurs at a distalPAS (and). The first motif is recognized by TDP-43 as well as by other RBPs (e.g., RBM24 and RBM38), while the second T/G-rich motif is predicted to be bound by several other RBPs, including TIA1 and HuR (). The presence of highly conserved miRNA and RBP-binding motifs in the distal region of the3′ UTR suggests that differential recognition of short versus long 3′ UTR isoforms by cis-regulatory elements may account for observed changes inmRNA expression upon TDP-43 knockdown. ^{24 33} Figure 7A Figure 7B ^{34 35 36} Supplemental Figure 8 Supplemental Table 9 ³⁷ Supplemental Figure 8 Supplemental Table 10 Supplemental Figure 8B MARK3 MARK3 MARK3 MARK3 MARK3 MARK3 MARK3 MARK3 MARK3 3

Using a large cohort of frontal cortex tissue from healthy controls (= 52) and FTLD-TDP patients (= 221), we found thatRNA expression is modestly increased in the frontal cortex in human disease (and). Using a smaller, independent cohort of patient tissue (and), we validated our finding of increasedRNA expression in the frontal cortex of ALS/FTD and FTLD-TDP patients relative to controls (, A and B). While we observed a small increase in the relative expression of the long 3′ UTRisoform in TDP-43 proteinopathy patient tissue, this did not reach statistical significance (). Given thatAPA is prominent in monocultured neuronal cells () and in isolated neuronal nuclei depleted of TDP-43 (), but not in bulk frontal cortex tissue (), it is likely thatRNA isoform expression is regulated by distinct mechanisms in nonneuronal cells or in neurons without TDP-43 pathology in FTLD-TDP. n n MARK3 MARK3 MARK3 MARK3 MARK3 Figure 7C Supplemental Table 11 Table 5 Supplemental Table 6 Supplemental Figure 9 Supplemental Figure 9C Figure 1, B and C Figure 6A Supplemental Figure 9C

Based on our finding that TDP-43 depletion results in an overall increase inexpression in iNeurons and in postmortem patient tissue, we next evaluated tau S262 phosphorylation, a known target of MARK3 kinase activity. We found that TDP-43 knockdown significantly increases tau S262 phosphorylation in iNeurons and that this effect is blocked by PCC0208017, a small molecule inhibitor of MARK3/4 () (). We similarly found that TDP-43 depletion in iPSC-MNs corresponds with a strong trend (= 0.069) toward increased tau S262 phosphorylation (). To determine whether MARK3 itself is sufficient to drive tau S262 hyperphosphorylation in human motor neurons, we transduced iPSC-MNs with lentivirus encoding a shRNA vector against, lentivirus encoding thegene, or control empty vector lentivirus. In iPSC-MNs subjected to MARK3 overexpression, we detected an approximately 2.4-fold increase in tau S262 phosphorylation (). Moreover, we observed an approximately 50% reduction in tau S262 phosphorylation in iPSC-MNs subjected to MARK3 shRNA knockdown (). MARK3 P MARK3 MARK3 3 3 ³⁸ Figure 7D Supplemental Figure 10A Supplemental Figure 10B Supplemental Figure 10B

While these results demonstrate that MARK3 protein levels correlate with tau S262 phosphorylation in iPSC-MNs and that inhibition of MARK3/4 kinase activity blocks increased tau S262 phosphorylation in iNeurons, we found no significant change in MARK3 protein expression in iNeurons or iPSC-MNs upon TDP-43 knockdown (and). Similarly, MARK3 protein levels were unchanged in ALS/FTD and FTLD-TDP patient tissue relative to controls (). We further confirmed that protein levels of the closely related tau S262 kinase MARK4 are unaffected by TDP-43 knockdown (). Given that we observed increased tau S262 phosphorylation upon TDP-43 knockdown, which was blocked by treatment with a MARK3/4 inhibitor, yet we observed no increase in MARK3 protein levels, we next examined whether MARK3 subcellular localization is altered by TDP-43 depletion in iNeurons. Indeed, we noted a significant increase in the subcellular localization of MARK3 to neurites relative to soma upon TDP-43 knockdown, which corresponded with a similar trend of increased neurite localization of tau pS262 (). Moreover, we observed a decrease in total tau localization in the neurites of iNeurons upon TDP-43 depletion (), suggesting broader dysregulation of tau dynamics following TDP-43 knockdown. Taken together, these results reveal a potentially important mechanistic link between TDP-43 and tau biology and suggest that TDP-43 dysregulation of neuronalAPA may contribute to altered cytoskeletal function in ALS/FTD and related neurodegenerative disorders. 3 3 3 3 Supplemental Figure 10C Supplemental Figure 11A Supplemental Figure 11B Supplemental Figure 11C Figure 7, E–G Figure 7, E–G MARK3

