TDP-43K145Q expressing neurons are susceptible to nuclear TDP-43 phase separation
We originally showed that TDP-43 acetylation is linked to RNA-binding deficiency, aggregation, and pathology52,53. We sought to expand these findings by exploring the behavior of acetylation-mimic TDP-43 variants in primary murine cortical neurons. We employed lentiviral vectors that encode either wild-type (TDP-43wt), acetylation-deficient (TDP-43K145R), and acetylation-mimic (TDP-43K145Q) variants to overexpress TDP-43 species in neurons, and then examined their subcellular localization by immunofluorescence microscopy. In the absence of acute cellular stress, most neurons overexpressing TDP-43K145Q showed distinct nuclear foci that were not apparent with TDP-43wt or TDP-43K145R constructs (Fig. 1a, b). When neurons were exposed to acute an oxidative stress (200µM sodium arsenite), a sensitizing trigger for TDP-43 dysfunction59–62, there was a significant increase in TDP-43 foci formation with all variants to some degree (Fig. 1a, c). The response in neurons expressing TDP-43K145Q was robust, resulting in the formation of numerous large, bright TDP-43-positive foci, as well as small TDP-43-positive nuclear puncta. These aberrant TDP-43 structures were absent from cells expressing TDP-43wt or acetylation-null TDP-43K145R, indicating that acetylation-mimic TDP-43K145Q alters TDP-43 conformation within the nucleus and sensitizes neurons to oxidative stress. By coupling high content wide-field microscopy with quantitative image analysis, we observed a three-fold increase in TDP-43 aggregation in neurons expressing TDP-43K145Q (Fig. 1c).
TDP-43 overexpression can result in general toxicity and altered TDP-43 function, depending on duration and the level of over-expression20,63,64. To avoid potentially confounding non-specific toxicity, we transitioned to a more physiologically relevant model to further elucidate the impact of RNA-binding deficient acetylation-mimic TDP-43. We employed CRISPR-Cas9 mutagenesis to introduce a single amino acid substitution at position 145 (K145Q) in the endogenous mouse Tardbp locus, thereby generating the TDP-43K145Q knock-in line (Fig. S1b). By targeting the native mouse gene, we both avoid TDP-43 overexpression and maintain critical auto-regulatory control of the endogenous Tardbp transcript18. A TDP-43K145Q founder line was propagated as heterozygotes and continually re-sequenced to confirm retention and propagation of the K145Q substitution (Fig. S1c, d). Sequencing did not detect any unintended mutations, insertions, or deletions throughout the propagation of this line. Both heterozygous and homozygous TDP-43K145Q mice were born at normal mendelian frequencies and showed no obvious developmental defects.
We first investigated the effects of TDP-43K145Q expression in neurons in vitro by isolating and culturing primary cortical neurons from homozygous TDP-43K145Q, hereafter referred to as TDP-43KQ/KQ mice, and compared them to TDP-43wt -derived neurons. Quantitative image analysis of untreated TDP-43KQ/KQ neurons showed mild nuclear TDP-43 depletion and slightly increased levels of cytoplasmic TDP-43 compared to TDP-43wt (Fig. 2a, c). Exposing neurons to acute oxidative stress induced larger and more abundant TDP-43-positive nuclear foci in acetylation-mimic TDP-43KQ/KQ neurons than in TDP-43wt neurons (Fig. 2a, b). The presence of nuclear TDP-43 foci has been linked to TDP-43 phase separation and loss-of-function65–67, suggesting that TDP-43 acetylation drives these phenomena. Therefore, to determine if TDP-43 function is impaired, we used a TDP-43-dependent CFTR splicing reporter, in which full-length GFP is fused to an mCherry gene is interrupted by exon 9 of CFTR68. Functional TDP-43 suppresses CFTR exon 9 inclusion69, thereby promoting expression of both GFP and mCherry, while loss of TDP-43 function allows GFP but not mCherry expression (Fig. 2d). A neuron-specific synapsin (hSyn) promoter was used to selectively deliver the CFTR reporter via lentiviral transduction of primary neurons. Using RT-PCR, we observed impaired splicing in TDP-43KQ/KQ neurons, as assessed by reduced exclusion of the CFTR exon 9 (Fig. 2e-f). Overall, these results indicate that a single endogenously expressed acetylation-mimic TDP-43K145Q mutation is sufficient to alter TDP-43 localization, induce nuclear phase separation, and impair splicing in a murine primary neuron culture model.
