https://orcid.org/0000-0001-5490-1697
Abstract
Responding effectively to external stress is crucial for neurons. Defective stress granule dynamics has been hypothesized as one of the pathways that renders motor neurons in amyotrophic lateral sclerosis (ALS) more prone to early death. Specifically, it is thought that stress granules seed the cytoplasmic TDP-43 inclusions that are observed in the neurons of most ALS patients, as well as ~50% of all frontotemporal dementia (FTD) patients. In this study, we tested this hypothesis in an intact mammalian nervous system. We established an in vivo heat stress paradigm in mice that effectively triggers the eIF2α pathway and the formation of stress granules in the CNS. In non-transgenic mice, we report an age-dependent decline in the formation of heat-induced stress granules, with 18-month-old animals showing a significant impairment. Furthermore, although neuronal stress granules were robustly observed in non-transgenic mice and SOD1G93A mice, they were largely absent in age-matched TDP-43M337V animals. The observed defect in stress granule formation in TDP-43M337V mice correlated with deficits in expression of key protein components typically required for phase separation. Lastly, while TDP-43 was not localized to stress granules, we observed complete nuclear depletion of TDP-43 in a subset of neurons, with the highest proportion being in the TDP-43M337V mice. Overall, our results indicate that mutant TDP-43 expression is associated with defective stress granule assembly and increased TDP-43 nuclear depletion in the mammalian nervous system, which could be relevant to ALS/FTD pathogenesis.
Introduction
The RNA-binding protein (RBP) TDP-43 is at the center of amyotrophic lateral sclerosis (ALS) pathology. In the vast majority of ALS cases, TDP-43 is predominantly cytoplasmic and is frequently observed as aggregated in post-mortem cases (1). This TDP-43 pathology occurs not only in ALS, but also in ~50% of patients with frontotemporal dementia (FTD), and in limbic-predominant age-related TDP-43 encephalopathy [LATE; (2)]. As an RBP, TDP-43 is involved in many aspects of RNA metabolism, such as splicing, mRNA stability and transport, which are misregulated in neurodegenerative diseases involving TDP-43 proteinopathies (3). In addition, TDP-43 can modulate stress granule dynamics (4). Stress granules are biomolecular condensates that form in the cytoplasm by liquid–liquid phase separation (LLPS) in response to an external insult. These transient structures are primarily composed of RBPs and polyadenylated mRNA, and their composition and dynamics are highly variable based on the nature and duration of the stress (5,6). Many stress granule associated proteins are nuclear, and translocate to the cytosol during stress. In fact, persistent stress leads to a reorganization of the mRNA and RBP network (7), which is likely in part because of the formation of stress granules in the cytoplasm. Our understanding of stress granule biology has expanded greatly in recent years, however many of the studies relating to stress granules have been done using in vitro model systems. Few studies have examined stress granule formation in vivo.
Stress granules have gained increasing attention in ALS because of the prevailing hypothesis that their formation leads to persistent granules that transition from a liquid phase to a solid structure, which could act as seeds for the TDP-43 inclusions observed in patients (8,9). Evidence supporting this includes the observation that several RBPs, which localize to stress granules in certain in vitro contexts, have been reported as aggregated in ALS-post-mortem tissue (10–12). However, mRNA, which is an obligate component of stress granules, is not detected in TDP-43 inclusions in ALS/FTD patients (13). Furthermore, recent studies in cell models indicate that TDP-43 aggregates and stress granules form independently (13,14). Therefore, an alternative hypothesis is that, in the context of disease, compromised stress granule formation may contribute to increased neuronal vulnerability. In support of this alternate view, in vitro data, in cells exposed to sodium arsenite, indicate that knockdown of TDP-43 with a silencing RNA leads to sub-optimal stress granule formation, with granules having a significantly reduced size, and compromised viability (4,15,16). Similarly, patient fibroblasts bearing the TDP-43A382T mutation and neuronal cells expressing only cytoplasmic TDP-43 (TDP-43ΔNLS) also have a defect in stress granule formation when exposed to arsenite (17,18). More recently, primary mouse motor neurons or ESC-derived motor neurons bearing the TDP-43M337V mutation have been shown to have defective stress granule assembly (19). One potential reason for these observed deficits in stress granule formation is that depletion of nuclear TDP-43 is associated with the loss of G3BP1—a core stress granule protein (20). Together, these studies demonstrate that TDP-43 mutations expressed at near physiological levels invoke phenotypes similar to those induced by the loss of nuclear TDP-43, suggesting that with respect to stress granule assembly, TDP-43 mutations are loss of function.
In this study, we test how stress granule formation is affected in an intact mammalian nervous system. Specifically, stress granule assembly is assessed in the lumbar spinal cord neurons of mice exposed to full body heat stress. We report an age-dependent decline in stress granule assembly in wild-type mice. In addition, we find stress granule assembly to be significantly impaired in mice bearing the TDP-43M337V mutation, whereas it is unaffected in the SOD1G93A mouse model. Our data suggest that altered stoichiometry between RNA and RBPs in the TDP-43M337V mouse model leads to the observed failure in stress granule assembly. Lastly, we observe neurons with TDP-43 nuclear depletion more frequently in mice that are unable to form stress granules.
Results
Hyperthermia as an in vivo inducer of stress granule assembly
Stress granules have been under the spotlight as a potential mechanism driving ALS pathogenesis. Although significant insights have been gained from work in transformed cell lines, few studies have explored the stress granule response in neurons or in an intact mammalian nervous system. From a machine-learning study using iPSC-derived motor neurons, heat stress was predicted to be the stressor that was the most physiologically relevant to ALS (21). Thus, to study stress granules in vivo, we established a whole-body hyperthermia stress paradigm by subjecting animals to a 20 min heat shock at 44°C [Fig. 1A; (22,23)]. Hyperthermia led to an average change in internal body temperature of 7.9°C in 4-month-old, non-transgenic (NTg) mice (Supplementary Material, Fig. S1A). We observed a steady increase in heart rate and a slight decrease in oxygen saturation over the course of hyperthermia, as expected [Fig. 1B; (22)]. Other mouse models tested displayed similar physiological parameters during the treatment (Supplementary Material, Fig. S1). In control mice that were kept anesthetized at room temperature, we observed a decrease in internal body temperature, in line with the anesthesia, and minor fluctuations in heart rate and oxygen saturation (Supplementary Material, Fig. S1).