This study includes DaPars analysis of 1 published transcriptome data set (), based on sorting of neuron nuclei containing or depleted of TDP-43 from 7 postmortem ALS/FTD patients (3 females and 4 males), where the study was not powered to evaluate sex as a biological variable. For experiments using postmortem human tissue, roughly equal numbers of males and females were analyzed (and), and sex was not evaluated as a biological variable. ²⁷ Supplemental Table 6 Supplemental Table 11

DaPars software, described previously (), allows for the joint analyses of multiple samples based on a 2-normal mixture model, which is used to calculate the PDUI value. Briefly, we extract a 3′ UTR annotation for each gene using the “DaPars_Extract_Anno.py” script within DaPars2. We then used the “samtools flagstat” command () to calculate the sequencing depth for each sample. Finally, we use DaPars2 to calculate the percentage of the PDUI divided by the total expression level of each transcript across samples. ^{25 50}

WT human neuroblastoma SH-SY5Y cells (ATCC) and SH-SY5Y cells with homozygous mutation of TDP-43(obtained from the lab of Don Cleveland; ref.) were maintained in 50% Eagle’s minimum essential medium (EMEM) (ATCC, 30-2003), 50% Ham’s F12 (ThermoFisher, 11765047), supplemented with 10% fetal bovine serum, and 50 U/mL penicillin-streptomycin. Cells were grown at 37°C and 5% CO. N352S ¹¹ 2

SH-SY5Y cells with stable integration of a doxycycline-inducible shRNA cassette targeting TDP-43 were obtained from the lab of Pietro Fratta (). Cells were treated with 1 μg/mL doxycycline (Sigma-Aldrich, D9891-1G) for 7 days to induce TDP-43 knockdown. ¹²

Theminigene construct comprises an insert containing the NanoLuc luciferase gene (FLAG-tagged at the C-terminus) and the3′ UTR cloned into the pJTI R4 DEST CMV pA vector (ThermoFisher) using BbsI and PmeI restriction enzyme sites. The HSV TK poly(A) signal was deleted. Theminigene construct comprises an insert containing the NanoLuc luciferase gene immediately followed by exon 3, intron 3, and exon 4 of thegene cloned into the pJTI R4 DEST CMV pA vector (ThermoFisher) using BbsI and PmeI restriction enzyme sites. Exon 4 is FLAG tagged at the C-terminus. Synthesis of the described inserts and assembly of the vectors was performed by Azenta. With the exception of the primers used to amplify thelong 3′ UTR, all primer pairs were designed to specifically amplify minigene construct products by containing either FLAG tag or NanoLuc luciferase gene sequence. MARK3 MARK3 CNPY3 CNPY3 MARK3

qRT-PCR primer sequences were as follows: CNPY3 minigene var1 3′ UTR fwd: GCATGTCAGAGACCTTTGAGAC; CNPY3 minigene var1 3′ UTR rev: GTCATCGTCTTTGTAGTCCTGCTTC; CNPY3 minigene var2 3′ UTR fwd: GAATCCTCGCCTCGGACTTG; CNPY3 minigene var2 3′ UTR rev: GGATGAAGGACAATCCCGAATC; MARK3 minigene fwd: GACGATGACGATAAATAACCCAGTGA; short 3′ UTR MARK3 rev: TAGAAGATGCAGACGTTATTGCC; long 3′ UTR MARK3 fwd: CAGGTTTACAGTTCATGCCTGT; long 3′ UTR MARK3 rev: CACACACAAGCAATGTTCACAAC.