To assess this model’s relevance to human neurons, we used CRISPR/Cas9 to generate a panel of human induced pluripotent stem cell (hiPSC) lines harboring acetylation-mimic TDP-43 (TDP-43K145Q.12 and TDP-43K145Q.18), acetylation-deficient TDP-43 (TDP-43K145R.2 and TDP-43K145R.12), or unmodified TDP-43 (TDP-43wt) (Fig. S2) and differentiated these lines into cortical neurons (Fig. S3). Untreated hiPSC-derived TDP-43K145Q cortical neurons were morphologically identical to TDP-43wt and TDP-43K145R neurons and showed similar patterns of TDP-43 localization (Fig. 2g, Fig. S4). All hiPSC-derived lines showed a granular nuclear TDP-43 localization pattern under normal conditions, consistent with physiologic de-mixing of nuclear TDP-4361. Following acute oxidative stress, TDP-43K145Q neurons showed TDP-43 nuclear clearing and the formation of large, intensely labeled TDP-43-positive foci (Fig. 2g, Fig. S4). In comparison, cortical neurons expressing TDP-43wt or TDP-43K145R maintained nuclear TDP-43 and formed small stippled TDP-43 foci. We note that TDP-43 nuclear clearing and foci formation was more robust in hiPSC-derived neurons compared to mouse neurons, suggesting human neurons may be more sensitive to the effects of RNA-binding deficient TDP-43. Similar to the mouse neuron analysis, hiPSC-derived TDP-43K145R neurons were indistinguishable from TDP-43wt neurons, again supporting charge neutralization as a primary driver of TDP-43 loss of function.
TDP-43 acetylation-mimic mice develop age-dependent cognitive and behavioral defects
To evaluate neurodegenerative phenotypes in TDP-43 acetylation-mimic mice, we aged homozygous TDP-43KQ/KQ mice and WT littermates and performed an extensive battery of behavioral analysis to assess cognitive and motor function, which reflect impairments that are commonly impacted in the spectrum of TDP-43 proteinopathies6. Since there were no significant differences between males and females in any behavioral analyses described below, we pooled both sexes into either WT or TDP-43KQ/KQ groups. At 12 months old, TDP-43KQ/KQ mice showed significant reduction of body weight compared to WT littermates, and this difference was maintained until end point analysis at 18 months old (Fig. 3a).
Evaluation of exploratory activity and locomotion in an open field test demonstrated that TDP-43KQ/KQ mice spend significantly more time in the center region (Fig. 3b), with no differences in total distance traveled (Fig. S5a), indicative of decreased anxiety-like behavior70. Acoustic startle testing revealed impaired prepulse inhibition (PPI) at 12 months old (Fig. S6a), indicative of deficits in sensorimotor gating, a form of inhibitory control, in TDP-43KQ/KQ mice71,72, which is a phenomenon that can be observed in early dementia73. In 18-month-old animals, evaluation of acoustic startle response and PPI was confounded by hearing impairments (Fig. S6b), however the altered activity reflected by increased time in the center region of an open field test was maintained at this advanced age (Fig. 3b). Thus, consistent patterns of behavioral disinhibition and reduced anxiety-like behavior were apparent in TDP-43KQ/KQ mice over time.
We next performed contextual and cue-dependent fear conditioning as an index of hippocampal and cortical function74–76. Context-dependent fear testing revealed reduced freezing times in TDP-43KQ/KQ mice, with trends observed at 12 months old and more significant deficits at 18 months, suggesting age-dependent impairments in contextual learning (Fig. 3c). Similarly, auditory cue-dependent fear testing revealed significant impairments in associative cue learning in TDP-43KQ/KQ mice at 12 months of age (Fig. 3d), a behavior thought to be mediated by the amygdala and higher-order cortical regions important in inhibitory control77. As mentioned above, general auditory defects in both genotypes at 18 months old confounded interpretations of any cue-dependent learning deficits at this advanced age (Fig. 3d).
Morris Water Maze testing was used to evaluate swimming ability and spatial learning78, which showed equivalent swim speeds, suggesting no motor impairments in TDP-43KQ/KQ mice at 18 months of age (Fig. S5b). While assessment of spatial learning showed a trend towards impaired acquisition learning (Fig. 3e, f), we observed more prominent defects in reversal learning after moving the location of the platform, as determined by significant delays in escape latency. These findings support deficits in cognitive flexibility in TDP-43KQ/KQ mice compared to WT littermates (Fig. 3g, h)79.
