TDP-43 skein-like inclusions are formed by BAG3- and HSP70-guided co-aggregation with actin binding proteins

Abstract

In multiple neurodegenerative diseases, the RNA binding protein TDP-43 forms cytoplasmic aggregates of distinct morphologies, including skein-like, small rounded granular, and large spherical inclusions. While the N-terminal self-oligomerization domain (NTD) regulates TDP-43 de-mixing into cytoplasmic droplets, inhibition of NTD-mediated oligomerization is shown here to promote the formation of skein-like inclusions. Utilizing proximity labeling/mass spectrometry, cellular stresses are shown to induce TDP-43 association with actin-binding proteins that include filamins and α-actinin. Small interfering RNA-mediated reduction of filamin in Drosophila ameliorates cell loss from cytoplasmic TDP-43, consistent with filamin/TDP-43 interaction enhancing cytotoxicity. TDP-43’s association with actin-binding proteins is mediated by BAG3, a HSP70 family nucleotide exchange factor (NEF) that regulates proteostasis of actin-binding proteins. BAG2, another HSP70 NEF facilitates the formation of small, rounded TDP-43 inclusions. Our evidence demonstrates that both TDP-43 self-oligomerization and its binding partners, including HSP70 and co-chaperones BAG2 and BAG3, drive formation of different types of TDP-43 inclusions.

Introduction

The formation within affected neurons of spherical, skein-like, diffuse, granular, or dash-like cytoplasmic inclusions of the RNA binding protein TDP-43 is a hallmark of multiple neurodegenerative diseases, including almost all instances of ALS¹ and limbic-predominant age-related TDP-43 encephalopathy (LATE)², half of cases of frontal temporal dementia (FTD), and many instances of Alzheimer’s disease (AD)³ and chronic traumatic encephalopathy (CTE)⁴. Within these aggregates, TDP-43 is often fragmented and accumulates post-translational modifications including phosphorylation^{5, 6}, ubiquitination^{5, 6}, and acetylation⁷. The mechanism(s) through which TDP-43 forms protein inclusions with different morphologies are not known.

Liquid-liquid phase separation (LLPS) has emerged as a mechanism for the assembly of de-mixed droplets containing RNA-binding proteins associated with ALS/FTD, such as FUS⁸, hnRNPA1⁹, and TDP-43^10–15. TDP-43-containing droplets, in a liquid-like phase separated state, have been observed to transition into gel/solid structures under prolonged stress, or even into amyloid-like fiber structures in vitro¹². TDP-43 is a primary component of the inclusions in ALS and FTD⁵. The structure of TDP-43 fibrils has been reported by one group using cryogenic electron microscopy of fibrils isolated from postmortem tissues of individuals with FTLD and ALS featuring TDP-43 proteinopathy^{16, 17}. However, two other groups using similar methods did not detect TDP-43 fibrils, instead identifying non-fibrillar TDP-43 aggregates^{18, 19}. These structural efforts findings highlight unsolved questions about the mechanisms underlying formation of the various TDP-43 inclusions.

Work from us²⁰ and others^{7, 11, 21, 22} has demonstrated that TDP-43 phase separation is regulated by post-translational modifications (PTMs) that affect its RNA-binding activity or self-interaction. Examples of such PTMs include acetylation^{7, 20, 22} on lysine residues involved in RNA binding and phosphorylation on the N-terminal¹¹ or low complexity²¹ domains. Our findings also indicate that TDP-43 with modifications can form de-mixed droplets in a different form²⁰. For instance, acetylated TDP-43 forms liquid anisosomes in the nucleus²⁰, while mis-localized cytoplasmic TDP-43 can form stress-granule independent de-mixed structures following transient association with stress granules formed in response to arsenite stress^{12, 23}. These results suggest that modifications on TDP-43 may regulate the association of TDP-43 with different proteins and the formation of diverse assemblies.

Here we use the combination of proximity labeling coupled with mass spectrometry, analysis of transcriptomic data from ALS and normal motor neurons, and TDP-43 aggregation in the fly to identify that BAG3, a nucleotide exchange factor (NEF) for the HSP70 chaperone family, acts in conjunction with HSP70 and the small heat shock protein HSPB8 to mediate formation of skein-like inclusions of TDP-43 that are tethered to actin filaments through actin binding proteins.

Results

Inhibition of TDP-43 self-interaction alters stress-induced TDP-43 inclusions

Prior efforts using NMR²⁴ or crystallography²⁵ have reported that the amino terminal domain (NTD) of TDP-43 forms dimers or homo-oligomers, while disruption of NTD-dependent oligomerization inhibits TDP-43 liquid-liquid phase separation but promotes TDP-43 pathological aggregation^{11, 26}. TDP-43 C-terminal fragments without the NTD domain are found in insoluble aggregates isolated from ALS tissues⁵. To test how the TDP-43 NTD affects cytoplasmic aggregation of TDP-43 in response to proteotoxic stress, we expressed a full-length TDP-43 variant with a defective nuclear localization sequence (NLS) (TDP-43^NLSm-Clover) and an NTD- and NLS-truncated TDP-43 variant (TDP-43^ΔNTD-Clover) missing the first 101 amino acids including the NLS. As expected, both proteins exhibited diffuse, cytoplasmic localization (Fig. 1a). Upon proteotoxic stress from exposure to sodium arsenite, full length TDP-43 (TDP-43^NLSm-Clover) phase separated into round, de-mixed droplets (Fig. 1a) after an initial transient recruitment to stress granules, consistent with our previous report²³. Surprisingly, while proteotoxic stress transiently induced NTD truncated TDP-43 (TDP-43^ΔNTD-Clover) into stress granules, skein-like inclusions subsequently formed (Fig. 1a–c, Extended Data Fig. 1a). Proteotoxic stress also induced similar skein-like inclusions from a 35KD variant lacking the first 90 amino acids of TDP-43 (Extended Data Fig. 1b). Formation of skein-like inclusions by the NTD-truncated TDP-43 was not affected by mutations that rendered it incompetent for RNA binding (Extended Data Fig. 1c–d), despite such mutations being known to promote liquid-liquid phase separation of full length nuclear²⁰ or cytoplasmic^{7, 23} TDP-43.

Figure 1. Disruption of NTD interaction between TDP-43 monomers affects the morphology of TDP-43 inclusions induced by proteotoxic stress.

(a) Representative live images of cells expressing TDP-43^NLSm-Clover and TDP-43^ΔNTD-Clover treated with sodium arensite. (b) Percentage of cells that form round or skein-like TDP-43 condensates shown in (a). N=four biological replicates representing 149, 214, 146, 215 cells (TDP-43^NLSm-Clover) and 145, 131, 176, 103 cells (TDP-43^ΔNTD-Clover), respectively. Bar: SD. (c) Box plot of the circularity of the de-mixing structures formed by TDP-43^NLSm-Clover and TDP-43^ΔNTD-Clover. Center lines (the medians); box limits (the 25th and 75th percentiles) as determined by R software; whiskers (1.5 times the interquartile range from the 25th and 75th percentiles), outliers (dots). Number of de-mixing structures are 2799 and 1938 from 6 fields of view per condition. (d) Representative live images of cells expressing TDP-43^{NLSm/2KQ-Clover} and TDP-43^{NLSm/2KQ/S48D-Clover} treated with sodium arensite. (e) Bar plot showing the percentage of cells that form round or skein-like TDP-43 condensates shown in (c). N=4 biological replicates representing 151, 159, 100, 186 cells (TDP-43^{NLSm/2KQ-Clover}) and 45, 44, 39, 55 cells (TDP-43^{NLSm/2KQ/S48D-Clover}), respectively. Bar: SD. (f) Box plot of the circularity of the de-mixing structures formed by TDP-43^NLSm-Clover and TDP-43^ΔNTD-Clover. Center lines (the medians); box limits (the 25th and 75th percentiles) as determined by R software; whiskers (1.5 times the interquartile range from the 25th and 75th percentiles), outliers (dots). N: 2595 and 1800 from 6 fields of view per condition. (g) FRAP analysis of TDP-43 skein-like inclusions. (h) FRAP curve of TDP-43 skein-like inclusions. Average normalized intensity is plotted, and standard deviation is plotted for each time point. N: 4 de-mixing structures from three independent replicates. Bar: s.d. (i) Representative images of skein-like TDP-43 inclusions preserved after mild cell permeabilization from more than 3 biological replicates. (j) Fluorescent intensity plot across the lines that are drawn in (i) at each time point.

We hypothesized that the different morphologies of TDP-43 inclusions could result from the loss of NTD oligomerization. Therefore, full length cytoplasmic TDP-43 (with and without mutations K145Q and K192Q to inhibit RNA binding) was modified with an S48D mutation (Fig. 1d–f and Extended Data Fig. 1e), a variant which has been reported to disrupt NTD self-interaction and reduce liquid-liquid phase separation (LLPS) of nuclear TDP-43¹¹. S48D mutation promoted both full-length TDP-43 variants to form skein-like inclusions in the cytoplasm in response to arsenite-induced stress, results consistent with the NTD-mediated self-interaction playing a crucial role in determining the type of TDP-43 inclusion. Additionally, absence of fluorescence recovery after photobleaching of the arsenite or proteosome inhibition-induced skein-like structures indicated that these structures were gel/solid-like (Fig. 1g–h and Extended Data Fig. 1f) and were resistant to solubilization after cell plasma membrane permeabilization, further supporting their insoluble nature (Fig. 1i,j and Extended Data Fig. 1g).

