TDP-43 aggregation inside micronuclei reveals a potential mechanism for protein inclusion formation in ALS

Amyotrophic lateral sclerosis (ALS) is a devastating progressive neurodegenerative disease with no known etiology. The formation of pathological protein inclusions, including RNA-binding proteins such as TDP-43 and rho guanine nucleotide exchange factor (RGNEF) are a hallmark of ALS. Despite intensive research, the mechanisms behind protein aggregate formation in ALS remains unclear. We have investigated the role of metabolic stress in protein aggregate formation analyzing how it is relevant to the co-aggregation observed between RGNEF and TDP-43 in motor neurons of ALS patients. Metabolic stress was able to induce formation of micronuclei, small nuclear fragments, in cultured cells. Notably, we observed the formation TDP-43 protein inclusions within micronuclei that co-aggregate with RGNEF and can be released to the cytoplasm. We observed that the leucine-rich domain of RGNEF is critical for its interaction with TDP-43 and localization in micronuclei. Finally, we described that micronuclei-like structures can be found in brain and spinal cord of ALS patients. This work is the first description of protein inclusion formation within micronuclei which also is linked with a neurodegenerative disease. The formation of TDP-43 inclusions within micronuclei induced by metabolic stress is a novel mechanism of protein aggregate formation which may have broad relevance for ALS and other neurodegenerative diseases.


Results
The LeuR domain of RGNEF localizes in TDP-43-positive micronuclei after metabolic stress. In order to analyze the effect of metabolic stress on TDP-43 cellular localization and its potential for protein aggregate induction, we tested several conditions including high and low glucose in the presence or absence of lactate. Compared to the control condition (25 mM glucose media; high glucose), we observed an increase of 7.7-fold of globular cytoplasmic structures containing nucleic acids and the nuclear RNA-binding protein TDP-43, which resemble micronuclei in HEK293T cells incubated with 30 mM lactate in low glucose media (0.6 mM glucose) and 3.4-fold in cells only in low glucose media. Notably, we observed a more robust increase in the number of these micronuclei-like structures when cells were incubated with 30 mM lactate in high glucose media compared to control (18.8-fold increase) (Fig. 1A). Under this condition, in which the lactate concentration is in the same range of the glucose concentration of the media, we observed an increase of mitochondrial metabolism ( Fig. 1B,C), increase in reactive oxygen species (ROS) concentration (Fig. 1D) and a mild cytotoxic effect over HEK293T cells (Fig. 1E), all consistent with a moderate stress condition. The stress under this condition is probably generated because gluconeogenesis is halted due to the high glucose concentration which may induce overload of the mitochondrial metabolic activity by lactate.
To confirm if these structures were actually micronuclei, we analyzed whether they contained nuclear markers together with nucleic acids. We observed that the structures formed after metabolic stress co-localized with Sirtuin 1, a nuclear deacetylate enzyme; TDP-43, and nucleic acids ( Fig. 2A). Also, we observed localization of TDP-43, nuclear pore complex proteins (NPCP), and nucleic acids (Fig. 2B) in the same structures, confirming they were micronuclei. These structures didn't show presence of the cytoplasmic protein poly-A binding protein (Supplementary Figure S1A) or an enrichment of ALS-linked protein SOD1 that is present in both nucleus and cytoplasm (Supplementary Figure S1B). Additionally, we also observed the formation of micronuclei after metabolic stress in the neuronal cell line SH-5YSY (Supplementary Figure S1C).
Since we observed the presence of TDP-43 in micronuclei after metabolic stress we decided to evaluate whether these TDP-43 -containing micronuclei were also enriched with endogenous full length RGNEF or an RGNEF fragment containing its LeuR domain in HEK293T cells. Leucine-rich (LeuR) domains have been described to be highly relevant for protein-protein interactions 35 . Because of this we hypothesize that the LeuR region of RGNEF is critical for the aggregate formation of RGNEF in motor neurons of ALS patients. To study the subcellular localization of RGNEF's LeuR under stress we made a construct expressing a flag-tagged version of the first 242 amino acids of RGNEF containing the LeuR domain (f-LeuR; Fig. 2C,D). The f-LeuR construct was expressed in HEK293T cells and incubated with 30 mM lactate in high glucose media to induce metabolic stress. Under metabolic stress, f-LeuR showed a change in its localization pattern, going from mainly cytoplasmic homogenous localization under basal conditions (Fig. 2E), similar to full length RGNEF 12,36 , to be observed highly concentrated in micronuclei structures (Fig. 2F). Interestingly, under metabolic stress, we observed micronuclei containing endogenous TDP-43 highly enriched with f-LeuR. In some cases, we were able to detected cells containing f-LeuR only in micronuclei (Fig. 3A). TDP-43 enriched micronuclei were also highly enriched in endogenous RGNEF despite the low levels of endogenous RGNEF in HEK293T cells 32 (Fig. 3B). It is worth noting that lactate increases the translocation of RGNEF to the nucleus in HEK293T cells (Supplementary Figure S1D). This effect of lactate may explain why RGNEF is enriched in micronuclei. As control for the specificity of LeuR domain over the localization of RGNEF in micronuclei, we performed an experiment expressing an RGNEF construct lacking the LeuR region (RGNEF-∆LeuR-myc; Supplementary Figure S2A). We observed that this construct was unable to localize in micronuclei under metabolic stress induced by lactate (Supplementary Figure S2B).
Our results indicate that metabolic stress induces the formation of micronuclei in HEK293T and SH-SY5Y cells and show that in HEK293T cells the micronuclei are enriched with f-LeuR, endogenous TDP-43 and endogenous RGNEF. These experiments also suggest that the LeuR region of RGNEF is necessary for its localization in micronuclei following metabolic stress.