iNeurons were generated from the KOLF2.1J iPSC line () in which neurogenin 2 (NGN2), under the control of a tetracycline-inducible promoter, was stably integrated into the adeno-associated virus integration site (AAVS) locus. iPSCs were cultured in mTeSR (STEMCELL Technologies, 100-0276). As previously described (), NGN2 expression was induced by doxycycline for 3 days; then immature iNeurons were replated and cultured for an additional 7 days (in the presence of doxycycline) prior to experimentation. 3 3 ^{51 52}

iPSCs from the KOLF2.1J line were maintained in mTeSR (STEMCELL Technologies, 100-0276). According to the manufacturer’s instructions, iPSCs were differentiated into motor neurons from days 0–14 (STEMCELL Technologies, 100-0871) and subsequently matured from days 14–28 (STEMCELL Technologies, 100-0872). iPSC-MNs were transduced using lentiviral vectors at day 28 and collected at day 38. Cells were cultured at 37°C and 5% CO. 2

Lentiviral constructs encoding GFP (VectorBuilder, LVM[VB010000-9298rtf]-C), MARK3-V5 (Addgene, 107235), shRNA control (Sigma, SHC002), tardbp shRNA (Sigma, TRCN0000016038), or MARK3 shRNA (Sigma, TRCN0000001564) were packaged into lentivirus by VectorBuilder. iPSC-MNs were transduced with 5 MOI lentivirus at day in vitro (DIV) 28 for 10 days, and iNeurons were transduced with 5 MOI lentivirus at DIV 10 for 10 days. 3

shRNA target sequences were as follows: control: CAACAAGATGAAGAGCACCAA; Tardbp: GCTCTAATTCTGGTGCAGCAA; and MARK3: TGTGTGTGAAGTGGTGTATAT.

iNeurons were treated with 5 μM MARK3/4 inhibitor (PCC0208017³⁸) or DMSO (vehicle) for 24 hours prior to harvesting cells for downstream assays. 3

RNA was isolated from cell lysates (QIAGEN, 74106) and reverse transcribed to generate cDNA (Thermo Fisher, 11756500). PCR was performed using 10–30 ng cDNA template, 0.5 μM forward and reverse primers, and OneDNA polymerase mastermix (NEB, M0488L). PCR products were separated by agarose gel electrophoresis using 1.5% agarose (VWR, 0710) in 1× TBE buffer at 150V. Taq

RT-PCR primer sequences were as follows: MARK3 fwd: GGTTTAAGCGGATATCGGGG; MARK3 prox rev: TAGAAGATGCAGACGTTATTGCC; and MARK3 dist rev: CACACACAAGCAATGTTCACAAC.

Relative fold change of CNPY3, SMC1A, and MARK3 was determined by qRT-PCR using SYBR Green Master Mix (Thermo Fisher, A25776) with hypoxanthine-guanine phosphoribosyltransferase (HPRT) as an endogenous control.

qRT-PCR primer sequences were as follows: CNPY3 var1 3′ UTR fwd: CCAGCATCCTCTGTCCTGA; CNPY3 var1 3′ UTR rev: GAAGAGGGCACAGCCAAG; CNPY3 var2 3′ UTR fwd: GACCACCTGGGATCTTCCT; CNPY3 var2 3′ UTR rev: CACATCGTGGATCTTGCTGAG; SMC1A prox 3′ UTR fwd: CCTGTCTGGATCCCTAAGCTG; SMC1A prox 3′ UTR rev: CTCCAGACCTAACATCACCTCTG’ SMC1A dist 3′ UTR fwd: GTTAGTCAGTAGCAGTAGGAGGAG; SMC1A dist 3′ UTR rev: GCATTCACAGGGAAATAAGGAAGAC; total MARK3 fwd: CAGTCTCCTCACCACAAAGTGC; total MARK3 rev: TGCTGGTCTGACTCCTTTTCGG; short 3′ UTR MARK3 fwd: GGTTTAAGCGGATATCGGGG; short 3′ UTR MARK3 rev: TAGAAGATGCAGACGTTATTGCC; long 3′ UTR MARK3 fwd: CAGGTTTACAGTTCATGCCTGT; long 3′ UTR MARK3 rev: CACACACAAGCAATGTTCACAAC; HPRT fwd: CATTATGCTGAGGATTTGGAAAGG; and HPRT rev: CTTGAGCACACAGAGGGCTAC.