Removal of RRM1 from TDP-43^ΔNTD-Clover, a domain that we have previously shown to be necessary for the sodium arsenite-induced transition of TDP-43 from liquid to gel/solid state²³, inhibited formation of skein-like structures (Extended Data Fig. 1h). Furthermore, deletion of the conserved region in the low complexity domain (which forms an α-helix¹⁴ and mediates multivalent interactions that induce LLPS¹⁴), also inhibited formation of skein-like inclusions (Extended Data Fig. 1i). A V5-tagged TDP-43 variant without the NTD also formed skein-like inclusions (Extended Data Fig. 1j), eliminating the possibility that the GFP tag was needed to influence skein-like inclusions.

The discovery that NTD interaction regulates TDP-43 assembly into round or skein-like inclusions prompted us to determine how co-expression of both full-length and amino-terminally truncated TDP-43 variants (as are found in ALS/FTD^{5, 6}) affects TDP-43 inclusions (Extended Data Fig. 2a). An increasing level of full-length TDP-43 shifted skein-like inclusions to round, droplet-like inclusions that contained both full-length and truncated TDP-43, with droplet formation enhanced by an RNA-binding incompetent full length TDP-43 (Extended Data Fig. 2b–c).

TDP-43 skein-like inclusions are associated with actin filaments

Next, we used APEX2 proximity labeling²⁷ of cells induced to transiently express NTD-truncated TDP-43^{ΔNTD-Clover-APEX2} to identify proteins that interact with TDP-43 in the skein-like structures. To specifically label TDP-43 skein-like inclusions, we performed a mild permeabilization step to remove diffusely positioned cytoplasmic TDP-43 before a brief incubation with biotin-phenol and hydrogen peroxide to biotinylate adjacent proteins (Fig. 2a). Streptavidin staining demonstrated robust labeling of TDP-43 skein-like structures (Fig. 2b), as expected. Tandem mass tag isotope labeling and quantitative mass spectrometry were then used to identify 66 proteins that were increasingly labeled under conditions where skein-like inclusions of TDP-43 formed (Fig. 2c–e; Supplementary Table 1–2). The most enriched proteins were actin-binding proteins, including actinins and filamins (Fig. 2c–e). Additionally, heat shock proteins, ubiquitin-proteasome proteins, tubulins, and caveolae proteins were also found to be increasingly bound to TDP-43 during inclusion formation (Fig. 2d,e). Conversely, RNA-binding proteins such as FUS, HNRNPU, MATR3, and the stress granule protein UBAP2L exhibited decreased binding to TDP-43 (Fig. 2d,e), suggesting dissociation of TDP-43 from stress granules to form the inclusions.

Figure 2. Associated proteins of TDP-43 skein-like inclusions identified by proximity labeling coupled with quantitative mass spectrometry.

(a) Schematic illustration of proximity labeling of TDP-43 skein-like inclusions and the identification of associated proteins by tandem-mass-tag quantitative mass spectrometry. (b) Representative images showing biotin labeling signals in cells expressing TDP-43^{ΔNTD-Clover-APEX2} under no stress or proteotoxic stresses from more than three biological replicates. (c) Volcano plot of differentially labeled proteins comparing NaAsO₂ treatment to no stress conditions. Differentially enriched proteins (with cut-off of p-Value <0.001, foldchange >1) are highlighted with names. Proteins also increasingly labeled under MG132 stress are indicated in orange. SC, spectral count. (d) Interaction network of the differentially enriched proteins. Node colors represent fold change, highlighting the degree of enrichment. (e) Heatmap depicting differentially enriched proteins under NaAsO₂ or MG132 treatment conditions.

The association of TDP-43-containing skein-like structures with filamentous actin was confirmed by affinity labeling actin bundles with LifeAct-mRuby²⁸ (Fig. 3a,b). One of the identified proteins, α-actinin, was observed to co-aggregate in the skein-like TDP-43 inclusions upon arsenite stress (Fig. 3c). Using human iPSC-derived motor neurons (Fig. 3d), monomeric infrared fluorescent protein (mIFP²⁹)-tagged α-actinin was observed to co-aggregate with TDP-43^ΔNTD-Clover within both neuronal cell bodies (Fig. 3e) and their processes (Fig. 3f), while mIFP alone did not (Extended Data Fig. 3a), supporting the specificity of the interaction with actinin. The TDP-43^ΔNTD-Clover inclusions associated with actin filaments (Fig. 3g). Furthermore, use of immunofluorescence (Extended Data Fig. 3b–e) and immunogold electron microscopy (Extended Data Fig. 3f) revealed that the interaction of TDP-43 with actin filaments and actin-binding proteins survived cell membrane permeabilization. Labeling of the RNA in the cells with 5-ethynyl uridine revealed that the TDP-43^ΔNTD-Clover skein-like inclusions were not enriched in RNA (Extended Data Fig. 3g), consistent with decreased proximity labeling of RNA binding proteins by TDP-43^{ΔNTD-Clover-APEX2} in stressed conditions (Fig. 2d,e).

Figure 3. TDP-43 skein-like inclusions associate with actin filaments.

(a) Representative live images showing the formation of TDP-43 skein-like structure along actin filaments after NaAsO₂ treatment in cells expressing TDP-43^ΔNTD-Clover and LifeAct^mRuby2 from more than three biological replicates. (b) Representative live images showing the formation of TDP-43 skein-like structure along actin filaments after MG132 treatment in the cells expressing TDP-43^ΔNTD-Clover and LifeAct^mRuby2 from more than three biological replicates. (c) Representative live images of TDP-43 co-aggregating with Actinin-1 after NaAsO₂ treatment in cells expressing TDP-43^ΔNTD-Clover and ACTN1^mIFP from more than three biological replicates. (d) Schematic diagram of experiments assessing the co-aggregation of TDP-43^ΔNTD-Clover and ACTN1^mIFP in human iPSC-derived motor neurons. (e) Representative live images showing TDP-43 co-aggregation with Actinin-1 after NaAsO₂ treatment in iPSC-derived motor neurons from more than three biological replicates. (f) Representative images of TDP-43 co-aggregating with Actinin-1 in the neurites of iPSC-derived motor neurons following NaAsO₂ treatment from more than three biological replicates. (g) Representative images of iPSC-motor neurons with TDP-43 inclusions and stained with phalloidin for visualization of actin filaments from more than three biological replicates.

Furthermore, enrichment of TDP-43 and actin-binding proteins in insoluble fractions was confirmed using biochemical fractionation (Extended Data Fig. 4a). Consistent with previous reports, a proportion (20–50%) of cytoplasmic TDP-43 became insoluble following transient MG132 or arsenite stress (Extended Data Fig. 4b–d). After proteotoxic stress exposure higher proportions of both actin and actin-binding proteins were insoluble (Extended Data Fig. 4c–d). The interaction between TDP-43 and actin-binding proteins was further confirmed through co-immunoprecipitation (Extended Data Fig. 4e) and super-resolution microscopy which revealed TDP-43 labeled particles surrounding actin filaments in cells exposed to arsenite-mediated stress (Fig. 4a).

Figure 4. The formation of TDP-43 skein-like inclusions relies on actin filaments.

(a) Super-resolution images of TDP-43^ΔNTD-Clover and actin filaments (stained with phalloidin) in the skein-like inclusions from more than three biological replicates. Left, a low magnification image of a cell with skein-like inclusions. Right, high magnification images showing TDP-43^ΔNTD-Clover surrounding the actin filaments in the skein-like inclusions. (b-c) Effects of nocodazole, Blebbistatin, cytochalasin D on the formation of TDP-43 inclusions. Cells co-expressing TDP-43^ΔNTD-Clover and LifeAct^mRuby2 are pre-stained with 1 μM Sir-tubulin for 1 hour, followed by treatment with 100 μM NaAsO₂ (b) or 10 μM MG132 (c), with or without 2 μM nocodazole, 25 μM Blebbistatin, or 3.3 μM cytochalasin D for 3 hours. (d) Quantification of cells forming skein-like inclusions under the conditions shown in (b) and (c). Number of cells analyzed: 170 (MG132), 173 (MG132+blebbistatin), 169 (MG132 + cytochalasin D), 187 (MG132 + nocodazole), 200 (NaAsO₂), 236 (NaAsO₂ + Blebbistatin), 161 (NaAsO₂ + cytochalasin D), and 148 (NaAsO₂ + nocodazole) from one representative biological replicate. (e) Representative images of cells expressing TDP-43^ΔNTD-Clover and LifeAct^mRuby2 that are treated with NaAsO₂ for 3 hrs before adding 50 μM blebbistatin from more than three biological replicates.

Bundles of assembled actin enhance formation of TDP-43 skein-like structures

Disruption of assembled actin bundles with the actin depolymerizing agent cytochalasin D³⁰ or the myosin inhibitor blebbistatin³¹, but not drugs (e.g., nocodazole) that disrupt microtubule assembly³², strongly inhibited the formation of TDP-43 skein-like structures, with TDP-43 instead forming round or amorphous particles in the absence of filamentous actin (Fig. 4b–d). When actin bundles were disrupted with blebbistatin treatment after the initial formation of skein-like inclusions of TDP-43, those inclusions remained (Fig. 4e), consistent with the formation, but not maintenance, of TDP-43 skein-like structures to be dependent on assembled actin filaments.