In vivo hyperthermia leads to physiological and molecular stress response. (A) Schematic of the hyperthermia stress, which was performed with the following mice: non-transgenic (NTg; 3–4 months, 12 months and 18 months), TDP-43M337V (3–4 months) and SOD1G93A (3–4 months). (B) Sample physiological reading from one control and one hyperthermia mouse (NTg, 4 months). Data points were obtained every 0.3 s for the 20 min stress duration. (C) Representative fluorescence images from NTg mice in lumbar spinal cords stained for either eIF2α or P-eIF2α. Scale bar, 20 μm.
As eIF2α-dependent translation levels should rapidly decrease with activation of the stress response (24), we first looked for phosphorylation of eIF2α at S51 in lumbar spinal cord. This phosphorylation event strongly inhibits the activity of the guanine exchange factor eIF2B, leading to a rapid decrease of Met-tRNA complexes and an overall downregulation of protein synthesis (25,26). Using antibodies specific for either eIF2α or its phosphorylated form, we observed a qualitative increase in phosphorylated eIF2α in the lumbar spinal motor neurons of hyperthermic animals, whereas the total levels of eIF2α remained similar (Fig. 1C). This supports that this hyperthermia paradigm effectively triggers a stress response in lumbar neurons in vivo.
Formation of stress granules in lumbar spinal cord neurons with hyperthermia
In cultured cell models, phosphorylation of eIF2α is recognized as the first step leading to the formation of canonical stress granules under a variety of stresses. Although the protein composition of stress granules is dependent on the external insult (27,28), polyadenylated mRNA is an obligate component of these condensates (29). Therefore, we first performed fluorescence in situ hybridization (FISH) with an oligo-dT probe to assess stress granule formation. In mice that underwent hyperthermia, we observed a clear partitioning of cytosolic mRNA that resembled stress granules in lumbar neurons (Fig. 2A). Although <10% of neurons from control mice displayed granules, hyperthermic mice exhibited stress granules in over 80% of neurons (Fig. 2B). This response was consistent between mice even though the temperature change over the duration of the hyperthermia was varied (Fig. 2B). Unlike patterns reported in vitro, the observed mRNA redistribution was more subtle and sometimes not in clearly defined puncta, thus making it more difficult to quantify the number of individual granules. For this reason, we also calculated a granularity index using MatLab, which is premised on Sobel edge detection, to measure stress granule formation. Using this index, which avoids bias for a heterogeneous cell population, we also observed a significant increase in cytosolic mRNA granularity in the neurons of mice subjected to hyperthermia compared with sham control mice (Fig. 2C). In addition, we noted that the granularity index was robust despite minor fluctuations in total temperature change within animals, suggesting that all mice have reached the temperature threshold to form granules in this paradigm (Fig. 2C). With polyadenylated mRNA labelling, we also observed larger puncta in the nucleus in both control and hyperthermia conditions (Fig. 2A). Analysis of these puncta showed no difference in their size or number between control and hyperthermia mice (Supplementary Material, Fig. S2). These puncta could represent Cajal bodies or nuclear speckles, however we do not think they represent paraspeckles, which would have been expected to increase with stress (30,31).
Stress granule formation in lumbar neurons following hyperthermia. (A) Representative fluorescence in situ hybridization (FISH) images of lumbar spinal cord neurons using oligo-dT probe from control and hyperthermia NTg mice. Yellow arrows highlight stress granules. Scale bar, 10 μm. (B) Measurement of % cells with stress granules by observing individual neurons from control and hyperthermia mice. Control N = 5, n = 41; Hyperthermia N = 6, n = 50. **** P < 0.0001. Graph comparing the temperature change to cells with stress granules. R2 = 0.1767, P = 0.4810. (C) Measurement of granules between control and hyperthermia conditions using the granularity index. Control N = 6, n = 76; Hyperthermia N = 5, n = 89. Unpaired t-test, **** P < 0.0001. Graph comparing the temperature change and granularity in individual mice. R2 = 0.03967, P = 0.0935. (D) Representative images of G3BP2, eIF3η, TiaR and HuR staining in lumbar neurons from control and hyperthermic mice. Scale bar, 20 μm. (E) Representative images of neurons from control or hyperthermia mice labelled with G3BP2 and TiaR. Scale bar, 20 μm.
To identify key protein components of these stress-induced mRNA granules, we labelled spinal cords for common protein markers of stress granules. Several stress granule markers, including G3BP2, eIF3η, TiaR and HuR, exhibited a granular pattern following hyperthermic stress (Fig. 2D). Interestingly, the staining patterns for these RBPs varied. eIF3η and TiaR granules were smaller and more numerous, whereas G3BP2 and HuR granules were larger and less abundant, suggesting variable protein composition of individual granules in a single neuron. We also observed a redistribution of TiaR from the nucleus to the cytosol with hyperthermia in neurons, which has been reported previously (32,33). Moreover, a partial colocalization between G3BP2 and TiaR was observed (Fig. 2E). To identify whether G3BP1—a central regulator of stress granule assembly—was also a constituent of these granules, we labelled spinal cords with four commercial antibodies that target different domains of the G3BP1 protein (Supplementary Material, Fig. S3). Although G3BP1 has previously been shown to localize to heat-induced stress granules in cultured cells (34), we were unable to conclusively detect G3BP1-positive granules that correlated specifically with hyperthermia. In addition, we stained spinal cords for six other reported stress granule protein markers, but similar to G3BP1, these antibody signals did not show clear recruitment to granular structures during hyperthermia and were also inconclusive (Supplementary Material, Fig. S4). Overall, hyperthermia treatment in mice leads to the formation of stress granules in lumbar neurons that contain mRNA and a subset of well-defined stress granule proteins.