Finally, we assessed motor function in these cohorts and were surprised to find no overt signs of motor impairment in TDP-43KQ/KQ mice, even at 18 months of age. TDP-43KQ/KQ mice do not differ from WT littermates in motor coordination as assessed by rotarod testing (Fig. S5c-d) or in grip strength as measured using digital force meters (Fig. S5e). Moreover, there were no differences in swim speed or distance traveled in an open field at any age tested (Fig. S5a, b). The preferential deficits in learning and behavioral control support an FTLD-like, rather than an ALS-like, phenotype in TDP-43KQ/KQ mice.
TDP-43KQ/KQ mice display progressive TDP-43 dysfunction and neurodegeneration in neocortex and hippocampus
Given the FTLD-like behavioral phenotype observed in TDP-43KQ/KQ mice, which was most pronounced at 18 months, we examined the neocortex and hippocampus of aged animals for characteristics of TDP-43 pathology including TDP-43 aggregation, mislocalization, and neuronal loss17,43. We performed immunohistochemical and immunofluorescent labelling, confocal microscopy, and automated quantitative image analysis to assess TDP-43 localization in 18-month-old TDP-43KQ/KQ. Quantification of NeuN-positive cells revealed significantly reduced neuron density in TDP-43KQ/KQ neocortex compared to WT mice (Fig. 4a, b). In contrast, using a Cresyl Violet stain, which labels the rough endoplasmic reticulum and nuclei of all cell types including glia80, we did not observe significant differences in cell density (Fig. 4c, d), suggesting preferential depletion of neurons in TDP-43KQ/KQ cortex.
We next examined the tissue for hallmarks of TDP-43 aggregation and did not detect prominent nuclear or cytoplasmic TDP-43 inclusions in the neocortex or hippocampus of TDP-43wt or TDP-43KQ/KQ mice at 18 months of age. Since TDP-43 expression may be altered in FTLD-TDP and other TDP-43 proteinopathies18, 81–83, we assessed TDP-43 immunoreactivity and abundance in NeuN-positive neurons in the neocortex and hippocampus, and then visualized the single-cell resolution data using SuperPlots84. We observed trends of elevated TDP-43 protein in the neocortex and the CA3 region of the hippocampus (Fig. 4e-f, h-i) and also a potential increase in the cytoplasmic to nuclear TDP-43 ratio within CA3 neurons (Fig. 4h, j). We note that these effects were not significant using microscopy and quantitative image analysis approaches, which may result from technical challenges in achieving accurate image segmentation at the subcellular level needed to separate nuclear and cytoplasmic TDP-43 intensity within tissue samples. Therefore, to more reliably detect alterations in TDP-43 localization and solubility, we turned to an alternative biochemical approach of sequentially fractionating isolated hippocampus and neocortex tissue to generate soluble (RIPA-extracted) and insoluble (Urea-extracted) protein fractions. TDP-43KQ/KQ mouse neocortex harbored insoluble phosphorylated TDP-43 at the disease-associated Ser409/410 locus85 (p409/410) at 12 months of age, which was even more prominent at 18 months (Fig. 5a, c), a timepoint at which TDP-43KQ/KQ mice show prominent behavioral and cognitive defects (Fig. 4). Notably, p409/410 immunoreactivity was minimal in the hippocampus of 12-month-old TDP-43KQ/KQ mice but increased dramatically by 18 months (Fig. 5d, f), coinciding with the onset of hippocampal-mediated learning deficits (Fig. 3c). Though p409/410 was elevated in TDP-43KQ/KQ mice, we were surprised to find that the total insoluble TDP-43 pool was not prominently altered, suggesting increases in TDP-43 phosphorylation may precede overt conversion towards insoluble TDP-43 accumulation.
We also examined soluble TDP-43 levels, as acetylation-induced loss of function may result in autoregulatory feedback that increases production of TDP-4318. The soluble TDP-43 pool was significantly increased in TDP-43KQ/KQ at 12 months in the neocortex and at 18 months in both neocortex (Fig. 5a, b) and hippocampus (Fig. 5d, e), consistent with increases in autoregulated TDP-43 protein levels. To assess soluble TDP-43 localization using a biochemical approach, we performed subcellular fractionation to isolate nuclear and cytoplasmic proteins from the neocortex and hippocampus, and found a striking increase in cytoplasmic TDP-43 in TDP-43KQ/KQ mice at both 12 and 18 months of age (Fig. 5g-l). The detection of hippocampal cytoplasmic TDP-43 at 12 months (Fig. 5j-l), which is not yet phosphorylated (Fig. 5d, f), suggests that cytoplasmic mislocalization occurs prior to TDP-43 phosphorylation86,87. Overall, our findings indicate that phosphorylated mislocalized TDP-43 coincides with neurodegeneration in TDP-43KQ/KQ mice, consistent with progressive FTLD-TDP44.