Filamin A co-aggregates with TDP-43 in ALS motor neurons

Examination of TDP-43 pathology in nine ALS nervous system tissues revealed 10–70% of surviving motor neurons had developed TDP-43 pathology, with about half (average 51%) of those containing skein-like inclusions (Extended Data Fig. 5a). Given the crucial role of actin filaments in the formation of TDP-43 skein-like inclusions in cells (Fig. 4b–d), the expression level of each candidate protein identified in proximity labeling was examined in the RNA-seq dataset of laser-captured motor neurons³³. Expression of several actin-binding proteins (e.g., filamin A, filamin B, filamin C, and actinin) was increased in motor neurons from individuals with ALS, with filamin A showing a particularly significant (4 fold) increase (Fig. 5a, Supplementary Table 3) despite an unchanged level of β-actin (Fig. 5b), while RNAs that are cryptically spliced when nuclear TDP-43 is reduced (e.g., STMN2^{34, 35}) were significantly reduced (Fig. 5c). While filamin A in non-neurologic disease control tissues was diffusely present within the cytoplasm of motor neurons (Fig. 5d), in ALS tissue samples filamin A was enriched in TDP-43 skein-like pathologic inclusions (Fig. 5e), with over 90% of motor neurons exhibiting skein-like pathology that included co-aggregation of TDP-43 with filamin A (Fig. 5f).

Figure 5. Elevation of actin binding proteins in motor neurons of ALS patients, the co-aggregation of filamin with TDP-43 inclusions in diseased motor neurons, and decrease of cytotoxicity from cytoplasmic TDP-43 when filamin (cher) is reduced in the Drosophila eye.

(a-c) RNA levels of FLNA (a), FLNB (a), FLNC (a), ACTN1 (a), ACTB (b) and STMN2 (c) in laser-captured motor neurons from the spinal cords of ALS and control patients, data adapted from Krach, et al.³³. N=8 (Ctrl patients) and 13 (ALS patients) Unpaired t-tests are used for statistical analysis, no adjustment. * means <0.01. Bar: SD. 89 indicates patient #89. (d) A representative image of healthy motor neurons showing nuclear TDP-43 and cytoplasmic filamin A. (e) Representative images of motor neurons from ALS patients with TDP-43 skein-like pathological inclusions and enrichment of filamin A. (f) Quantification of the percentage of motor neurons with TDP-43 pathology exhibit filamin A enrichment. N: three ALS patients (from 29, 30, 8 motor neurons with skein-like inclusions). (g) Quantification of cell viability under the conditions of expression of TDP-43^ΔNTD-Clover or Clover with or without co-expression of filamin A. Expression of TDP-43^ΔNTD-Clover or Clover is driven by TetO promoter and induced by the addition of doxycycline. N: 3 replicate wells. Bar: SD. (h) Knock-down of the filamin homolog (cher) in GMR>TDP-43^ΔNLS fly model ameliorates the eye degeneration phenotype. (i) Quantification of eye degeneration scores in the conditions with or without cher knock-down. N=27 (Ctrl) and 25 (Cher-RNAi) Drosophila). Unpaired t-tests are used for statistical analysis. Bar: SD.

As a test of whether filamin A contributes to toxicity accompanying TDP-43 pathology, we overexpressed filamin A in cells expressing TDP-43^ΔNTD-Clover or Clover alone. In cells expressing TDP-43^ΔNTD-Clover, increased filamin A significantly exacerbated cell death (Fig. 5g). We then extended this test to photoreceptors in the adult fly. Selective expression of cytoplasmic TDP-43^ΔNLS led to severe eye degeneration³⁶ (Fig. 5h). Remarkably, shRNA knockdown of cher, the fly filamin homolog, strongly alleviated degeneration in the Drosophila eye induced by cytoplasmic TDP-43 (Fig. 5i), an outcome suggesting that actin-binding proteins, such as filamin, may promote the toxicity caused by cytoplasmic TDP-43.

BAG3 drives skein-like TDP-43 inclusions

In addition to actin binding proteins, a group of heat shock chaperones was identified in our list of proteins co-enriched with TDP-43 in skein-like inclusions, including several HSP70 family members, HSPB1, and two nucleotide exchange factors (NEFs) of HSP70, namely BAG3 and HSPH1 (Fig. 3e). The NEFs of HSP70 include HSPH1 and six members of the BAG family (BAG1 to BAG6)³⁷. BAG3 forms a complex with small heat shock proteins, particularly HSPB8, via the two isoleucine, proline, valine (IPV) motifs^38–40 (Extended Data Fig. 5b). BAG3 is the major NEF induced by proteotoxic stress^{41, 42} and has been reported to regulate the clearance of damaged actin-binding proteins, such as filamins⁴³. Dysfunction of BAG3 can lead to myofibrillar myopathy and cardiomyopathy³⁷. From published snRNA-seq data of human spinal cord^{44, 45} and the Allen mouse spinal cord immunocytochemistry data⁴⁶, BAG3 is one of the highly expressed NEFs in alpha motor neurons (Fig. 6a, Extended Data Fig. 5c–f), the neuron primarily affected in ALS and develops TDP-43 pathology.

Figure 6. BAG3-HSPB8 complex co-assembles with TDP-43 skein-like structures and is crucial for the assembly of the skein-like inclusions.

(a) Expression of BAG3 in the adult mouse spinal cord. The image is ISH data from the Allen mouse spinal cord atlas⁴⁶ https://mousespinal.brain-map.org/imageseries/show.html?id=100020020 and depicts a cervical slice from a P56 mouse. (b) Representative live images of cells expressing BAG3^mRuby2 and TDP-43^ΔNTD-Clover or TDP-43^ΔNTD-Clover under MG132 treatment. (c) Quantification of cells with skein-like inclusions at different time points. N=109, 185, 111, 118 cells at 2 hr and 105, 127, 105, 111 cells at 3 hr (TDP-43^ΔNTD-Clover) per replicate and N=102, 122, 134, 190 cells at 2 hr and 105, 131, 130, 128 cells at 3 hr (BAG3^mRuby2 and TDP-43^ΔNTD-Clover) per replicate. (d) Representative images of BAG3 and TDP-43^ΔNTD-Clover in cells under no stress, 100 μM NaAsO₂ or 10 μM MG132. Cells were permeabilized with 50 μg/mL digitonin for 5 mins before fixation. (e) Representative images of HSPB1, HSPB8 and TDP-43^ΔNTD-Clover in cells under no stress, 100 μM NaAsO₂ or 10 μM MG132. Cells were permeabilized with 50 μg/mL digitonin for 5 mins before fixation. (f) Representative live images of TDP-43^ΔNTD-Clover skein-like or granule-like inclusions following 100 μM NaAsO₂ treatment in the cells transfected with control siRNA, siBAG3, siHSPB8 or siHSPB1. (g-h) Quantification of cells with granule-like or skein-like inclusions under the conditions described in (f). N: four biological replicates. Cell number quantified in each replicate experiment: 183, 175, 192 (control siRNA); 159, 192, 160 (siBAG3); 225, 201, 176 (siHSPB8); 123, 143, 154 (siHSPB1). Bar: SD.

While endogenous BAG3 (Extended Data Fig. 6a) and fluorescently tagged BAG3^mRuby2 (Fig. 6b) accumulated in the cytoplasm, as expected, in response to proteasome inhibition or arsenite stress BAG3 co-aggregated with TDP-43 in the skein-like inclusions which formed (Fig. 6b, d). In such proteotoxic stress-induced TDP-43 skein-like structures, HSPB1 and HSPB8 were also enriched (Fig. 6e). Moreover, forced increase in BAG3 led to more MG132-induced skein-like structures containing TDP-43, which were also enriched in BAG3 (Fig. 6b–c, Extended Data Fig. 6b). Furthermore, an increased amount of BAG3 became insoluble following application of proteotoxic stress (Extended Data Fig. 4b–c).

To assess if BAG3, HSPB8, or HSPB1 promoted formation of TDP-43 skein-like inclusions, we transfected cells with siRNAs to lower accumulation of each (Extended Data Fig. 6c–e) and induced TDP-43 inclusion formation with arsenite exposure. While a decrease in HSPB1 promoted TDP-43-containing skein-like inclusions (Fig. 6f–h), reduction in BAG3 or HSPB8 sharply reduced their formation, with small, round granule-like inclusions forming instead (Fig. 6f–h).

Reduced HSP70 activity promotes TDP-43 skein-like structure formation

BAG3 has been reported to regulate HSP70 refolding activity in a concentration-dependent manner⁴⁷, with inhibition of HSP70 activity at high BAG3 levels alleviated by expression of a HSP70 binding deficient mutant of BAG3 (BAG3^R480A)⁴⁷. Indeed, in cells exposed to arsenite stress, increased BAG3 level drove assembly of more skein-like inclusions of cytoplasmic TDP-43^ΔNTD-Clover (Extended Data Fig. 6f, g), accompanied by reduction in HSP70 activity (measured with a luciferase assay) (Extended Data Fig. 6i). The BAG3-dependent increase in stress-induced TDP-43 skein-like inclusions was abolished by comparable accumulation of the HSP70 binding-deficient BAG3^R480A (Extended Data Fig. 6b, h), which (as expected) did not affect HSP70 activity (Extended Data Fig. 6i). Correspondingly, expression of fluorescently tagged BAG3 (but not similarly tagged BAG3^R480A) increased the sensitivity of TDP-43^ΔNTD-Clover expressing cells to HSP70 inhibition (Extended Data Fig. 6j).