To evaluate totalRNA expression, our study cohort included 273 postmortem cases classified into 2 groups: healthy controls (= 52) and FTD with TDP-43 pathology (FTLD-TDP) (= 221). A separate cohort, in which phosphorylated TDP-43 (pTDP-43) was previously quantified by ELISA and the presence or absence of theCE was previously determined, was used to evaluateandAPA. This cohort () consisted of healthy controls with low, but detectable levels of pTDP-43 and no detectableCE (= 11) and ALS/FTD or FTLD-TDP cases with low or high levels of pTDP-43 and confirmedCE expression (= 30). Finally, to evaluate MARK3 protein levels, our study cohort consisted of healthy controls with low pTDP-43 (= 5) and ALS/FTD or FTLD-TDP cases with high pTDP-43 (= 11). A summary of patient data is included inand. RNA was extracted from frontal cortex tissue following the manufacturer’s protocol using the RNAeasy Plus Mini Kit (QIAGEN) and as previously described (,). Up to 3 cuts of the sample was used for extraction and only the high-quality RNA samples were processed for downstream analysis. RNA concentration was measured by using Nanodrop technologies (Thermo Fisher) and the RNA integrity number (RIN) was evaluated by Agilent 2100 Bioanalyzer (Agilent Technologies). Subsequently, 500 ng of the total RNA extracted was reverse transcribed into cDNA using the High-Capacity cDNA Transcription Kit (Applied Biosystems) per the manufacturer’s instructions. cDNA samples, in triplicate, with SYBR GreenER qPCR SuperMix (Invitrogen), were further evaluated for the quantitative real-time PCR (qRT-PCR) on a QuantStudio 7 Flex Real-Time PCR System (Applied Biosystems). Relative quantification of totallevels was determined using the ΔΔCt method and normalized to 2 endogenous controls,and, while quantification ofandAPA was normalized to. Protein was extracted from frontal cortex tissue using RIPA lysis buffer, and MARK3 (Cell Signaling Technology, 9311S, 1:2000) was normalized to GAPDH (Meridian Life Science, H86504M, 1:10,000). MARK3 n n UNC13A MARK3 CNPY3 UNC13A n UNC13A n n n MARK3 GAPDH RPLP0 MARK3 CNPY3 GAPDH Table 5 Supplemental Table 6 Supplemental Table 11 ^{13 53}

Protein extracts were prepared in RIPA lysis buffer (Thermo Fisher, 89900) with protease and phosphatase inhibitors (Thermo Fisher, 87786). Lysates were sonicated at 4 °C (Bioruptor Pico), denatured in LDS sample buffer (Thermo Fisher, NP0007) with a reducing agent (Thermo Fisher, NP0009), and heated at 70 °C for 10  minutes. Lysates were electrophoresed by sodium dodecyl sulfate–polyacrylamide gel electrophoresis using a 4%–20% gradient polyacrylamide gel (Bio-Rad, 5678093) and transferred to a 0.45 μm polyvinylidene difluoride membrane (Bio-Rad, 1704157). Blots were incubated with primary antibodies diluted in 5% nonfat milk in TBST (TRIS-buffered saline, 0.05% Tween-20) overnight at 4 °C and with secondary antibodies diluted in 5% nonfat milk in TBST (Thermo Fisher, A16078 [goat anti-mouse HRP, 1:3000], A16110 [goat anti-rabbit HRP, 1:3000], A32728 [goat anti-mouse Alexa Fluor 647, 1:1000], A32733 [goat anti-rabbit Alexa Fluor 647, 1:1000]) for 60  minutes at RT. Detection was performed using enhanced chemiluminescence substrate (Genesee Scientific, 20–300S) and blots were imaged on a ChemiDoc MP (Bio-Rad). Primary antibodies were as follows: TDP-43 (Proteintech, 12892-1-AP, 1:3000), MARK3 (Cell Signaling Technology, 9311S, 1:2000), MARK4 (Proteintech, 20174-1-AP, 1:2000), CNPY3 (Proteintech, 15215-1-AP, 1:1000), SMC1A (abcam, ab9262, 1:2500), phospho-tau Ser262 (Thermo Fisher, 44-750-G, 1:2000), and TAU-5 (ThermoFisher, AHB0042, 1:2000).