Formation of stress granules in vivo is compromised with aging
Because aging is the biggest risk factor associated with ALS and other neurodegenerative diseases, we first looked at stress granules in vivo in aged non-transgenic mice. Our previous data in cultured primary motor neurons indicate that stress granule formation is less efficient in aged neurons (16). In addition, it has been demonstrated that senescent cells have difficulty forming stress granules (35). To assess whether initiation of the stress response and formation of stress granules proceeded normally in vivo in aged animals, we performed hyperthermia stress with mice at 12 and 18 months of age. Interestingly, although S51 phosphorylation of eIF2α was increased ~5-fold after hyperthermia in the spinal cord of 4-month-old mice, we observed a weakened response in the older mice (~2-fold at 12 months, ~3-fold at 18 months; Fig. 3A and B). Next, we looked at the formation of stress granules using polyadenylated mRNA labelling. Similar to animals at 4 months of age, stress granule formation proceeded normally at 12 months despite the reduction in P-eIF2α (Fig. 3C and D). However, at 18 months of age, stress granule formation was significantly compromised in lumbar neurons. Thus, in 12-month-old mice, P-eIF2α activation is compromised but stress granule formation proceeded normally, whereas both eIF2α phosphorylation and stress granule formation are deficient in 18-month-old mice.
Aged mice display time dependent defect in stress granule formation following hyperthermia. (A) Western blots of eIF2α and P- eIF2α from spinal cord lysates of control or hyperthermia mice at different ages. (B) Quantification of western blots from (A) relative to loading control tubulin. 4 months: control N = 6, hyperthermia N = 6; 12 months: control N = 3, hyperthermia N = 3; 18 months: control N = 3, hyperthermia N = 3. Unpaired t-test between control and hyperthermia for each genotype, * P < 0.05. (C) Representative FISH images from control and hyperthermia mice at 3–4, 12 and 18 months of age in lumbar neurons. Scale bar, 10 μm. (D) Quantification of stress granules by cells with stress granules and granularity index between control and hyperthermia mice. Data for the 4-month mice is the same as in Figure 2. Cells with SG: 4 months N = 6, n = 50; 12 months N = 3, n = 32, 18 months N = 3, n = 30. Granularity index: 4 months N = 6, n = 89; 12 months N = 3, n = 57; 18 months N = 3, n = 56. Kruskal–Wallis followed by multiple comparisons test was used to determine significance. *** P < 0.001, **** P < 0.0001.
TDP-43 M337V mice have a defective stress granule response
We next looked to see whether mice with ALS-causing mutations were able to effectively launch the stress response. It has previously been shown that in tau transgenic mice at 8 months of age, hyperthermic stress is associated with a defect in stress granule formation in cortical and hippocampal neurons (23). To analyze the stress response in the context of ALS, we looked at the SOD1G93A mouse model (36) and the TDP-43M337V mouse model (19). We selected the SOD1G93A mouse model because it has been reported that mutant SOD1 can interact with G3BP1 and disrupt stress granule dynamics (37). On the other hand, we chose homozygote TDP-43M337V mice because they express only two additional copies of human TDP-43M337V in addition to the mouse TDP-43 (Supplementary Material, Fig. S5A). This low-level human TDP-43 expression is important when looking at biomolecular condensates since TDP-43 expressed at higher levels is prone to aggregation and therefore could interfere with our analysis. As previously reported, YPet-tagged TDP-43M337V localizes to both the nucleus and the cytoplasm (Supplementary Material, Fig. S5B). Although TDP-43 aggregates have been detected in mice with high overexpression of the TDP-43M337V mutation, the TDP-43M337V mice used in our study do not show cytoplasmic TDP-43 aggregation (19,38). At 4 months of age, when we perform our in vivo stress paradigm, these mice do not have detectable changes in hang time, grip strength or weight compared with non-transgenic littermates (Supplementary Material, Fig. S5C and E). Published work with these TDP-43M337V mice indicate defects in endosomal trafficking, as well as differences in splicing of common TDP-43 targets (39,40). Importantly, in vitro study of ESC-derived motor neurons from this mouse model have been demonstrated to have defective stress granule assembly and an altered stress granule interactome in response to oxidative stress (19,41).
Analysis of eIF2α phosphorylation showed that both TDP-43M337V and SOD1G93A mice have increased levels of P-eIF2α after hyperthermia (Fig. 4A). Specifically, TDP-43M337V mice show a ~4-fold increase and SOD1G93A mice show a ~3-fold increase in P-eIF2α signal in spinal cord lysates after hyperthermia, respectively (Fig. 4B). Surprisingly, we found that in TDP-43M337V homozygote mice, stress granule assembly was disrupted as measured by polyadenylated mRNA labelling. However, the formation of heat-induced stress granules in SOD1G93A mice proceeded normally (Fig. 4C and D). Analysis of 4-month heterozygous TDP-43M337V mice and 12-month homozygous TDP-43M337V mice also showed no stress granule formation after hyperthermia treatment (Supplementary Material, Fig. S6). We further investigated whether protein markers that robustly labelled stress granules in non-transgenic mice were affected in TDP-43M337V mice. Spinal cords of presymptomatic TDP-43M337V mice that underwent hyperthermia treatment showed, similar to polyadenylated mRNA, that the protein markers TiaR and HuR failed to produce a cytosolic granular pattern in response to stress (Fig. 4E). Taken together, these results suggest that the expression of mutant TDP-43, but not mutant SOD1, leads to impaired stress granule formation.
TDP-43M337V mice have a defect in stress granule formation following acute hyperthermia. (A) Western blots of eIF2α and P-eIF2α from spinal cord lysates of NTg, TDP-43M337V and SOD1G93A mice from control or hyperthermia conditions. The blot for the NTg mice is replicated from Figure 2. (B) Quantification of western blots from (A) relative to loading control tubulin. NTg control N = 6, hyperthermia N = 6; TDP-43M337V control N = 4, hyperthermia N = 5 and SOD1G93A control N = 7, hyperthermia N = 7. Unpaired t-test between control and hyperthermia for each genotype. Data for NTg is the same as Figure 3. (C) Representative images with oligo-dT probe from sham and hyperthermia-exposed NTg, TDP-43M337V and SOD1G93A mice. Scale bar, 10 μm. (D) Quantification of cells displaying stress granules and granularity index following hyperthermia. The NTg mice are the same group used in Figure 2 and 3. Cells with SG: NTg N = 6, n = 50, TDP-43M337V N = 5, n = 46 and SOD1G93A N = 5, n = 46. Granularity index: NTg N = 6, n = 89; TDP-43M337V N = 5, n = 57 and SOD1G93A N = 3, n = 43. Kruskal–Wallis test with multiple comparisons, * P < 0.05, ** P < 0.01. (E) Representative images of TiaR and HuR staining in TDP-43M337V mice that underwent control or hyperthermia treatment. Scale bar, 20 μm.