Furthermore, nearly complete inhibition of HSP70 activity (by addition (50 μM) of the HSP70 inhibitor VER1550008) drove rapid collapse of nuclear anisosomes²⁰ (initially comprised of shells enriched in RNA binding deficient TDP-43 and liquid cores of HSP70) into a single large, intranuclear aggregate (Extended Data Fig. 7a,b). Addition of 1/10^th the inhibitor level to produce much milder HSP70 inhibition did not affect the morphology and liquidity of anisosomes, with no induction of large round granules (Extended Data Fig. 7a–b). However, in cells with cytoplasmic TDP-43^ΔNTD-Clover, mild HSP70 inhibition led to a faster and increased formation of skein-like inclusions in response to proteasome inhibition or arsenite stress (Extended Data Fig. 7c–f).

BAG2 mediates rounded, droplet-like TDP-43 inclusions

Each NEF member competes with the others for binding to HSP70 and it has been suggested that each has its own client proteins⁴⁸. Included here is our previous demonstration²³ that BAG2 binds to rounded, de-mixed, cytoplasmic condensates formed by full-length TDP-43, consistent with reports of stress-induced BAG2 condensation directing clients to proteasome degradation⁴⁹. To test if other NEFs compete with BAG3 to bind to TDP-43 and affect TDP-43 skein-like inclusions, fluorescently tagged variants of BAG1 (BAG1S, BAG1M, and BAG1L), BAG2, and HSPH1 were expressed in TDP-43^ΔNTD-Clover expressing cells (Fig. 7a). BAG1S, BAG1M, and HSPH1 were not enriched on TDP-43 skein-like inclusions upon proteotoxic stress from proteosome inhibition or arsenite exposure (Fig. 7b–d, Extended Data Fig. 8a–c). Similarly, the primarily nuclear BAG1L did not influence formation of cytoplasmic TDP-43 inclusions under stress conditions, but did recruit nuclear TDP-43 to the periphery of the nucleolus in an HSP70 binding-dependent manner (Extended Data Fig. 8d–f). An increased level of BAG2 drove cytoplasmic TDP-43 into round inclusions in response to proteotoxic stress (Fig. 7e) in a manner dependent on both its binding to HSP70 and its oligomerization coiled-coil domain (e.g., instead of rounded droplets, TDP-43^ΔNTD-Clover formed skein-like inclusions upon overexpression of the HSP70 binding-deficient BAG2^I160A (Extended Data Fig. 8g) or the oligomerization-deficient BAG2⁴⁹ (Extended Data Fig. 8h). Even in the handful of cells expressing elevated BAG2 and in which skein-like TDP-43 inclusions were formed, BAG2 continued to be recruited into rounded BAG2-positive droplets (Fig. 7f–g).

Figure 7. Different BAG proteins facilitate NTD-truncated TDP-43 to form morphologically different inclusions.

(a) Schematic representation of the structures of major HSP70 NEFs in human cells. (b) Representative images of cells co-expressing HSPH1^mRuby2 and TDP-43^ΔNTD-Clover before and after treatment with 100 μM NaAsO₂ or 10 μM MG132. (c) Representative images of cells co-expressing of BAG1s^mRuby2 and TDP-43^ΔNTD-Clover before and after treatment with 100 μM NaAsO₂ or 10 μM MG132. (d) Representative images of cells expressing BAG1S^mRuby2 and TDP-43^ΔNTD-Clover before and after treatment with 100 μM NaAsO₂ or 10 μM MG132. (d) Representative images of cells expressing BAG1M^mRuby2 and TDP-43^ΔNTD-Clover before and after treatment with 100 μM NaAsO₂ or 10 μM MG132. (e) Representative images of cells expressing BAG2^mRuby2 and TDP-43^ΔNTD-Clover before and after treatment with 100 μM NaAsO₂ or 10 μM MG132. (f) Representative images showing both BAG2-depenent TDP-43 granule-like inclusions and BAG2-independent TDP-43 skein-like inclusions in cells expressing BAG2^mRuby2 and TDP-43^ΔNTD-Clover after 4 hours of 100 μM NaAsO₂ treatment. (g) Quantification of cells expressing BAG2^mRuby2 and TDP-43^ΔNTD-Clover that form skein-like inclusions, granule-like inclusions, or both types of inclusions after 4 hour of 100 μM NaAsO₂ or 10 μM MG132 treatment. Cell number: 111 for the NaAsO₂ treated group and 110 for the MG132 treated group.

Discussion

The discovery that intrinsically disordered domains (IDRs) can induce the phase separation of RNA-binding proteins^{8, 9}, including TDP-43^10–15, provoked many efforts to determine how RNA-binding proteins (and the chaperones responsible for generating and maintaining their folding) create phase separated, de-mixed condensates under conditions conducive to phase separation^{11, 26, 50}. Here we have shown that in addition to IDRs, the N-terminal oligomerization domain of TDP-43 mediates multivalent interactions that enhance its phase separation, regulate TDP-43 aggregation, and influence the morphology of TDP-43 inclusions. We have discovered that the BAG3 and HSPB8 (probably complexed together⁵¹) mediate co-aggregation of TDP-43 with numerous actin-bundling proteins, including actinins and filamins. The skein-like inclusions formed by TDP-43 and actin-binding proteins are associated with actin filaments, and the skein-like morphology of these inclusions depends on the integrity of actin filaments.

Increased expression of filamin A (Fig. 5g), similar to the increased expression of filamin A in ALS motor neurons³³, exacerbates toxicity in cells expressing TDP-43 without its N-terminal domain, while reduction in filamin mitigates neurodegeneration caused by cytoplasmic TDP-43 accumulation in a Drosophila TDP-43 model. Indeed, expression of multiple filamins, including filamin A, increases both in the aging brain⁵² and in FTLD-TDP frontal cortex^{53, 54}. Filamin A has also been reported to enhance the aggregation of tau⁵⁵. When combined with a reported increase in F-actin in brain during aging⁵⁶ or in tau-mediated neurodegeneration⁵⁷, elevation of these aggregation-prone actin bundling proteins and/or disruption of proteostasis may play a general role in different age-associated neurodegenerative diseases.

When coupled with our identification that HSP70 NEFs recruit TDP-43 into different cellular structures (Fig. 7), our findings offer experimental support that different HSP70 NEFs can recruit clients and HSP70 into different cellular compartments for refolding⁴⁸, degradation^{43, 58, 59}, or aggregation. The collective evidence supports a model (Extended Data Fig. 9) in which action of these NEFs on specific HSP70 variants can drive (and/or regulate) formation of TDP-43-containing inclusions of divergent morphologies. Accumulation of different NEFs in various cell types, which compete with each other for binding to HSP70 and its substrates⁴⁸, may underlie the formation of diverse proteinopathies, including regulating the aggregation of tau^{58, 60–63}, α-synuclein^{62, 64}, and others⁶⁵. Moreover, mutations in the genes encoding BAG3, HSPB8 and HSPB1 are causative of motor neuron diseases^66–68 (Extended Data Fig. 5b). Finally, increase in BAG3 during aging^{59, 70} points to a potential mechanism underlying age-dependent proteinopathy formation through aging-associated regulation of neuronal NEFs. These results collectively demonstrate that modifications on TDP-43 and its binding partners, especially HSP70 and its NEFs, affect formation of different types of TDP-43 inclusions.

Methods

All experiments performed in this study were approved by the institutional ethical review committee of University of California, San Diego.

Plasmids

Plasmid information is provided in Supplementary Table 4. The vectors were constructed using Gibson assembly or double-restriction digestion cloning methods. The uniport ID of the three BAG1 isoforms are Q99933–1 (BAG1L), Q99933–3 (BAG1M) and Q99933–4 (BAG1S).

Cell culture and stable cell line construction

Cell lines used in this paper are: HEK293T (ATCC: CRL-11268), U2OS (ATCC: HTB-96) and the human inducible NGN2, ISL1, and LHX3 iPSC line, which was a gift from Dr. Michael Ward’s group at NIH. Routine maintenance of these model cell lines follows the guideline provided by ATCC. In brief, U2OS and HEK293T cells were cultured in complete DMEM supplemented with 10% fetal bovine serum, while iPSC cells were cultured on Matrigel-coated plates in mTeSRTM plus medium.

Lentivirus was produced in HEK293t cells by transfection of lentiviral plasmid and packaging plasmids pMD2.G and psPAX2 using TransIT-VirusGEN^® Transfection Reagent (Mirus, MIR6705). Two days post transfection, the culture medium containing the lentivirus was filtered through a 0.45 μm filter and used to infect U2OS cell line. Two days post infection, the medium was replaced with a medium containing 20 μg/ μL blasticidin or 2 μg/ μL puromycin for selection. Single clones of TRE-TDP-43^ΔNTD-Clover cell line were sorted using SH800S Cell Sorter (Sony). The co-expression cell lines are generated on the same clone of TRE-TDP-43^ΔNTD-Clover cell line to make sure similar expression of TDP-43^ΔNTD-Clover across the lines.