iNeurons plated in 8-well chamber slides (Thermo Fisher, 154941PK) were fixed in 4% PFA for 15 minutes, permeabilized, and blocked in 0.3% Triton X-100 in 2% normal donkey serum (NDS) for 30 minutes, and incubated with primary antibodies in 1.5% NDS for overnight at 4°C. Secondary antibodies were diluted in 1.5% NDS and incubated with samples for 60 minutes at room temperature. Hoescht was added to visualize nuclei (Thermo Fisher, H1399). Slides were mounted using ProLong Glass Antifade Mounting Media (Thermo Fisher,). Cells were imaged on a Nikon A1R confocal microscope using a ×60 objective. Antibodies used included the following: MARK3 (Cell Signaling Technology, 9311S, 1:200). phospho-tau Ser262 (Thermo Fisher, 44-750-G, 1:200), and TAU-5 (Thermo Fisher, AHB0042, 1:500). Fluorescence intensity of tau pS62, total tau, or MARK3 was quantified as described (). Each raw image was maximally projected and fluorescence intensity thresholds were set using identical values for all conditions per experiment. Thresholded areas occupied by a fluorescence signal for tau pS62, total tau, or MARK3 were measured; then fluorescence area was divided by the cell area, which was determined by subtracting the background fluorescence. To quantify the neurite/soma expression levels of these proteins, the neurite values were defined by the area 10 μm away from the soma. All quantitation is represented as combined results for all the repeats of a particular experiment (= 10–15 images analyzed per condition). 3 P36984 ⁵⁴ n

WT and TDP-43SH-SY5Y cells were plated on a 96-well plate and treated for 120 minutes with 100 μM HO. A fluorometric intracellular ROS detection kit was used to detect ROS levels at 30 and 120 minutes (Sigma, MAK144) using a Varioskan microplate reader (Thermo Fisher). N352S 2 2

GO analysis was performed using ShinyGO, version 0.77 (), for GO Biological Processes and GO Molecular Function pathway databases. GO analysis was performed for APA events in protein-coding genes with ΔPDUI 0.1 and< 0.05. Graphical representations of the top 10 pathways with an FDR adj.< 0.05 cut-off were generated in R, version 4.3.1, utilizing the ggplot and sjPlot packages. ⁵⁵ P P

As indicated, for pairwise comparisons, statistical significance was determined by unpaired 2-tailed Student’stest or by Mann-Whitneytest (). For pairwise comparisons in which the controls of distinct biological replicates were normalized to 1, statistical significance was determined by a 2-tailed, 1-sampletest in which experimental conditions were compared with a value of 1. For multiple comparisons, statistical significance was determined by 1-way ANOVA with post hoc Tukey’s test. Hypergeometric distribution was used to determine the statistical significance of Venn diagram overlaps. The significance level (α) was set at 0.05 for all experiments. Data were compiled and analyzed using Microsoft Excel or Prism 10 (GraphPad). t U t Figure 7C

This study does not include research conducted on animals or human subjects. For more information on deidentified postmortem human tissue used in this study, seeand. Supplemental Table 6 Supplemental Table 11

Full documentation and workflow of the DaPars2 algorithm is publicly available on github:Values for all data points in graphs are reported in thefile. Deidentified information on postmortem human samples is available inand. Any additional information on underlying data can be made available by the corresponding author upon request. https://github.com/3UTR/DaPars2/tree/v2.1↗ Supporting Data Values Supplemental Table 6 Supplemental Table 11