TDP-43 interacts in vitro and co-localizes in vivo with f-LeuR. The co-localization of RGNEF's
LeuR domain with TDP-43 in micronuclei suggests that RGNEF and TDP-43 may interact through RGNEF's LeuR domain and be part of a protein complex. This idea is consistent with our previous observation showing co-immunoprecipitation between TDP-43 and full-length RGNEF 11 . Here, we transfected HEK293T cells with plasmids expressing TDP-43-myc and f-LeuR and then we crosslinked the proteins using DTSSP. We then performed immunoprecipitation using anti-myc antibody. Interestingly, after Western Blotting and probing www.nature.com/scientificreports www.nature.com/scientificreports/ with anti-flag antibody, we were able to detect the presence of f-LeuR in a protein complex with an electrophoretic shift of approximately 440 KDa (Fig. 3C, left panel). When we treated the samples with the reducing agent β-mercaptoethanol, which dissociates the crosslinking between the proteins, that complex was eliminated and a new band appeared at approximately 60 KDa. After stripping, we confirmed the presence of TDP-43 in those complexes using antibody against this protein (Fig. 3C, right panel). As control, we performed this experiment expressing RGNEF-∆LeuR-myc. Under the same experimental conditions, we were unable to observe the immunoprecipitation of a high molecular weight complex containing TDP-43 (Supplementary Figure S2C).
To analyze if this in vitro interaction has an in vivo correlate, we injected rats with rAAV9 that expressed f-LeuR or GFP as control ( Fig. 3D and Supplementary Figure S3D, E,F). We observed a high degree of co-localization between endogenous TDP-43 and f-LeuR in brain neuronal cells (Fig. 3D). This observation strongly suggests that RGNEF and TDP-43 are also able to interact through RGNEF's LeuR domain under physiological conditions in vivo.
These results suggest that TDP-43 together with RGNEF forms inclusions within micronuclei and that micronuclei facilitate the co-aggregation between TDP-43 and RGNEF, but not with FUS/TLS. Also, our results suggest that the TDP-43 aggregates observed within micronuclei are pathological in nature.
Micronuclei undergo a disruptive process and are present in ALS tissues. Micronucleus disruption has been previously described as the loss of micronuclei envelope permeability and its collapse implying the cessation of nuclear functions 39 . This may imply the release of the material previously in the micronuclei to the cytoplasm. To evaluate if a similar phenomenon occurs in HEK293T cells under metabolic stress, we analyzed the integrity of the nuclear membrane in cells containing TDP-43 inclusion-positive micronuclei after metabolic stress. Interestingly, we observed micronuclei with intact membranes (continuous membrane) (Fig. 5A) and also, micronuclei at different stages of the disruptive process, showing partial (discontinuous membrane) (Fig. 5B) or complete loss of the nuclear membrane marker (collapsed membrane) plus absence of detectable nucleic acids (Fig. 5C,D). Notably, between some of the cells with collapsed micronuclei we were able to also observe massive TDP-43 cytoplasmic aggregates (Fig. 5D,E,F). These observations support the idea that TDP-43 aggregates are likely released to the cytoplasm from disrupted micronuclei (Supplementary Figure S3E).
Our findings in vitro suggested that micronuclei formation in cells may be a relevant pathogenic mechanism in neurodegenerative diseases such as ALS. This would suggest the presence of observable micronuclei in tissues of ALS patients. We performed immunohistochemical analysis of ALS patient tissues to determine the presence of micronuclei. Notably, we were able to find micronuclei-like structures in the hippocampus (Fig. 6A-C)  Figures S3F,G). We were able to observe a micronucleus-like structure positive for both RGNEF and TDP-43 (Fig. 6A) and also in a β-tubulin III positive cell (Fig. 6C), findings that strongly support our observations in vitro.