All the cell lines are authenticated by UCSD IRB approval for project 16155. There is no mycoplasma contamination obtained from a regular test for mycoplasma quarterly.

Neuron differentiation

iNGN2-ISL1-LHX3 iPSCs(Fernandopulle et al., 2018⁶⁹) are first induced to differentiate to neural precursor cells by culturing in induction medium (DMEM/F12, 1x N2 supplement, 1x GlutaMAX, 1x non-essential amino acids, 0.2 μM compound E, 2 μg/mL doxycycline, 10 μM ROCK inhibitor Y-27632) for 1 day and then replaced with induction medium (no ROCK inhibitor) for another day. Then neural precursor cells are then treated with Accutase and plated onto 8-well chamber slides (iBidi, 80827) or 24-well #1.5 glass-like polymer bottom plates (Cellvis, P24–1.5P) that are coated with poly-L-ornithine and laminin with neural culture medium (Neurobasal medium, 1xN2 supplement, 1xB27 supplement, 1x GlutaMAX, 1x non-essential amino acids, 10 ng/mL BDNF, 10 ng/mL GDNF, 1 μg/mL laminin). The medium is replaced every two-three day.

Neuron infection

Filtered lentivirus solution from HEK293T culture is concentrated using the Lenti-X Concentrator (Takara, 631231) according to the manufacturer’s instructions. The pellets are resuspended in 20 mM Hepes, pH 7.4, at 1/100th of the original volume. The concentrated lentivirus solution is then added to the 14-day differentiated neuronal culture. After 1–2 day of infection, the lentivirus is removed, and expression is checked after 3–4 days. Experiments are conducted 7–14 days post-infection.

Proximity labeling and enrichment of biotinylated protein

Prior to labeling, U2OS cells were treated with the indicated reagents and permeabilized with 50 μg/mL digitonin in PBS for 4 minutes. After one wash with PBS, reaction buffer containing 100 μM biotin phenol (Iris-Biotech, 41994–02-9) and 1 μM H2O2 was added, and the reaction proceeded for 1 minute. The reaction was immediately quenched with ice-cold quenching buffer (1x PBS, 10 mM sodium azide, 10 mM sodium ascorbate, 2.5 mM Trolox). The cells were scraped into the quenching buffer and centrifuged at 3000 rpm for 3 minutes. The cells were then washed three times with the quenching buffer, each time centrifuged at 3000 rpm for 3 minutes to collect the cell pellets. Cells were lysed in lysis buffer (100 mM NaPO4, pH 8.0, 8 M urea, 1% SDS, 10 mM sodium azide, 10 mM sodium ascorbate, 5 mM Trolox, 10 mM TCEP) and passed through an insulin syringe 15 times to shear DNA. After sonication in a Diagenode bioruptor sonication system for 10 minutes (30 s on and 30 s off) at 4 °C, protein lysates were cleared by centrifugation. Protein concentration was measured using the 2-D Quant Kit (GE Healthcare, Cat# 80648356), following the manufacturer’s instructions. After alkylation with 20 mM iodoacetamide for 15 minutes, 1 mg of protein samples were aliquoted and equilibrated to the same volume with lysis buffer. After dilution with an equal volume of ddH2O to reduce the concentration of urea to 4 M and SDS to 0.5%, the samples were incubated with streptavidin magnetic AccuNanobeads (Bioneer, Cat# TA-1015–1) at 4°C overnight.

Protein digestion and TMT labeling

After three washes with wash buffer 1 (100 mM TEAB, PH 8.0, 4 M Urea, 0.5% SDS) and four washes with wash buffer 2 (100 mM TEAB, PH 8.0, 4 M urea), the beads were resuspended in 100 mM TEAB, 2 M urea, supplemented with 10 ng/uL trypsin and 5 ng/uL Lys-C for pre-digestion at 37 °C on a thermo-mixture shaking at 1,000 rpm. The pre-digested products were collected, and an additional 10 ng/ μL trypsin was added for overnight digestion at 37 °C. Digested peptides from each sample were labeled with TMT six-plex labeling reagents (Thermofisher, Cat# 90061) according to manufacturer’s instruction. Briefly, TMT reagents were solubilized in anhydrous acetonitrile and added to peptides from each sample according to the labeling design in Supplementary Table 1. After 1-hr reaction at RT, 5% hydroxylamine was added and incubated for 15 mins to quench the reaction. Equal volumes of peptides from each sample in the same group were pooled together and concentrated using a SpeedVac to remove acetonitrile. The samples were acidified with formic acid (1%, final concentration) and desalted using Pierce C18 spin columns (Thermofisher, 89870).

Liquid chromatography-Mass spectrometry analysis

The TMT labeled samples were analyzed on an Orbitrap Eclipse mass spectrometer (Thermo Fisher). Samples were injected directly onto a 25 cm, 100 μm ID column packed with BEH 1.7 μm C18 resin (Waters). Separation was performed at a flow rate of 300 nL/min on an nLC 1200 (Thermo Fisher). Buffer A consisted of 0.1% formic acid in 5% acetonitrile and B consisted of 0.1% formic acid 80% acetonitrile. The gradient used was 0–25% B over 75 min, followed by an increase to 40% B over 30 min, an increase to 100% B over another 10 mins, and held at 100% B for 5 mins, for a total run of 120 mins.

Peptides were eluted directly from the tip of the column and nano-sprayed into the mass spectrometer by applying 2.5 kV voltage at the back of the column. The Eclipse was operated in data-dependent mode. Full MS1 scans were collected in the Orbitrap at 120k resolution. The cycle time was set to 3 seconds, during which the most abundant ions per scan were selected for CID MS/MS in the ion trap. MS3 analysis with multi-notch isolation (SPS3) was utilized for detection of TMT reporter ions at 7.5k resolution⁷¹. Monoisotopic precursor selection was enabled, and dynamic exclusion was used with an exclusion duration of 60 seconds.

Quantitative mass spectrometry data analysis

The raw data was processed by Rawconverter⁷² to extract MS2 and MS3 spectra, with correction of each precursor ion peak to its monoisotopic peak when appropriate. MS2 and MS3 spectra were searched against a complete human protein database downloaded from Uniprot, with the addition of APEX2 and Clover protein sequences, using the search algorithm ProLuCID⁷³. The searching parameters were: precursor mass tolerance of 50 ppm, fragment ion tolerance of 500 ppm for CID spectra and of 20 ppm for HCD spectra, minimum peptide length of 6 amino acids and static modifications for carbamidomethylation of cysteine and TMT tags on lysine residues and peptide N-termini (+229.162932 Da). Identified PSMs were filtered to an FDR of ≤1% at the PSM level with DTASelect2⁷⁴. The FDR was calculated based on the number of PSMs that matched sequences in the reverse decoy database. TMT quantification of reporter ions from MS3 spectra is done by Census2⁷⁵ with a filter of over 0.6 for isobaric purity. Normalized intensity based on weighted normalization was used to calculate the ratio of reporter ions corresponding to the indicated groups. Ratios of each protein from two forward labeling groups and two reverse labeling groups (Supplementary Table 1) were used to calculate P-value through student t-test, two-sided, no adjustment. The volcano plot was generated using the R package.

Human post-mortem tissues

Human tissues were obtained from the UCSD ALS tissue repository that was created following HIPAA-compliant informed consent procedures approved by Institutional Review Boards (either Benaroya Research Institute, Seattle, WA IRB# 10058 or University of California San Diego, San Diego, CA IRB# 120056) and de-identified at time of acquisition. Informed consent was obtained on all subjects and there was no compensation.

Tissue samples were obtained from patients who met the modified El Escorial criteria for definite ALS. Control nervous systems were obtained from non-neurological patients when life support was withdrawn, or from patients on hospice. Tissues were acquired using a short-postmortem interval acquisition protocol usually under 6 hours. Tissues were immediately dissected in the autopsy suite, placed in labelled cassettes and fixed in neutral buffered formalin for at least 2 weeks before being dissected and paraffin embedded for indefinite storage. Patient information is provided in Supplementary Tables 5–8.

Histology analysis in postmortem tissue

On day one, PPFE sections were deparaffinized with Citrisolv (FISHER brand #04–355-121) and hydrated through a series of ethanol dilutions. Sections were permeabilized with 1% FBS (Atlanta Biologicals #511150) and 0.2% Triton X-100 (Sigma #65H2616). Following permeabilization, antigen retrieval was performed in a high pH solution (Vector # H- 3301) in a pressure cooker for 20 min at 120 °C. Next, sections were blocked with 2% FBS in 1 × PBS for 60 min and then incubated with primary antibody overnight. Primary antibodies were diluted in 2% FBS in 1X PBS. On day two, after three washes with 1 × PBS, slides were incubated with secondary antibodies diluted in 2% FBS in 1X PBS for 60 min at room temperature and further washed three times with 1 × PBS. CNS auto-fluorescence was quenched with 0.1% Sudan Black in 70% ethanol for 15 seconds. Slides were cover-slipped using ProLong Gold Antifade Mountant with or without DAPI. We analyzed two to four 8- μm sections per patient.

siRNA transfection

ON-TARGETplus SMARTpool siRNAs (Horizon Discovery) targeting BAG3, HSPB1, HSPB8 or HSPA1A were transfected into U2OS cells at a concentration of 3 fmol/5,000 cells using Lipofectamine RNAiMAX Transfection Reagent (Thermofisher, 13778075) for three or four days before the experiment.