Disease-linked transcriptomic and splicing defects in acetylation-mimic TDP-43KQ/KQ mice
TDP-43 acetylation drives RNA dissociation and loss of TDP-43 function52, implying that reduced RNA-binding capacity may impact transcriptional regulation and mRNA splicing36,50,88. We performed total RNA sequencing on neocortex and hippocampus tissue from 18-month-old TDP-43wt or TDP-43KQ/KQ mice to determine how TDP-43 acetylation affects RNA profiles in vivo. We identified nearly 400 differentially expressed genes (DEGs) in each brain region in TDP-43KQ/KQ mice compared to TDP-43wt, after correcting for underlying batch effects. As expected by acetylation-induced loss of TDP-43 function and subsequent autoregulation, the Tardbp transcript was increased in both the hippocampus and neocortex (Supplementary Tables S1, S2). Follow-up RT-qPCR analysis confirmed a 2- to3-fold increase in Tardbp expression in neocortex and hippocampus tissue (Fig. S7), which correlates with increased TDP-43 protein levels in these regions (Fig. 5b, e).
We then clustered the DEGs based on their up- or down-regulation and the brain region affected (Fig. 6), which revealed similar, but distinct, patterns of transcriptional alterations (Fig. 6, Fig. S8 Supplementary Tables S1, S2). To investigate the potential biological implications of the altered transcriptome, we performed Gene Ontology (GO) term enrichment analyses on DEGs identified in each of the six clusters, which revealed that similar pathways were affected in cortical and hippocampal tissues (Supplementary Table S3). In both brain regions, the most highly downregulated genes were involved in developmental processes, including many related to CNS development and maintenance, such as neurogenesis (e.g., Sema5b, Rnd2, Brinp1), gliogenesis (e.g., Tlr2, Olig2, Sox10), and myelination (e.g., Nkx2-2, Nkx6-2, Sox10). Sema5b was the most dramatically reduced transcript in the hippocampus, and the third most in the cortex, (Supplementary Tables S1, S2) with a two-fold reduction in expression in TDP-43KQ/KQ mice. Downregulated genes in both the hippocampus and neocortex were also enriched for terms related to synapse homeostasis and transmembrane signaling, however the dysregulated pathways were distinct. Genes related to GABAergic synapses (e.g., Gad1, Gad2, Abat, Gnb5) were selectively downregulated in the neocortex, while trans-synaptic signaling and ion transport mechanisms (e.g., Homer3, Camk4a, Nsmf, Cnih2) were decreased in the hippocampus.
In contrast, there was significant upregulation of cellular stress response genes (Sesn1, Nrros, Plat, Klf15) and many apoptotic regulators (e.g., Trp53inp1, Pmaip1, Bcl2l1, Plekhf1) in both the hippocampus and neocortex, along with an over-representation of GO terms related to metabolism, localization, cell adhesion (Fig. 6, Supplementary Table S3). Several pathways were uniquely altered, however, suggesting regionally-specific transcriptional effects of acetylation-mimic TDP-43 in TDP-43KQ/KQ mice. For example, coagulation and complement cascades were only upregulated in the hippocampus (e.g., F3, Plat, CD59a). Intriguingly, while genes associated with trans-synaptic signaling were decreased in the hippocampus, another set of genes involved in this same pathway was upregulated in the cortex (e.g., Syt7, Synpo, Nptx1, Spg11) (Supplementary Table S3).