Discussion
The understanding of the mechanisms involved in the formation and spread of protein aggregates in ALS and other neurodegenerative diseases is a highly relevant topic currently in biomedicine. The present study has investigated a potential novel mechanism of neurodegeneration-related protein aggregation. Additionally, these results www.nature.com/scientificreports www.nature.com/scientificreports/ provided a deeper understanding of the mechanism of the strong co-aggregation of TDP-43 with RGNEF, a novel protein in ALS pathology 11,12 .
In order to induce the formation of protein inclusions we evaluated how cellular metabolic stress induced by lactate affects the subcellular localization of a protein carrying only the LeuR region of RGNEF (f-LeuR), a www.nature.com/scientificreports www.nature.com/scientificreports/ domain broadly implicated in protein-protein interactions. Interestingly, we observed lactate-induced formation of micronuclei in two human cell lines; HEK293T and the neuronal-like cell line SH-SY5Y. We selected these dividing cellular models since micronuclei are generated only as consequence of errors in the mitosis 33,34 . The HEK293T cell line was our main cellular model because in addition it is not derived originally from cancer 40 . We didn't observe micronuclei formation induction using osmotic and oxidative acute stress 32 .
Recent attention has been brought to micronuclei because of the study of massive chromosomal rearrangements known as chromothripsis 41 , especially in the cancer field 34 . However, despite neurodegeneration and cancer sharing several molecular aspects 42 , the study of micronuclei formation in neurodegeneration has been limited to reports outside the nervous system in Alzheimer's disease (AD) and Parkinson's disease (PD). Micronuclei have been described as potential biomarkers for neurodegeneration as they are observed in higher frequency in peripheral lymphocytes, skin fibroblasts, and buccal mucosa cells from AD patients and peripheral lymphocytes from PD patients 43 .
The micronuclei we observed were enriched with endogenous TDP-43 which co-localized with f-LeuR and endogenous RGNEF. These results suggested RGNEF's LeuR is critical for the localization of full RGNEF in micronuclei under cellular metabolic stress. Our observation that a protein including the LeuR of RGNEF is part of a protein complex containing TDP-43, and also co-localizes in vivo with TDP-43, not only confirmed our previous observation about the interaction between TDP-43 and full length RGNEF 11 , but also indicated that the LeuR domain may be critical for the formation of the RGNEF-TDP-43-containing aggregates observed in motor neurons of ALS patients 11,12 .
A key finding of this study was that TDP-43 forms inclusions inside the micronuclei induced by lactate stress in culture cells. As has been described previously, TDP-43 can form aggregates due to the dysregulation of protein-protein [44][45][46][47] or protein-RNA 48,49 interactions. This suggests that the interior of the micronuclei may generate a microenvironment of altered protein/RNA homeostasis leading to an aberrant TDP-43 phase separation and consequent induction of aggregation. This process could be analogous to the current postulated mechanism in which stress granules become protein inclusions [14][15][16][17][18][19] . Remarkably, the disrupted micronuclei we observed in cultured cells show that TDP-43 aggregates inside micronuclei are likely released to the cytoplasm (Fig. 5D-F).
Evidence that supports our in vitro observations as a model of a pathophysiological phenomenon, comes from the observation of micronuclei-like structures in samples from three ALS patients, which were absent in neurologically healthy patients. We focused our study in hippocampus and spinal cord because both present TDP-43 pathology in ALS 50 and have been described to have neurogenesis 51,52 . Our main interest was to search for micronuclei in neurons; however, we cannot discard the possibility that micronuclei-like structures are also present in other types of neural cells.
We hypothesize that micronuclei formation is likely an early event in the ALS pathogenic process, as in our model micronuclei are generated as a result of a stress condition and only after that we observed aggregate formation. This could explain why we observed a relatively low incidence of micronuclei-like structures in the brain or spinal cord from ALS patients with advanced stages of the disease. A comprehensive quantitative analysis of micronuclei in early and late stage cohorts of ALS will illuminate this possibility, but this is beyond the scope of the current study.
In summary, this work shows that the LeuR domain of RGNEF may be critical for its co-aggregation with TDP-43 but most notably that the formation of TDP-43 inclusions within micronuclei could be one of the mechanisms of aggregate formation in neurodegenerative diseases such as ALS. Our observations open the door for a new fascinating field of study where the analysis of micronuclei formation and dynamics could be implicated in the pathogenesis of neurodegenerative diseases.
Human samples. Post-mortem frozen and formalin-fixed, paraffin-embedded tissues were collected as part of the ALS protocol at London Health Sciences Centre (London, Ontario, Canada). Ethics review and approval was granted by The University of Western Ontario Research Ethics Board for Health Sciences Research Involving Human Subjects (HSREB -Protocol #103735) for the use of brain and spinal cord tissues and for access to medical records for research purposes. Informed consent was obtained from all participants. All research was performed in accordance with relevant guidelines/regulations. Three sporadic ALS cases without known mutations in the coding sequence of SOD1, FUS/TLS and TDP-43 and without C9ORF72 expanded repeats, and three neurologically healthy controls were analyzed in this study.