Basal levels of cytosolic mRNA and RBPs are decreased in TDP-43 M337V mice
Although we saw a clear defect in stress granule assembly in TDP-43M337V mice, this was not accompanied by a loss of eIF2α phosphorylation. Lack of stress granules in the presence of decreased translation could indicate a defect in LLPS. For LLPS to occur, cells must have an adequate concentration of RBPs and RNA, which both contribute to this process (42,43). When performing FISH to examine polyadenylated RNA in TDP-43M337V mice, we noticed an overall lower cytoplasmic signal intensity compared with non-transgenic mice (Fig. 2A and4C). Quantification of cytosolic mRNA intensity of these images revealed that the TDP-43M337V mRNA signal was 23% less than of that in non-transgenic mice (Fig. 5E). Intriguingly, SOD1G93A mice, which did not show a defect in stress granule assembly, displayed cytosolic mRNA levels comparable with non-transgenic mice. We also measured the protein levels of eIF4A1, which is described as a major RNA chaperone that prevents RNA–RNA condensation and limits the recruitment of mRNAs to stress granules (44). We found that eIF4A1 levels were increased in TDP-43M337V mouse spinal cords, but not in SOD1G93A, compared with non-transgenic mice (Fig. 5A–D).
Decreased level of RBPs and cytosolic mRNA in untreated TDP-43M337V mice. (A-B) Western blots and quantification of eIF4A1, TiaR, HuR and G3BP1 in spinal cord lysates from NTg and TDP-43M337Vmice at 3–4 months of age. N = 5–7 per group. Tubulin was used as the loading control for quantification. (C-D) Western blots and quantification on the same proteins as A in SOD1G93A animals. N = 4–8 per group. Kruskal–Wallis followed by multiple comparisons test determined significance. * P < 0.05, ** P < 0.01. (E) Quantification of cytoplasmic oligo-dT FISH signal in NTg, TDP-43M337V or SOD1G93A mice. NTg N = 4, n = 60 TDP-43M337V N = 4, n = 70; SOD1G93A N = 4, n = 75. Kruskal–Wallis followed by multiple comparisons test determined significance. **** P < 0.0001. (G-H) Measurement of TiaR or HuR signal in neurons from 3- to 4-month old NTg or TDP-43M337V mice. TiaR: NTg N = 5, n = 103; TDP-43M337V N = 5, n = 106. HuR: NTg N = 5, n = 96; TDP-43M337V N = 5, n = 94. Unpaired t-test to determine significance. * P < 0.05, **** P < 0.0001.
Because protein–protein and protein-RNA interactions also contribute to stress granule formation, we measured the levels of TiaR and HuR in the spinal cords of non-transgenic, TDP-43M337V and SOD1G93A mice, since these proteins were observed as localized to stress granules in lumbar neurons (Fig. 1D). Similar to polyadenylated mRNA, protein levels of both TiaR and HuR were decreased in TDP-43M337V mice, but not in SOD1G93A mice (Fig. 5A–D; TiaR was actually increased in SOD1G93A mice). We also observed a significant decrease of G3BP1 protein levels in TDP-43M337V mice (Fig. 5A–D). Lastly, we analyzed the levels of cytoplasmic TiaR and HuR in confocal images of motor neurons from non-transgenic and TDP-43M337V mice by immunofluorescence. We found that similar to spinal cord lysates, cytoplasmic TiaR and HuR levels were decreased in the neurons of TDP-43M337V mice (Fig. 5F and G). Collectively, the decreased levels of RBPs and cytosolic mRNA coupled with the increase in eIF4A1 in TDP-43M337V mice suggest that the components necessary for LLPS-driven stress granule formation are insufficient to achieve the appropriate percolation threshold for assembly of these structures.
Acute hyperthermia induces TDP-43 cytoplasmic localization, but not aggregation. Western blots (A) and quantification (B) for TDP-43 of spinal cord lysates from control and hyperthermia mice normalized to tubulin. (C) Representative TDP-43 images in neurons in control and hyperthermia 4-month-old non-transgenic mice. Neurons with TDP-43 nuclear depletion have nuclei circled in yellow. TDP-43 perinuclear accumulations are highlighted with yellow arrows. Scale bar, 20 μm. (D) Representative TDP-43 images from lumbar spinal cords of 3–4-month-old NTg, TDP-43M337V and SOD1G93A mice. Scale bar, 50 μm. (E) Quantification of lumbar spinal neurons with nuclear TDP-43 depletion during hyperthermia in different genotypes. Each point represents an individual mouse. NTg N = 4, n = 230; TDP-43M337V N = 4, n = 244; SOD1G93A N = 4, n = 187. The Kruskal–Wallis test was used to determine significance, * P < 0.05. (F) Representative images of total TDP-43, hTDP-43 M337V (YPet) in a neuron that shows TDP-43 nuclear loss. Scale bar 20 μm.