We next sought to draw parallels between the TDP-43KQ/KQ mouse and the human FTLD-TDP transcriptome. Comparison of the DEGs in TDP-43KQ/KQ mice to the mouse orthologs of those found in FTLD-TDP frontal or temporal cortex tissue89 revealed marked overlap between our mouse and their human data sets (Fig. 6, see “FTLD-TDP Cortex” bar), particularly in the hippocampus (Supplementary Table S4; p = 0.0003 "Hippocampus Down" vs Downregulated in FTLD-TDP frontal cortex; p = 0.0014 "Hippocampus Down" vs Downregulated in FTLD-TDP temporal cortex; p = 0.0048 "Hippocampus Up" vs Upregulated in FTLD-TDP frontal cortex). A similar alignment comparing DEGs in mouse striatum following TDP-43 knockdown90 also identified commonly altered genes, particularly in the TDP-43KQ/KQ downregulated gene sets (Fig. 6, see “TDP-43 knockdown” bar; Supplementary Table 4; p = 0.019 "Hippocampus Down" vs Downregulated in TDP-43-KD; p = 0.046 "Cortex Down + Hippocampus Down" vs Upregulated in TDP-43-KD). Together, this data defines distinct functional signatures including altered synaptic gene expression and stress response signaling that are reflective of acetylation-induced TDP-43 dysfunction and the progression of FTLD-TDP.
Alternative splicing defects, particularly impaired repression of cryptic exons, due to TDP-43 dysregulation are implicated in FTLD and ALS pathogenesis91–96. In line with our findings above that TDP-43KQ/KQ primary cortical neurons in vitro show splicing deficits (Fig. 2e-f), we identified widespread splicing alterations in vivo. Analysis of TDP-43KQ/KQ mouse neocortex samples identified 289 differentially spliced genes (DSGs), with 81.7% of loci containing at least one cryptic splice junction, and 29.8% containing two cryptic splice junctions (Supplementary Table S5). In the hippocampus, we found 126 DSGs, 77.0% of which contain a cryptic splice junction and 41.3% that are formed by two cryptic splice sites (Supplementary Table S6). The alternative splicing events were relatively consistent between brain regions, as over 70% of the DSGs identified in the hippocampus were also present in the cortex. Among the most significant DSGs identified were several known TDP-43 splicing targets (e.g., Kcnip2, Pdp1, Poldipp3, Ppfibp1, Dnajc5. Tmem2, Sort1)24,90,92, transcripts associated with particular neurodegenerative diseases (e.g., Mapt, Atxn1, Lrrk2)97–103, and also many robustly altered transcripts that are not well-characterized but have been linked to neurodegeneration (e.g., Nrxn3, Nos1, Arfgef2, Arhgap10, Lrp8, Smarca4, Rims2). Of all identified DSGs, the most substantially altered transcript was Sort1, encoding the Sortilin-1 (SORT1) protein. Exclusion of an exon toward the 3’ end of the Sort1 transcript was reduced by 55.9% in TDP-43KQ/KQ cortex and by 57.0% in the hippocampus (Fig. 7a, b).
SORT1 is a highly expressed neurotrophic factor receptor that binds progranulin (PGRN) and regulates endosomal/lysosomal function through a pathway that is genetically linked to FTLD-TDP104–108. Recent studies suggest that TDP-43 depletion results in inappropriate inclusion of the Sort1 exon 17b and production of a soluble and putatively toxic SORT1 variant that is increased in FTLD-TDP patients90,109,110. To confirm that the altered splicing in our sequencing data was Sort1 exon 17b inclusion, we performed qPCR using primers specific for the mouse Sort1 + ex17b transcript, the appropriately spliced variant (Sort1-WT), and all Sort1 variants (Sort1 total) on tissues isolated from TDP-43KQ/KQ or WT mice110. This analysis showed a nearly 8-fold increase in the Sort1 + ex17b transcript in the cortex (Fig. 7c) and hippocampus (Fig. 7d) of both 12- and 18-month-old animals, which is accompanied by an approximately 25% decrease in Sort1-WT transcripts. Interestingly, the total level of Sort1 transcript varied with age, with slightly increased levels at 12 months and reduced levels at 18 months of age, suggestive of a negative feedback mechanism regulating Sort1 expression.
To determine whether the Sort1 + ex17b transcript identified in TDP-43KQ/KQ mice generates a distinct SORT1 protein, we immunoblotted hippocampus and cortex homogenates and identified a higher molecular weight SORT1 in TDP-43KQ/KQ compared to WT mice (Fig. 7e, f), consistent with Sort1 exon 17b inclusion. Notably, this abnormal SORT1 protein showed decreased stability in TDP-43KQ/KQ mice, as suggested by reduced protein levels in acetylation-mimic animals, compared to controls (Fig. 7g, h). These data indicate failure to repress cryptic exon inclusion and an altered SORT1-PGRN axis, indicating an aberrant splicing pattern in TDP-43KQ/KQ mice that faithfully resembles human FTLD-TDP.