Rats.
Female SAS Sprague-Dawley (Charles River), 12 weeks old and 250 g of weight, were used for the brain intraventricular injection experiment. All procedures involving animals, surgeries, and animal maintenance were in accordance with the Canadian Council for Animal Care and the University Council on Animal Care guidelines for research. Ethics review and approval was granted by the Animal Care Committee of The University of Western Ontario (Protocol #2017-035). High-Fidelity DNA Polymerase (Thermo Scientific) and the products were cloned into the vectors pcDNA 3.1/ myc-His A (Invitrogen) or pcDNA 3.1 (Invitrogen) respectively. A flag tag was added in the 5′ extreme of the coding sequence of LeuR and full length RGNEF for later detection using anti-flag antibodies. The sequences for the self-complementary adeno-associated viruses (scAAVs) were designed as described in the Figure S3. The sequences from the 5′ ITR to the 3′ ITR-∆trs were fully synthetized and cloned into the pUC18 vector (GeneScript) originating the vectors pscAAV-GFP and pscAAV-L-rich. All constructs were confirmed by sequencing.

Antibodies. See Supplementary
Metabolic stress conditions. To induce metabolic, stress cells were incubated with 30 mM lactate (DL-Lactic acid sodium salt, Sigma-Aldrich) in 10% FBS DMEM containing 25 mM glucose (high glucose-HG) and 1 mM pyruvate or 0.6 mM glucose (low glucose-LG) media (glucose given by the FBS present in the media).

Transfections of cells under metabolic stress. HEK293T cells maintained 7 days under metabolic stress
(HG + lactate) or under control conditions (HG) were seeded onto coverslips previously treated with attachment factor (Gibco) in 6-well plates at 250,000 cells/ml 24 hours before transfection. Transfections were performed using Lipofectamine 2000 (ThermoFisher Scientific) at 70% of confluency using 2.5 µg of DNA in a ratio (µg) DNA:(µl) lipofectamine 1:2.5. Cells were maintained under metabolic stress for 48 hours after the transfection. Then, cells in the coverslips were fixed in 4% paraformaldehyde in PBS for 15 min and processed for immunofluorescence. For the cross-linking-immunoprecipitation experiment transfections were performed using the same protocol, but using 6-well plates without coverslips and without metabolic stress.

Micronuclei quantification. After maintaining HEK293T cells in high and low glucose media in presence
of absence of 30 mM lactate during 10 days, immunofluorescence detecting TDP-43 and staining with Hoechst was performed. TDP-43 and Hoechst positive micronuclei were quantified by triplicate under all the conditions described with an average of 91.08 + 3.24 cells per field. Data is presented as the percentage of cells containing at least one micronucleus over the total amount of cells.
Mitochondrial metabolic activity analysis. To evaluate the mitochondrial metabolic activity under metabolic stress we used MitoTracker Red CM-H2XRos. This nonfluorescent chemical passively diffuses across the plasma membrane and becomes fluorescent and accumulates in active mitochondria after oxidation. After 2 or 20 days under metabolic stress, cells were incubated with 400 nM MitoTracker Red CM-H2XRos (Invitrogen) at 37 °C for 45 min. Then, the cells were fixed and processed for immunofluorescence. The data were presented as relative MitoTracker fluorescence intensity divided by the number of cells for normalizing.