Cell death assay

Cell survival rate was quantified using CellTiter-Glo^® luminescent cell viability assay (Promega, G7570) according to the manufacturer’s instruction. Briefly, the CellTiter-Glo^® glow buffer was mixed with the CellTiter-Glo^® glow substrate, and 100 μL of the reagent was added to each well of a 96-well plate after equilibrating the plate to room temperature for 30 mins. After 2 minutes of shaking and 10 minutes of incubation, the luminescence was measured using plate reader.

Luciferase assay for chaperone activity measurement

Luciferase activity was measured using Firefly Luciferase Assay Kit 2.0 (Biotium #30085–1) according to the manufacturer’s instruction. Briefly, U2OS cells were plated on 12-well plates at 1×105 cells per well and infected with AAV1-Luciferase at an MOI of 50,000 for three days before one day of doxycycline treatment to induce the expression of TDP-43ΔNTD-Clover. The cells were treated with sodium arsenite as indicated, then lysed in 250 μL of lysis buffer per well. For luciferase measurement, 20 uL of lysates were used, and freshly prepared luciferin-containing assay buffer was added. The firefly luminescence was measured with a Tecan Spark plate reader using an integration time of 1 second.

Live cell imaging

U2OS cells were plated onto the 96-well plate (Greiner #655866) in DMEM medium without phenol red and treated as indicated. Images were taken by a CQ1 benchtop high-content analysis with a 40x/1.2 objective under a constant CO2 flow.

Fluorescence recovery after photobleaching (FRAP)

U2OS cells for FRAP experiments were cultured on an 8-well chamber slide (iBidi, 80827) in DMEM without phenol red supplemented with 10% fetal bovine serum (FBS) and Antibiotic-Antimycotic (Thermofisher, 15240062). Expression of TDP-43ΔNTD-Clover was induced for 24 hours as indicated by adding 1 μg/mL doxycycline to the culture medium. FRAP analysis was performed on a Zeiss LSM880 Aryscan microscope with 40x/1.2 W objective. The intensity of the fluorescent signal is controlled in the detection range through changing the laser power, digital gain and off-set. Bleaching was conducted by 488-nm or 561-nm line correspondingly and the laser power and iteration of bleaching are optimized to get an efficient bleaching effect. Fluorescence recovery was monitored at 1 second intervals for 2 minutes. In the focal-bleach experiment, roughly half (partial bleach) or all (full bleach) of de-mixing structures was photobleached to determine the molecular mobility with diffuse pool or inside the structure.

The FRAP data were quantified using ImageJ. The time series of fluorescence intensity of condensates were calculated. The averaged relative intensity and standard error were plotted to calculate dynamics.

Immunofluorescence

For immunofluorescence, U2OS cells were cultured on 8-well chamber slides (iBidi, 80827) or #1.5 glass-like polymer bottom 96-well plate (Cellvis, P96–1.5P) in DMEM supplemented with 10% fetal bovine serum (FBS) and Antibiotic-Antimycotic (Thermofisher, 15240062). After the indicated treatments, cells were fixed with 4% PFA in PBS and permeabilized with 0.2% Triton X-100 for 10 minutes. After blocking with 2% BSA in PBS, 0.05% Triton X-100 for 2 hours, cells were incubated for 1 hour at room temperature with primary antibody in blocking solution. After three washes with PBS, cells were incubated with Alexa647-labeled, Cy3-labeled or Alexa488-labeled secondary antibody at 1:500 dilution or Alexa647-streptavidin at 1:1000 dilution for APEX labeling experiments, in blocking solution for 30 minutes at room temperature. After three washes with PBS and DAPI staining, cells were kept in PBS for imaging. Primary antibody dilutions for used in this study were as follows: anti-BAG3 (rabbit polyclonal, Novus Biologicals, NBP2–27398, Lot#102119) at 1:300, anti-HSPB1 (goat polyclonal, Santa Cruz, sc-1048; rabbit polyclonal, Stressmarq, SMC-161B) at 1:300, anti-HSPB8 (rabbit polyclonal, Stressgen, NBP2–87836) at 1:300, anti-filamin A (mouse monoclonal, Sigma, MAB1680-C) at 1:300, anti-Actinin (rabbit polyclonal, Proteintech, #11313–2-AP) at 1:300.

Stochastic Optical Reconstruction Microscopy (STORM) Imaging

U2OS cell lines with inducible TDP-43^ΔNTD-Clover were cultured on 8-well chamber slides (iBidi, 80827) and treated with doxycycline for 1 day to induce the expression of TDP-43^ΔNTD-Clover. The cells are then treated with the indicated stressor, permeabilized with 50 ug/mL digitonin, 0.5 mM EGTA, 3 mM MgCl₂ in PBS for 3 minutes, and fixed with 3% PFA, 0.1% glutaraldehyde in PBS for 10 minutes at room temperature. After treatment with 0.1% NaBH4 in PBS for 7 minutes, the cells were washed three times with 1x PBS. The samples were then blocked with 3% BSA, 0.2% triton for 90 mins before staining with nGFP-Alexa647 in 2% BSA, 0.05% Triton for 30 mins. After washing three times with 2% BSA, 0.05% Triton and once with 1x PBS, the cells were post-fixation with fixation buffer for 10 mins. This was followed by three washes with 1× PBS, and during the second wash, actin was stained with phalloidin-CF568 for 30 minutes. TetraSpeck beads solution (1:500 in PBS) was then added and allowed to settle for 15 minutes before removal and a final wash with PBS.

Before STORM imaging, prepare the STORM imaging buffer freshly. To make the GLOX buffer, mix 14 mg of glucose oxidase, 50 μL of 17 mg/mL catalase, and 200 μL of Buffer A (10 mM Tris, pH 8.0, 50 mM NaCl). Then mix 10 μL of GLOX and 100 μL of 1 M MEA into 1 mL of Buffer B (50 mM Tris, pH 8.0, 10 mM NaCl, 10% glucose) to create the STORM imaging buffer. Add the imaging buffer to the chamber before imaging.

Imaging was performed on an Eclipse Ti2-E microscope with N-STORM illumination (Nikon) using a 405 nm laser for activation and 647 nm and 561 nm lasers for reporting. The imaging mode was continuous, capturing 10,000 frames per channel. Images were analyzed with N-STORM analysis software, which detected molecules based on the average peak intensity of the molecule spot in each channel. Drift correction with beads was used to align molecules across different frames.

RNA labeling

RNA labeling in the cells were done with Click-iT^™ RNA Alexa Fluor^™ 594 Imaging Kit (Thermofisher, C10330) following the manufacturer’s instructions. Briefly, 0.5 mM 5-ethynyl uridine was added to the cell culture and incubated overnight. Then the cells were treated with proteotoxic stressors as indicated in the figures and fixed for click chemistry labeling and imaging.

Cell lysate fractionation

U2Os cells were cultured in 6-well plates and each well of cells were lysed in 300 μL of RIPA buffer (25 mM Tris•HCl pH 7.6, 150 mM NaCl, 1% NP-40, 1% sodium deoxycholate, 0.1% SDS) supplemented with 1000 unit/mL Benzonase (Millipore, 70746) and 1x Halt^™ Protease and Phosphatase Inhibitor Cocktail (Thermofisher, 78440). The cell lysates were passed through 28G insulin syringes (Fisher Scientific, 14–829-1A) eight times and lysed on ice for 20 minutes. The cell lysates were then centrifuged at 18,000 g for 20 minutes and the supernatants were collected as the soluble fraction. The pellet fractions were washed once with 500 μL of RIPA buffer and solubilized in 100 μL of 1xLDS (Thermofisher, B0007) running buffer as the insoluble fraction.

Immunoprecipitation

U2OS cells were treated with 100 μM sodium arsenite or 10 μM MG132 for 4 hr before being collected with a cell scraper. Cell pellets from one 10-cm dish were lysed in 1 mL of 1x IP lysis buffer (25 mM Hepes pH 7.4, 150 mM NaCl, 1 mM EDTA, 1% NP-40 and 5% glycerol) supplemented with 1000 unit/mL Benzonase (Millipore, 70746) and 1x Halt^™ Protease and Phosphatase Inhibitor Cocktail (Thermofisher, 78440). The cell lysates were passed through an 28G Insulin syringe (Fisher Scientific, 14–829-1A) eight times and lysed on ice for 20 minutes. The release of TDP-43 inclusions in cell lysates were checked under microscope. Then cell debris was removed by spinning at 1000 g, 15 minutes at 4 °C, twice. The inclusion-containing supernatant was collected and incubated with GFP-Trap Magnetic Particles M-270 (Chromotek) at 4 °C for 4 hours. After incubation, 5 uL of samples were taken to examine the efficiency of GFP binding on the beads under fluorescence microscope. Then, the beads were collected by magnet stand and washed with 1 mL of lysis buffer four times before elution with 1xLDS (Thermofisher, B0007) running buffer at 80 °C for 30 minutes on a thermomixer. One percent of input and ten percent of IP sample were used for western analysis.