TDP-43 nuclear depletion in a subset of neurons after hyperthermia
TDP-43 is known to regulate the assembly and dynamics of stress granules (15,16), but is not considered a core stress granule protein (45) and is thus not always found to be localized to these structures (13,14). In our hyperthermia model, total spinal cord protein levels of TDP-43 were not significantly changed between control and hyperthermic conditions (Fig. 6A and B). Interestingly, although TDP-43 was not found to be localized to stress granules in any of the ALS mouse models studied, we did occasionally observe neurons showing complete nuclear depletion of TDP-43 and diffuse cytoplasmic labelling (Fig. 6C). In some of the neurons with complete loss of the nuclear pool, we observed a perinuclear clustering of TDP-43 in the cytoplasm. Given their singularity within cells, these structures cannot be considered as stress granules (Fig. 6C, yellow arrows). Since TDP-43 nuclear depletion is often correlated with cytoplasmic TDP-43 aggregation, and it is hypothesized that stress granules may give rise to TDP-43 inclusions, we further investigated TDP-43 nucleocytoplasmic localization following hyperthermia in TDP-43M337V and SOD1G93A mice (Fig. 6D and E). Although we did not observe TDP-43 mislocalization in control mice, >25% of spinal neurons in hyperthermia-exposed TDP-43M337V mice displayed a complete loss of nuclear TDP-43 concomitant with diffuse cytoplasmic labelling. Using a GFP antibody to recognize the YPet tag on human TDP-43, we demonstrated that in neurons with this phenotype, both the mouse and human TDP-43 are depleted from the nucleus (Fig. 6F). In contrast, in non-transgenic and SOD1G93A mice, where stress granule formation proceeded normally, few neurons displayed TDP-43 nuclear depletion. These results suggest that TDP-43 nucleocytoplasmic distribution in lumbar neurons within an intact mammalian nervous system can be altered by external stress and is unrelated to stress granule formation.
Discussion
ALS onset is hypothesized to be a multistep process in which genetic susceptibility interacts with aging and environmental factors to cause motor neuron degeneration (46). In this study, we have established a model for studying stress granule formation within an intact mammalian nervous system using hyperthermia. With this model, we were able to address how an environmental insult—in this case a 20 min heat exposure—impacts stress granules in a time-dependent or a gene-dependent manner. This work effectively addresses how time, genetics and environmental stress components impact the stress response in the CNS of mice in the context of ALS. We demonstrate that stress granule assembly in lumbar spinal cord neurons is disrupted in aged 18-month-old wild-type mice, as well as in 4-month-old, presymptomatic mice bearing two copies of the TDP-43M337V mutation. Our work highlights that the stress granule defect described here precedes motor deficits.
The general appearance of neuronal stress granules that we observed in this study is quite different compared with what is reported for in vitro models. In vitro studies in non-neuronal cell lines exhibit stress granules that are mostly circular, with nearly complete co-labelling of multiple stress granule protein markers. However, it is published that stress granules in cultured motor neurons are pleomorphic (less circular), and more numerous (16). This variability in shape was also noted in spinal neurons in vivo. A potential explanation for the difference in appearance of stress granules is because of the neuronal RNA composition, which can impact condensate size (47). Alternatively, treatment with anesthetic prior to hyperthermia could impact the stress granule response. Because of this intrinsic morphological heterogeneity as well as noted compositional differences, we opted to use polyadenylated mRNA as a means to quantify stress granules as it is an obligate component of these structures. Although we tested a total of 11 stress granule proteins, only a handful demonstrated a clear granular pattern in hyperthermic neurons. Moreover, we noted that of the markers with conclusive labelling (TiaR, HuR, G3BP2 and eIF3ƞ), there was a variability in staining patterns suggesting non-homogenous protein composition for individual stress granules within a single neuron. The concept of different stress granule proteomes and/or stress granule cores has been suggested in a number of cell-based stress granule studies (6,48,49) but remains to be confirmed in vivo. Our data thus support stress granule heterogeneity in the mammalian nervous system and emphasize that the composition of these structures is dependent on both cell type and the nature of the stress.
When we looked at stress granule formation in aged mice at 12 and 18 months, we noted that although eIF2α phosphorylation was reduced in the mice at both time points, stress granule assembly was impaired only at 18 months. The reduced increase in P-eIF2α could be due to the increased basal levels we see in older mice. In fact, cells that have elevated basal levels of P-eIF2α have been shown to have altered stress granule dynamics (23). We also observed that at 18 months, mice that underwent hyperthermia had a less robust increase in basal body temperature compared with all other genotypes tested. However, by comparing the average percent of cells with stress granules and the granularity index in 4-month-old mice to the increase in body temperature, we are confident that the reason for the stress granule defect is not due to the inability of these 18 months animals to reach a threshold temperature. The absence of stress granule formation could be related to an increase in stress granule disassembly machinery in neurons, as was suggested as an underlying reason for failed stress granule assembly in senescent cells (35). Because stress granules are composed of many aggregation-prone RBPs, it is commonly hypothesized that persistent stress granules leads to aggregate formation with age (50,51). However, our data directly counter this and instead suggest that aged neurons form stress granules less efficiently in vivo. This would align with the report that TDP-43 inclusions observed in ALS/FTD patient neurons are devoid of polyadenylated mRNA (13,14).
In homozygote TDP-43M337V mice, we observed that stress granule assembly is defective in response to hyperthermia. Previous work in a knock-in FUSR521C mouse model of ALS stressed with intragastric delivered arsenite showed Tia-1 positive granules in neurons, but they were absent in non-transgenic littermates (52). However, since stress granules were not detected in arsenite-treated non-transgenic mice, this argues that the Tia-1 positive granules detected in FUSR521C mice may not represent stress granules but instead may be Tia-1 containing protein aggregates. This echoes recent work in cultured cells where the formation of mutant FUS assemblies co-segregate with stress granule-related RBPs (53). Other work in Drosophila shows that after traumatic brain injury, G3BP1 granules are formed and persist (54). However, <2% of these granules overlap with other stress granule markers such as PABP and FMR1. The acute stress protocol we have used does not allow us to examine the long-term persistence of stress granules, and the defect we describe here could differ with an increased stress exposure. In addition, we did not examine the kinetics of stress granule assembly and disassembly in TDP-43M337V mice in this study. However, the fact that mice expressing two additional copies of the TDP-43M337V mutation show a defect in stress granule formation suggests that persistent stress granules are unlikely to form. In contrast, the TDP-43M337V mice have a higher percentage of cells displaying TDP-43 nuclear depletion following heat stress. Together, these data suggest that persistent stress granules and TDP-43 mislocalization are independent of one another in this model. Moreover, TDP-43 aggregation is not observed in TDP-43M337V mice (19). However, we acknowledge that since our model is acute stress, we cannot exclude the possibility that subsequent TDP-43 aggregation could occur following a longer exposure or multiple exposures. In SOD1G93A mice, we observe robust stress granule formation with a low percentage of cells with TDP-43 nuclear depletion. It is noteworthy that TDP-43 pathology is not broadly reported in mutant SOD1 models (55). In the context of SOD1 mutations, we cannot rule out that stress granules may persist, but we anticipate that if this occurs, it is in the absence of TDP-43 mislocalization and subsequent aggregation. Collectively, these data raise important questions regarding the relationship between stress granule formation and TDP-43 pathology.