Detection of reactive oxygen species (ROS).
The generation of ROS under metabolic stress was studied using carboxy-H2DCFDA. This chemical is nonfluorescent until the acetate groups are removed by intracellular esterases, and oxidation by ROS occurs within the cell. The oxidized compound is largely retained intracellularly. After 2 or 20 days under metabolic stress, the cells were incubated with 20 µM carboxy-H2DCFDA (Invitrogen) at 37 °C for 1 hour. Then, cells were left in recovery for 2 hours and observed under the confocal microscope (green channel). To quantify specific fluorescence due to ROS activity, an autofluorescence control of HEK293T cells not treated with carboxy-H2DCFDA was used. Data was presented as relative ROS fluorescence intensity divided by the confluence of cells for normalization. The confluence was obtained by measuring the area cover by cells (observed by autofluorescence in the blue channel) using the ImageJ software. Cytotoxicity analysis. Cells were seeded in white 96-well plates at 8,000 cells/ml per well. The cytotoxicity was measured using the CytoTox-Glo ™ Cytotoxicity Assay kit (Promega) according to the manufacturer's protocol after 2 or 20 days under metabolic stress. This kit quantifies "dead-cell protease activity", which is released from cells that have lost membrane integrity. To obtain the percentage of cell death, the values obtained after the stress condition or control were normalized against total protease activity obtained after cell lysis using digitonin.
Cross-linking experiment. Cross-linking was performed incubating 500 µ g of protein lysate of HEK293T cells transfected with pcDNA-flag-L-rich and pcDNA-TDP-43-myc with 3,3′-Dithiobis(sulfosuccinimidylpropionate) (DTSSP; Thermo Scientific) to a final concentration of 1.3 mM in PBS buffer pH 7.2. After 30 min at room temperature, the cross-linking was quenched by the addition of 1 M Tris, pH 7.5 to a final concentration of 37 mM. Immunoprecipitation and immunoblot. Immunoprecipitations were performed using the Dyanbeads Protein G Immunoprecipation Kit (Invitrogen) according the manufacturer's instructions. After the cross-linking reaction, each sample was immunoprecipitated using 2 µg of mouse anti-myc (Cedarlane) or mouse IgG (Sigma-Aldrich) as control. A group of samples was treated with 50 mM β-mercaptoethanol (Sigma) at 100 °C for Immunofluorescence (tissues). Archival paraffin-embedded sections of brain from ALS cases and brain from rats injected with the AVVs post-fixed and embedded in paraffin, were serially sectioned at 6 μm thickness. Tissue sections were deparaffinized using standard protocols and following antigen retrieval (10 mM sodium citrate, 0.05% Tween 20, pH 6), the sections were incubated with the primary antibodies: goat anti-GFP, goat anti-flag, rabbit anti-TDP-43, or mouse anti-β tubulin III. Primary antibodies were detected using secondary antibodies conjugated to Alexa Fluor 488 or Alexa Fluor 555. Nuclei were stained with DAPI (5 µg/mL) and tissue then were visualized by scanning confocal microscopy.
Confocal imaging and 3D analysis. Cells and tissues from immunofluorescence experiments were visualized by scanning confocal microscopy (Leica TCS SP8). The 3D reconstruction analysis of set of confocal images were performed using the Leica 3D analysis tool from LAS X software.
Co-localization images. Intensity Correlation Analysis 53 using ImageJ software was performed to obtain the co-localization images. The co-localized pixels are shown as PDM (Product of the Differences from the Mean) images. PDM = (red intensity-mean red intensity) × (green intensity-mean green intensity). In the co-localization images, blue color indicates a low level of co-localization while yellow and white indicate a high degree of co-localization.
Protein modeling. Molecular modeling was performed using I-Tasser software (http://zhanglab.ccmb.med. umich.edu/I-TASSER/), which predicts protein structure using modeling by iterative threading assembly 54,55 . The models were visualized using RasTop 2.2 (http://www.geneinfinity.org/rastop). Statistical analysis. The statistical analyses were performed with GraphPad Prism 8.3 software using one-way or two-way ANOVA with Tukey post-hoc analysis to obtain exact p values. All data used for graphics are presented in the Supplementary Tables S2-S5. All the analysis passed normality (Shapiro-Wilk test) and equal variance test (Brown-Forsythe test). The results of the normality test are presented in the Supplementary  Table S6. Data were expressed as mean ± SD. Data was judged to be statistically significant when p < 0.05. For experiments carried out using culture cells, n of 1 represents and independent plate of cells under control or treatment condition.