Western blot

Protein samples were loaded onto 12% Criterion^™ TGX^™ Precast Midi Protein Gel (Bio-Rad) and transferred onto Immobilon^®-FL PVDF Membrane (Millipore). Then membrane was blocked with Intercept Blocking Buffer (LICOR) and incubated with primary antibodies: BAG3 rabbit polyclonal antibody, Novus Biologicals, NBP2–27398, Lot#102119, at 1:1000 dilution, HSPB8 rabbit polyclonal antibody, Stressgen, NBP2–87836, at 1:1000, filamin A mouse monoclonal antibody, Sigma, MAB1680-C, at 1:1000, Actinin rabbit polyclonal antibody, Proteintech, #11313–2-AP) at 1:1000, HSPB1 monoclonal antibody, StressMarq, 5D12-A12 at 1:1000 dilution, HSP70 monoclonal antibody, Enzo ADI-SPA-810 at 1:1000 dilution, GAPDH polyclonal antibody, Cell Signaling 14C10, 1:2000 dilution and mouse monoclonal, Abcam, ab8245, TDP-43 C-terminal antibody, Proteintech, 12892–1-AP at 1:1000 dilution, actin rabbit antibody, Abcam, ab8227, at 1:1000 dilution, overnight at 4 °C. After four washes with TBST, membrane was incubated with secondary antibody (1:20,000 for IRDye^® 800CW donkey anti-mouse, LICOR and IRDye^® 680RD donkey anti-rabbit IgG, LICOR or 1:5,000 for HRP donkey anti-mouse, Invitrogen). Then the membrane was imaged by LICOR Odyssey^® Imager.

Drosophila eye degeneration assay

All Drosophila stocks were maintained on yeast-cornmeal-syrup food at 25°C. UAS-TDP43ΔNLS/CyO; GMR-GAL4/TM6,Tb flies³⁶ were crossed to Canton-S or UAS-cher RNAi flies (BDSC #35755). UAS-TDP43ΔNLS/+; GMR-GAL4/+ and UAS-TDP43ΔNLS/+; GMR-GAL4/cher RNAi offspring were selected and aged at 25 °C for 10 days. Eye degeneration was scored using a previously described method³⁶. Briefly, points were added if there were necrotic patches, complete loss of interommatidial bristles, retinal collapse, loss of ommatidial structure, and/or depigmentation.

Plotting

Bar plots were created with Graphpad Prism 9. The heat map was generated at http://www.heatmapper.ca/expression/. Protein interaction network was generated with Cytoscape_3_9_1.

Statistics & Reproducibility

No statistical methods were used to pre-determine sample sizes, but our sample sizes are similar to those reported in previous publications¹². Data distribution was assumed to be normal, but this was not formally tested. No data was excluded from the analysis. The experiments were not randomized. The Investigators were not blinded to allocation during experiments and outcome assessment.

Extended Data

Extended Data Fig. 1. Skein-like inclusion formation does not rely on the RNA binding capacity of TDP-43, but the RRM1 and low complexity domains are required.

(a) Representative images of cells expressing TDP-43 ^ΔNTD-Clover treated with NaAsO₂ for 1 to 4 hours. Cells are stained with G3BP1 to show stress granules from more than three biological replicates. An arrow indicates a cell with TDP-43 ^ΔNTD-Clover in both stress granules and skein-like inclusions at 2 hours of arsenite treatment. (b-e) Representative live images of cells expressing TDP-43^{90–414-Clover} from more than three biological replicates (b), TDP-43^{90–414–2KQ-Clover} from more than three biological replicates (c), TDP-43^{102–414–2KQ-Clover} from more than three biological replicates (d), TDP-43^{NLSm-S48D-Clover} from more than three biological replicates (e) before and after 4 hours of 100 μM NaAsO₂ treatment. Black-white inverted fluorescent images are shown. n=4 biological replicates representing 149, 214, 146, 215 cells (TDP-43^NLSm-Clover) and 37, 55, 58, 37 cells (TDP-43^{NLSm-S48D-Clover}), respectively. Error bar: SD. (f) FRAP analysis of TDP-43 skein-like inclusions induced by MG132 from more than three biological replicates. High magnification images show no fluorescence recovery of photo bleaching points within 1 min. (g) Representative images of MG132-induced skein-like TDP-43 inclusions preserved after mild cell permeabilization. (h-i) Representative live images of cells expressing TDP-43^{174–414-Clover} (h) and TDP-43 ^{ΔNTD- Δ320–343-Clover} (i) before and after 4 hours of 100 μM NaAsO₂ treatment from more than three biological replicates. Black-white inverted fluorescent images are shown. N=4 biological replicates for TDP-43^{NLSm-S48D-Clover}. (j) Representative images of cells expressing TDP-43 ^ΔNTD-V5 under conditions of no stress, NaAsO₂, or MG132 treatment from more than three biological replicates. TDP-43 ^ΔNTD-V5 is indicated by V5 staining.

Extended Data Fig. 2. Concentration-dependent regulation of formation of skein-like and granule-like TDP-43 inclusions.

(a) Schematic of the experiment design to test the presence of different levels of cytoplasmic full-length TDP-43 alongside consistent expression levels of TDP-43 ^ΔNTD-Clover. The aim is to determine how the balance between full-length TDP-43 and the C-terminal fragment influences the formation of different types of TDP-43 de-mixing structures. (b) Representative live images of cells expressing similar levels of TDP-43 ^ΔNTD-Clover and either low or high levels of full-length RNA-binding-competent TDP-43^NLSm-mRuby2 before and after 4 hours of NaAsO₂ treatment from more than three biological replicates. (c) Representative live images of cells expressing similar levels of TDP-43 ^ΔNTD-Clover and either low or high levels of full-length RNA-binding-incompetent TDP-43^{NLSm/2KQ-mRuby2} before and after 4 hours of NaAsO₂ treatment from more than three biological replicates.

Extended Data Fig. 3. TDP-43 skein-like inclusions are associated with actin filaments and actin binding proteins.

(a) Representative image showing no enrichment of mIFP on arsenite-induced TDP-43 inclusions in iPSC-derived motor neurons. (b-c) Representative images of skein-like TDP-43 inclusions colocalized with filamin A (FLNA, detected by immunostaining) in the non-permeabilized or permeabilized cells from more than three biological replicates. (d) Representative images of skein-like TDP-43 inclusions colocalized with actin filaments stained by phalloidin from more than three biological replicates. (e) Representative images of skein-like TDP-43 inclusions colocalized with actinin, detected by immunostaining from more than three biological replicates. (f) Representative images of cells withs TDP-43 skein-like inclusions stained with Click-iT RNA imaging kit from more than three biological replicates.

Extended Data Fig. 4. TDP-43, actin binding proteins, and BAG3 are enriched in the insoluble fractions after cell exposure to proteotoxic stress.

(a) Schematic diagram illustrating the biochemical analysis of soluble and insoluble fractions of cells expressing TDP-43 ^ΔNTD-Clover or Clover. (b) Analysis of the levels of TDP-43 ^ΔNTD-Clover, endogenous TDP-43, FLNA, β-Actin, Actinin, BAG3, HSPA1A, GAPDH in the soluble fractions of cells under no stress, 100 μM NaAsO₂ or 10 μM MG132 treatment. Data are from three biological replicates. (c) Analysis of the levels of TDP-43 ^ΔNTD-Clover, endogenous TDP-43, FLNA, β-Actin, Actinin, BAG3, HSPA1A, GAPDH in the insoluble fractions of cells under no stress, 100 μM NaAsO₂ or 10 μM MG132 treatment. Data are from three biological replicates. (d) Quantification of the levels of TDP-43 ^ΔNTD-Clover, endogenous TDP-43 and FLNA in both soluble and insoluble fractions. n=3 independent replicates. Error bar: SD. (e) Specific interaction of TDP-43 with FLNA and actin under NaAsO₂ or MG132 stress conditions detected by immunoprecipitation.

Extended Data Fig. 5. BAG3 is associated with skein-like structures of TDP-43.

(a) Representative images of motor neurons from ALS patients displaying different types of TDP-43 pathological structures. A dot-plot quantifies the percentage of motor neurons exhibiting different types of TDP-43 pathological structures. Number of motor neurons quantified are 400, 164, 114, 105, 90, 144, 70, 70, 80 from nine patients. Error bar: SD. (b) Schematic representation of the interaction domains of HSP70, BAG3 and HSPB8, HSPB1. (c-e) Expression levels of BAG1, BAG2, BAG3, BAG4, BAG5, BAG6 and HSPH1 in different cell types within the adult human spinal cord. Data are adapted from scRNA-seq datasets published by Yadav, et al., in Nature Neuroscience⁴⁴ (c, d) and by Gautier, et al., in Neuron⁴⁵ (e). Center lines show the medians; box limits indicate the 25th and 75th percentiles as determined by R software; whiskers extend 1.5 times the interquartile range from the 25th and 75th percentiles, outliers are represented by dots. (f) Expression levels of BAG1, BAG2, BAG4, BAG5 and HSPH1 in different cell types within the adult mouse spinal cord. Images are ISH data from the Allen mouse spinal cord atlas (https://mousespinal.brainmap.org/, https://mousespinal.brainmap.org/imageseries/show.html?id=100024593, https://mousespinal.brain-map.org/imageseries/show.html?id=100038126, https://mousespinal.brain-map.org/imageseries/show.html?id=100037229, https://mousespinal.brain-map.org/imageseries/show.html?id=100048120, https://mousespinal.brain-map.org/imageseries/show.html?id=100021840 ).