The molecular mechanism underlying defective stress granule assembly in TDP-43M337V mice remains unclear. Our data suggest that there is an imbalance in the stoichiometry between RBPs and mRNA, both essential components of stress granules, and we posit that this is a contributing factor to the defective assembly we observed in spinal cord neurons in vivo. Although we cannot pinpoint the reason for the changes in RBP and mRNA levels, we speculate that they could be due to alterations in the TDP-43 interactome, as has been reported in TDP-43M337V ESC-derived motor neurons exposed to oxidative stress (41). Protein–protein interactions are essential for stress granule formation (56). In TDP-43M337V mice, levels of TiaR, and G3BP1, which are crucial RBPs that initiate stress granule formation (45,57), and HuR—a well-known stress granule marker—are all reduced. RNA–RNA interactions are also necessary for stress granules to form (43). To this end, we also observed decreased levels of cytoplasmic polyadenylated mRNA and increased levels of the RNA chaperone eIF4A1 in TDP-43M337V mice. Therefore, key components that likely contribute to achieving the percolation threshold for LLPS are altered in these mice, supporting the observed defect in stress granule formation in response to hyperthermia. In contrast, in SOD1G93A mice where levels of RBPs and RNA are comparable with non-transgenic mice, stress granule assembly proceeds normally. From a biophysics perspective, two-component phase-separated systems must be at a precise stoichiometry for optimal phase separation to occur (58,59). This implies that the loss of only one component can lead to many disruptions in the stress granule network with the outcome being that the condensates would not form. As an example, actin polymerization, which also relies on phase separation, is defective when stoichiometry of key components is changed (60).
Historically, stress granules are thought to serve as hubs for mRNA protection during times of cellular stress (42). Thus, based on this proposed function, a defect in stress granule formation, as we observed here, could contribute to increased neuronal vulnerability. Moreover, since TDP-43M337V mice do not exhibit motor symptoms when we observe the stress granule defect, our work suggests that defective stress granule assembly precedes clinical disease onset in this model. Future studies are needed to evaluate whether disrupted stress granule dynamics can influence clinical disease progression and/or symptom onset. Overall, our work in a whole-body hyperthermia model of cell stress demonstrates that the loss of stress granule assembly links aging with TDP-43 mutations and is therefore a point of convergence in the pathogenesis of ALS and other diseases involving TDP-43.
Materials and Methods
Animals
The use of animals and all procedures were performed in according to guidelines of the Canadian Council on Animal Care and were approved by the CRCHUM animal care committee (CIPA). The non-transgenic group of mice were a combination of B6/N mice (Charles River), and non-transgenic littermates from TDP-43M337V and SOD1G93A mice bred in-house. All mice showed similar phenotypes for the hyperthermia protocol and were therefore pooled for analyses.
Hyperthermia treatment
Only males were used for hyperthermia treatment. Mice were anesthetized by 4% isoflurane until they were unresponsive, then injected with 0.2 ml/10 g anesthetic cocktail (1 mg/ml ketamine, 0.5 mg/ml xylazine and 0.1 mg/ml acepromazine). The starting core temperature was taken immediately (PhysioSuite, Kent Scientific) before it decreased due to anesthesia. Mice were placed in an incubator set to 44°C supplemented with 1 l/min O2 and a 20 min timer was started. MouseOx® (Starr Life Sciences, USA) was used to monitor heart rate and oxygen saturation over the duration of hyperthermia. Any mouse that displayed an oxygen saturation of <50% was immediately sacrificed even if treatment was not completed. After hyperthermia, internal body temperature was taken again, and mice were immediately sacrificed. Animals were perfused using 4% PFA and tissues were subsequently cryopreserved, embedded in OCT and sectioned. Tissue collection was performed by dissection and snap freezing tissues in liquid N2. Change in heart rate was measured by taken the highest heart rate in the last 5 min and subtracting the lowest heart rate in the first 5 min. Change in oxygen saturation was calculated by taking the lowest O2 measurement in the last 5 min and subtracting the highest O2 measurement in the first 5 min.
Behavior
All behavior assays were performed on 2 consecutive days. For the hang time test, mice were placed on a wire frame and turned upside-down headfirst. A timer was started and mice were monitored until they fell or until 2 min were surpassed. Three trials, each minimum 15 min apart, were done each day, for a total of six trials. Hanging impulse was calculated by multiplying the average hang time and weight for each mouse. Forelimb grip strength was measured using a grip-strength meter (Bioseb, USA). Five consecutive measurements were taken for each mouse and the average was calculated to determine grip strength. Both males and females were used for behavior testing on the TDP-43M337V mice.
Immunofluorescence (spinal cords) & FISH
Floating sections of lumbar spinal cords (30 μm) were used for staining. For staining with G3BP1 antibodies, staining was done with and without demasking (using 10 mm citrate pH 6.0 for 30 min at 95°C) before blocking. For all other antibodies, sections were washed in Sorenson’s PBS (13-mm NaH2PO4, 87-mm Na2HPO4, pH 7.6) then blocked with 3% BSA followed by incubation of primary antibodies (see table) overnight at 4°C. The following day, spinal cords were washed and incubated with fluorescently conjugated secondary antibodies (1:200) to the desired species, stained with Hoescht and mounted on slides with ProLong Antifade reagent (Invitrogen). See Table 1 for antibody dilutions. For fluorescence in situ hybridization, sections were washed in 5X SSC buffer, acetylated with acetic anhydride (3 μl/ml), permeabilized with Triton-X-100 (0.5% in 5X SSC), then incubated in hybridization buffer (0.1 M TEA pH 8, 25% formamide, 1 mg/ml yeast RNA, 0.01% BSA, 10% dextran sulfate) for 2 h at room temperature. The oligo-dT probe (1.3 ng/μl) was diluted in hybridization buffer then added to spinal cords for incubation overnight at 37°C. The following day, spinal cords were washed in 5X SSC buffer, stained with Hoescht and mounted with ProLong Antifade reagent (Invitrogen). Images were acquired with 20X (1.25 NA) or 63X oil (1.7 NA) objectives on a confocal microscope (SP5; Leica) using LAS-X software (Leica, Germany). Fiji (ImageJ) was used for determining cytoplasmic oligo-dT intensity and cytoplasmic TiaR and HuR intensity. Based on location in the ventral horn, nucleus size and neuronal size, most neurons imaged are presumed to be motor neurons.
Antibodies used in this study. The following antibodies were used at the indicated dilutions in this study
| Antibody | Company | Cat # | Dilution |
|---|---|---|---|
| Actin (Ms) | MP Biomedicals | 69 100 | 1:10 000 (WB) |
| TDP-43 (Rb) | Proteintech | 10 782-2-AP | 1:5000 (WB) 1:200 (IF) |
| TDP-43 (Ms) | Cedarlane | MAB7778 | 1:1000 (WB) 1:250 (IF) |
| GFP (Rb) | Covance | MMS-118P | 1:200 (IF) |
| α-Tubulin (Ms) | Abcam | ab24610 | 1:20 000 (WB) |
| eIF2α (Rb) | Cell Signaling (WB) Bethyl Labs (IF) | 9722 A300-721A | 1:1000 (WB) 1:100 (IF) |
| P-eIF2α (S51) (Rb) | Cell Signaling | 9721 | 1:500 (WB) 1:100 (IF) |
| TiaR (Ms) | BD Transduction Labs | 610 352 | 1:2000 (WB) 1:100 (IF) |
| HuR (Ms) | Santa Cruz | SC5261 | 1:500 (WB) 1:500 (IF) |
| G3BP1 (Ms) | Abcam | ab56574 | 1:200 (IF) |
| G3BP1 (Ms) | BD Transduction Labs | 611 126 | 1:200 (IF) |
| G3BP1 (Rb) | Millipore | 05-1938 | 1:200 (IF) |
| G3BP1 (Rb) | Proteintech | 13 057-2-AP | 1:200 (IF) |
| Caprin1(Rb) | Proteintech | 15 112-1-AP | 1:100 (IF) |
| eIF4G (Rb) | Cell Signaling | 2498 | 1:100 (IF) |
| eIF3η (Ms) | Santa Cruz | SC137214 | 1:50 (IF) |
| FUS (Rb) | Abcam | Ab23439 | 1:100 (IF) |
| G3BP2 (Rb) | Cedarlane | A302-040A | 1:100 (IF) |
| SNRNP200 (Rb) | Abcam | ab176715 | 1:100 (IF) |
| Tia1 (Rb) | Abcam | ab40693 | 1:200 (IF) |
| UBAP2L (Rb) | Abcam | ab138309 | 1:100 (IF) |
| Antibody | Company | Cat # | Dilution |
|---|---|---|---|
| Actin (Ms) | MP Biomedicals | 69 100 | 1:10 000 (WB) |
| TDP-43 (Rb) | Proteintech | 10 782-2-AP | 1:5000 (WB) 1:200 (IF) |
| TDP-43 (Ms) | Cedarlane | MAB7778 | 1:1000 (WB) 1:250 (IF) |
| GFP (Rb) | Covance | MMS-118P | 1:200 (IF) |
| α-Tubulin (Ms) | Abcam | ab24610 | 1:20 000 (WB) |
| eIF2α (Rb) | Cell Signaling (WB) Bethyl Labs (IF) | 9722 A300-721A | 1:1000 (WB) 1:100 (IF) |
| P-eIF2α (S51) (Rb) | Cell Signaling | 9721 | 1:500 (WB) 1:100 (IF) |
| TiaR (Ms) | BD Transduction Labs | 610 352 | 1:2000 (WB) 1:100 (IF) |
| HuR (Ms) | Santa Cruz | SC5261 | 1:500 (WB) 1:500 (IF) |
| G3BP1 (Ms) | Abcam | ab56574 | 1:200 (IF) |
| G3BP1 (Ms) | BD Transduction Labs | 611 126 | 1:200 (IF) |
| G3BP1 (Rb) | Millipore | 05-1938 | 1:200 (IF) |
| G3BP1 (Rb) | Proteintech | 13 057-2-AP | 1:200 (IF) |
| Caprin1(Rb) | Proteintech | 15 112-1-AP | 1:100 (IF) |
| eIF4G (Rb) | Cell Signaling | 2498 | 1:100 (IF) |
| eIF3η (Ms) | Santa Cruz | SC137214 | 1:50 (IF) |
| FUS (Rb) | Abcam | Ab23439 | 1:100 (IF) |
| G3BP2 (Rb) | Cedarlane | A302-040A | 1:100 (IF) |
| SNRNP200 (Rb) | Abcam | ab176715 | 1:100 (IF) |
| Tia1 (Rb) | Abcam | ab40693 | 1:200 (IF) |
| UBAP2L (Rb) | Abcam | ab138309 | 1:100 (IF) |
Antibodies used in this study. The following antibodies were used at the indicated dilutions in this study
| Antibody | Company | Cat # | Dilution |
|---|---|---|---|
| Actin (Ms) | MP Biomedicals | 69 100 | 1:10 000 (WB) |
| TDP-43 (Rb) | Proteintech | 10 782-2-AP | 1:5000 (WB) 1:200 (IF) |
| TDP-43 (Ms) | Cedarlane | MAB7778 | 1:1000 (WB) 1:250 (IF) |
| GFP (Rb) | Covance | MMS-118P | 1:200 (IF) |
| α-Tubulin (Ms) | Abcam | ab24610 | 1:20 000 (WB) |
| eIF2α (Rb) | Cell Signaling (WB) Bethyl Labs (IF) | 9722 A300-721A | 1:1000 (WB) 1:100 (IF) |
| P-eIF2α (S51) (Rb) | Cell Signaling | 9721 | 1:500 (WB) 1:100 (IF) |
| TiaR (Ms) | BD Transduction Labs | 610 352 | 1:2000 (WB) 1:100 (IF) |
| HuR (Ms) | Santa Cruz | SC5261 | 1:500 (WB) 1:500 (IF) |
| G3BP1 (Ms) | Abcam | ab56574 | 1:200 (IF) |
| G3BP1 (Ms) | BD Transduction Labs | 611 126 | 1:200 (IF) |
| G3BP1 (Rb) | Millipore | 05-1938 | 1:200 (IF) |
| G3BP1 (Rb) | Proteintech | 13 057-2-AP | 1:200 (IF) |
| Caprin1(Rb) | Proteintech | 15 112-1-AP | 1:100 (IF) |
| eIF4G (Rb) | Cell Signaling | 2498 | 1:100 (IF) |
| eIF3η (Ms) | Santa Cruz | SC137214 | 1:50 (IF) |
| FUS (Rb) | Abcam | Ab23439 | 1:100 (IF) |
| G3BP2 (Rb) | Cedarlane | A302-040A | 1:100 (IF) |
| SNRNP200 (Rb) | Abcam | ab176715 | 1:100 (IF) |
| Tia1 (Rb) | Abcam | ab40693 | 1:200 (IF) |
| UBAP2L (Rb) | Abcam | ab138309 | 1:100 (IF) |
| Antibody | Company | Cat # | Dilution |
|---|---|---|---|