Extended Data Fig. 6. An elevated level of BAG3 increases the formation of TDP-43 skein-like structure and the sensitivity of TDP-43^DNTD-Clover expressing cells to HSP70 inhibition in a HSP70-binding dependent manner.

(a) Representative images showing the localization of endogenous BAG3 in cells stained with BAG3 antibody from more than three biological replicates. (b) Western blot analysis of BAG3 in cells without the mRuby2-tagged BAG3 transgene or in cells stably expressing BAG3^mRuby2 or BAG3^R480A-mRuby2. (c-d) Western blot demonstrating the reduction of BAG3 (c) and HSPB1 (d) following siRNA treatment. (e) Decrease of HSPB8 following siRNA treatment detected by immunostaining and quantification of mean fluorescent intensity in cells. Volcano plots show the mean fluorescent intensities of HSPB8 and HSPB1 in cells treated with siHSPB8 or control siRNA. Center lines show the medians; box limits indicate the 25th and 75th percentiles as determined by R software; whiskers extend 1.5 times the interquartile range from the 25th and 75th percentiles, outliers are represented by dots. (f) Schematic illustrating the assessment of BAG3 association with TDP-43 ^ΔNTD-Clover on skein-like structures. (g-h) Representative live images of cells co-expressing TDP-43 ^ΔNTD-Clover with BAG3^mRuby2 (g) or BAG3^R480A-mRuby2 (h) upon 100 μM NaAsO₂ treatment. (i) Measurement of HSP70 activity before, during, and after the removal of arsenite stress from more than three biological replicates. Error bar: SD. (j) The effect of expressing BAG3^mRuby2 and BAG3 ^R480A-mRuby2 on cell resistance to HSP70 inhibition. Data are from N=3 biological replicates. Number of quantified cells per replicate are 136, 168 and 202 (TDP-43 ^ΔNTD-Clover, no Dox); 140, 140, 156 (TDP-43 ^ΔNTD-Clover, Dox); 275, 286, 258 (TDP-43 ^ΔNTD-Clover and BAG3^mRuby2, no Dox); 234, 178, 233 (TDP-43 ^ΔNTD-Clover and BAG3^mRuby2, Dox); 198, 161, 210 (TDP-43 ^ΔNTD-Clover and BAG3 ^R480A-mRuby2, no Dox); 175, 166, 164 (TDP-43 ^ΔNTD-Clover and BAG3 ^R480A-mRuby2, Dox), respectively. Error bar: SD.

Extended Data Fig. 7. A mild decrease of HSP70 activity promotes the formation of TDP-43 skein-like inclusions induced by proteotoxic stress.

(a) Assessment of HSP70 chaperone activity under different levels of HSP70 inhibitor by examining the disruption of nuclear TDP-43 liquid annisosome from more than three biological replicates. (b) Quantification of cells exhibiting large aggregated TDP-43 granules under different levels of HSP70 inhibitor treatment. N=3 biological replicates. Cell numbers are 82, 117, 127 (5 μM VER155008); 114, 90, 144 (10 μM VER155008); 110, 86, 78 (50 μM VER155008). Error bar: SD. (c) Representative live images of cells expressing TDP-43 ^ΔNTD-Clover treated with 10 μM MG132 and low concentrations of HSP70 inhibitor. (d) Representative live images of cells expressing TDP-43 ^ΔNTD-Clover treated with 100 μM NaAsO2 and low concentrations of HSP70 inhibitor from more than three biological replicates. (e) Quantification of cells with skein-like TDP-43 inclusions under the conditions described in (c) and (d). N=3 biological replicates. Cells number quantified per replicate are 205, 177, 220 (2 hr, NaAsO2, no HSP70i); 197, 159, 204 (3 hr, NaAsO2, no HSP70i); 210, 163, 211 (4 hr, NaAsO2, no HSP70i); 200, 225, 205 (2 hr, NaAsO2, 5 μM HSP70i); 225, 225, 224 (3 hr, NaAsO2, 5 μM HSP70i); 232, 217, 220 (4 hr, NaAsO2, 5 μM HSP70i); 234, 213, 172 (2 hr, NaAsO2, 10 μM HSP70i); 231, 211, 190 (3 hr, NaAsO2, 10 μM HSP70i); 253, 232, 197 (4 hr, NaAsO2, 10 μM HSP70i); 195, 207, 185 (2 hr, MG132, no HSP70i); 149, 138, 130 (3 hr, MG132, no HSP70i); 65, 86, 76 (4 hr, MG132, no HSP70i); 152, 173, 168 (2 hr, MG132, 5 μM HSP70i); 137, 138, 119 (3 hr, MG132, 5 μM HSP70i); 106, 96, 77 (4 hr, MG132, 5 μM HSP70i); 199, 203, 171 (2 hr, MG132, 10 μM HSP70i); 175, 113, 129 (3 hr, MG132, 10 μM HSP70i); 108, 74, 86 (4 hr, MG132, 10 μM HSP70i). Error bar: SD.

Extended Data Fig. 8. Association of BAG2 and BAG1L with TDP-43 de-mixed structures requires their ability to bind HSP70.

(a-c) Quantification of cells co-expressing TDP-43 ^ΔNTD-Clover and HSPH1^mRuby2 (a), BAG1S^mRuby2 (b) or BAG1M^mRuby2 (c) that form skein-like inclusions, granule-like inclusions, or both types of inclusions after 4 hour of 100 μM NaAsO₂ or 10 μM MG132 treatment. Number of cells quantified from one representative biological replicate: 284 cells (HSPH1^mRuby2, NaAsO₂), 247 cells (BAG1S^mRuby2, NaAsO₂), 252 cells (BAG1M^mRuby2, NaAsO₂); 255 cells (HSPH1^mRuby2, MG132), 236 cells (BAG1S^mRuby2, MG132), 193 cells (BAG1M^mRuby2, MG132). (d) Representative live images of cells co-expressing TDP-43 ^ΔNTD-Clover and BAG1L^mRuby2 before and after treatment with 10 μM MG132 and 100 μM NaAsO₂ from more than three biological replicates. (e) Representative live images of cells co-expressing TDP-43 ^ΔNTD-Clover and BAG1L^2RA-mRuby2 before and after treatment with 10 μM MG132 and 100 μM NaAsO₂ from more than three biological replicates. (f) Quantification of cells from one representative biological replicate co-expressing TDP-43 ^ΔNTD-Clover and BAG1L^mRuby2 that form skein-like inclusions, granule-like inclusions, or both types of inclusions after 4 hour of 100 μM NaAsO₂ or 10 μM MG132 treatment. Number of cells quantified from one representative biological replicate: 230 cells in NaAsO₂ group, 160 cells in MG132 group. (g) Representative live images of cells co-expressing TDP-43 ^ΔNTD-Clover and BAG2^I60A-mRuby2 before and after treatment with 10 μM MG132 and 100 μM NaAsO₂ from more than three biological replicates. (h) Representative live images of cells co-expressing TDP-43 ^ΔNTD-Clover and BAG2 ^{Δ20–61-mRuby2} before and after treatment with 10 μM MG132 and 100 μM NaAsO₂ from more than three biological replicates.

Extended Data Fig. 9. Working model for formation of different, morphologically distinct types of TDP-43 inclusions regulated by TDP-43 oligomerization and different NEFs of HSP70.

Created in BioRender. Lu, S. (2025) https://BioRender.com/ppkmwcc

Data availability

All the numerical source data are provided in the source data table. Quantitative mass spectrometry raw data have been deposited to the public database MassIVE with the dataset identifier MSV000098098 and ProteomeXchange with identifier PXD068255. The data can be downloaded from ftp://MSV000098098@massive-ftp.ucsd.edu. Previously published laser captured motor neuron RNA-seq dataset is available from the Gene Expression Omnibus under accession code GSE76220³³. Previously published human spinal cord snRNA seq datasets are available from the Gene Expression Omnibus under accession codes GSE190442⁴⁴, GSE222322⁴⁴ and GSE228778⁴⁵, and at searchable website https://vmenon.shinyapps.io/humanspinalcord/⁴⁴. Further information on the post-mortem samples analyzed are available from the UCSD ALS tissue repository (contact information please find in https://health.ucsd.edu/specialties/neuro/specialty-programs/als-clinic/pages/default.aspx). The images of ISH data for BAG3, BAG1, BAG2, BAG4, BAG5 and HSPH1 are from the Allen mouse spinal cord atlas⁴⁶ (https://mousespinal.brain-map.org/imageseries/show.html?id=100020020, https://mousespinal.brainmap.org/imageseries/show.html?id=100024593, https://mousespinal.brain-map.org/imageseries/show.html?id=100038126, https://mousespinal.brain-map.org/imageseries/show.html?id=100037229, https://mousespinal.brain-map.org/imageseries/show.html?id=100048120, https://mousespinal.brain-map.org/imageseries/show.html?id=100021840). More representative images are deposited to Figshare at https://figshare.com/s/91c0eaea2f9c1f14a4b0. All other data supporting the findings of this study are available from the corresponding author on reasonable request.