| Actin (Ms) | MP Biomedicals | 69 100 | 1:10 000 (WB) |
| TDP-43 (Rb) | Proteintech | 10 782-2-AP | 1:5000 (WB) 1:200 (IF) |
| TDP-43 (Ms) | Cedarlane | MAB7778 | 1:1000 (WB) 1:250 (IF) |
| GFP (Rb) | Covance | MMS-118P | 1:200 (IF) |
| α-Tubulin (Ms) | Abcam | ab24610 | 1:20 000 (WB) |
| eIF2α (Rb) | Cell Signaling (WB) Bethyl Labs (IF) | 9722 A300-721A | 1:1000 (WB) 1:100 (IF) |
| P-eIF2α (S51) (Rb) | Cell Signaling | 9721 | 1:500 (WB) 1:100 (IF) |
| TiaR (Ms) | BD Transduction Labs | 610 352 | 1:2000 (WB) 1:100 (IF) |
| HuR (Ms) | Santa Cruz | SC5261 | 1:500 (WB) 1:500 (IF) |
| G3BP1 (Ms) | Abcam | ab56574 | 1:200 (IF) |
| G3BP1 (Ms) | BD Transduction Labs | 611 126 | 1:200 (IF) |
| G3BP1 (Rb) | Millipore | 05-1938 | 1:200 (IF) |
| G3BP1 (Rb) | Proteintech | 13 057-2-AP | 1:200 (IF) |
| Caprin1(Rb) | Proteintech | 15 112-1-AP | 1:100 (IF) |
| eIF4G (Rb) | Cell Signaling | 2498 | 1:100 (IF) |
| eIF3η (Ms) | Santa Cruz | SC137214 | 1:50 (IF) |
| FUS (Rb) | Abcam | Ab23439 | 1:100 (IF) |
| G3BP2 (Rb) | Cedarlane | A302-040A | 1:100 (IF) |
| SNRNP200 (Rb) | Abcam | ab176715 | 1:100 (IF) |
| Tia1 (Rb) | Abcam | ab40693 | 1:200 (IF) |
| UBAP2L (Rb) | Abcam | ab138309 | 1:100 (IF) |
Immunoblotting
Spinal cords were weighed, resuspended in lysis buffer (50-mm Tris pH 7.5, 1-mm EDTA and 150-mm NaCl) with protease (10-μg/ml leupeptin, 10-μg/ml pepstatin A and 10 μg/ml chymostatin) and phosphatase (HALT™ phosphatase inhibitor cocktail, Thermo) inhibitors, then homogenized using a Dounce homogenizer. NP-40 (1%) and SDS (1%) were immediately added to the lysate. Lysates were incubated on ice for 10 min, at room temperature for 10 min, and then centrifuged at 13 000g for 20 min. Supernatants were collected and protein concentration was measured using the BCA Protein Assay Kit (Thermo Scientific, USA). Twenty-five microgram of lysates were loaded for standard SDS-PAGE and subsequently transferred onto nitrocellulose membranes (0.45 μm). Membranes were incubated with primary and HRP-conjugated secondary antibodies (Jackson Immunoresearch, USA, 1:5000) then visualized with ECL western blotting substrate (Pierce, USA). For antibody dilutions, see Table 1. Densitometry measurements were performed using ImageJ and normalized to tubulin or actin from the same membrane.
Image analysis
Cells were scored as positive for stress granules if they displayed at minimum 2 puncta that measured at least 0.15 μm2. The granularity index was calculated in MatLab by using the Sobel operator for edge detection. Sobel detection works by calculating the image gradient and predicting whether a point is classified as an edge. Edges were detected in 2–3 cytosolic sections per cell and then divided by the size of each section to give the granularity index with the unit edges/pixel. Granularity index was then normalized to control for each genotype. Analysis of nuclear puncta in neurons was done using ImageJ. Briefly, individual nuclei were selected from oligo-dT FISH images and spots were identified using the analyze particles function with a threshold of 0.5–15 μm2. Cells with TDP-43 nuclear depletion were counted manually. Images were taken in lumbar spinal cords focused on the ventral horn, and a cell was counted if it had a nuclear size of at least 20 μm2.
Statistics
Unpaired t-tests (Mann–Whitney) or Kruskal–Wallis tests were used to measure statistical significance (Prism, GraphPad). A P-value lower than 0.05 was considered significant.
Acknowledgements
We thank all CVV lab members, Daryl Bosco, Steph Weber and Tatyana Shelkovnikova for helpful insights and discussion. We also thank Yousra Khalfallah and Celine Desseille for their early work on the hyperthermia model, and Aurélie Cleret-Buhot from the CRCHUM imaging platform for technical support.
Conflict of Interest statement. The authors declare no conflict of interest.
Funding
This project was funded by Canadian Institutes of Health Research (CVV). AD was supported by ALS Canada and CIHR Doctoral Awards, and MG was supported by an ALS Canada/Fondation Vincent Bourque/Brain Canada Doctoral Award.
