Dear Editor,

TAR DNA-binding protein 43 kDa (TDP-43) undergoes liquid–liquid phase separation (LLPS) and forms reversible, cytoprotective nuclear bodies (NBs) under stress.1 Abnormal liquid-to-solid phase transition condenses TDP-43 into irreversible pathological fibril, which is associated with neurodegenerative disorders including amyotrophic lateral sclerosis (ALS) and frontotemporal degeneration (FTD).2,3 However, the mechanisms by which cells maintain the highly aggregation-prone protein in the liquid-like phase and prevent TDP-43 NBs from aggregation under stressed conditions remain elusive. Molecular chaperones play a crucial role in maintaining protein homeostasis. Heat shock protein (Hsp) 70 can interact with TDP-434 and increase of Hsp70 suppresses TDP-43-mediated toxicity in fly models.5 Importantly, ALS patients with TDP-43 aggregates exhibit significantly decreased Hsp70 levels,6 suggesting that Hsp70 maintains TDP-43 proteostasis and that its dysregulation may be involved in pathological aggregation of TDP-43 RNA–protein granules in ALS and related diseases.

To investigate the role of Hsp70 in TDP-43 NBs, first we examined whether Hsp70 was associated with TDP-43 in NBs. Both endogenous and red fluorescent protein (RFP)-tagged Hsp70 was diffuse and mostly localized in the cytoplasm (Supplementary information, Fig. S1). Upon stress (250 μM arsenite, 30 min), RFP-Hsp70 signal increased in the nucleus and was co-localized with TDP-43 NBs (Fig. 1a, b). As a control, overexpression (OE) of an untagged RFP was dispersed in both the nucleus and the cytoplasm, and it did not co-localize with stress-induced TDP-43 NBs (Supplementary information, Fig. S2a–d). Similarly, another Hsp70 family protein, Hsc70, was also recruited into TDP-43 NBs when stressed (Supplementary information, Fig. S2e–h). Furthermore, we generated a stable knock-in (KI) cell line, Hsc70EGFP-KI, in which the coding sequence of enhanced green fluorescent protein (EGFP) was inserted into the N-terminus of Hsc70 (Fig. 1c). The Hsc70EGFP-KI protein expressed by its own promoter at the endogenous level was also co-localized with TDP-43 NBs in stressed cells (Fig. 1d, e). Purified full-length (FL) TDP-43 fused with the stabilizing maltose-binding protein (MBP)7 or cleaved by tobacco etch virus (TEV) protease phase separated in vitro and formed liquid droplets (LDs) (Fig. 1f; Supplementary information, Fig. S3). Hsp70 did not phase separate on its own but co-phase separated with TDP-43. Similar co-LLPS behavior was also observed with Hsc70 in vitro (Supplementary information, Fig. S4). Furthermore, the dynamics of TDP-43 LDs dropped significantly after 40 min of in vitro incubation, which was markedly suppressed by addition of Hsp70 (Fig. 1g; Supplementary information, Fig. S5).

Fig. 1: Hsp70 chaperones TDP-43 in liquid-like phase and prevents it from pathological aggregation.
figure 1

a, b Representative images and the intensity profile along the indicated line of HeLa cells expressing RFP-Hsp70 with TDP-43-HA in response to stress (250 μM of arsenite, 30 min). Arrows, co-localization of Hsp70 with TDP-43 NBs. See also Supplementary information, Figs. S1 and S2. c Western blot analysis confirming expression of Hsc70EGFP-KI in the stable KI cell line. The anti-Hsp70s antibody recognizes both Hsp70 and Hsc70. d, e Representative images of Hsc70EGFP-KI and TDP-43-HA in control (no stress) or stressed Hsc70EGFP-KI cells (250 μM of arsenite, 30 min). Arrows, co-localization of Hsc70EGFP-KI with stress-induced TDP-43 NBs. f In vitro co-LLPS of TDP-43-MBP (50 μM) and Hsp70 (10 μM) at indicated condition. See also Supplementary information, Fig. S3. g FRAP analyses of TDP-43-MBP LDs formed in the in vitro LLPS assay. n = 6. See also Supplementary information, Fig. S5. h, i Representative images showing the effect of si-Hsp70s on the assembly of TDP-43 NBs in stressed HeLa cells. G3BP, a marker for SGs. See also Supplementary information, Fig. S7. jm Representative images showing stress-induced EGFP-TDP-43 NBs in HeLa cells (j) and the regions (dashed circle) photo-bleached in the FRAP assays in km at indicated time. The fluorescent intensity (FI) recovery curves are shown on the right, n = 8. n The CCK-8 assay showing enhanced TDP-43 cytotoxicity by si-Hsp70s in prolonged stress. n = 3. o The thioflavine T (ThT) fluorescence assay of TDP-43 LCD (20 μM) with different concentrations of Hsp70 as indicated. n = 3. p Schematic of the main functional domains in TDP-43. q Negative-staining TEM images of the ThT samples at 30 h in o. r Residue-specific intensity changes of signals in the 2D 1H-15N HSQC spectra of 15N-labeled TDP-43 LCD with different concentrations of Hsp70 as indicated. The CR in the TDP-43 LCD is indicated in the dashed box. See also Supplementary information, Fig. S9. su The FRAP assay evaluating the impact of OE of Hsp70s on the dynamics of different types of EGFP-TDP-43-K181E NIs. The regions (dashed circles) of the same size are photo-bleached. n = 20 each group, pooled results of five independent biological repeats. See also Supplementary information, Fig. S14a–e. v, w Representative images of EGFP-TDP-43 K181E NIs co-expressed with RFP (v) or RFP-Hsp70 (w) in 293T cells that are co-immunostained for pTDP-43 (Ser409/410). x, y Quantification of the percentage of cells showing pTDP-43+ NIs (x) and the relative size of pTDP-43+ NIs (normalized to area labeled by DAPI) (y). n = ~25 cells in x and 6 in y. See also Supplementary information, Fig. S14f, g. Data are means ± SEM; Two-way ANOVA in g, km, su, Student’s t-test in n, x, y; *P < 0.05 and ***P < 0.001; ns, not significant. Scale bars, 5 μm in a, b, d, e, hj, su, 10 μm in f, v, w, 1 μm in km and 200 nm in q.

To understand the physiological significance of co-localization and co-phase separation of Hsp70 with TDP-43 NBs, we sought to determine the impact of downregulation of Hsp70 on TDP-43 NBs. Among the mammalian Hsp70 family members, HSPA8 (encoding Hsc70) was expressed the most abundantly in HeLa cells, whereas HSPA1A (encoding Hsp70) increased the most dramatically in response to stress (Supplementary information, Fig. S6). We then used small interference RNA (siRNA) to downregulate HSPA1A and HSPA8 simultaneously (shown as si-Hsp70s for simplicity). It significantly reduced the assembly of TDP-43 NBs but not stress granules (SGs) (Fig. 1h, i; Supplementary information, Fig. S7a–h), suggesting a specific function of Hsp70 in assisting the assembly of TDP-43 NBs. Indeed, we showed that Hsp70 promoted TDP-43 LLPS in vitro (Supplementary information, Fig. S7i, j).

We then examined how knockdown (KD) of Hsp70 affected the dynamics of TDP-43 NBs by the FRAP assay (Fig. 1j–m). Interestingly, although si-Hsp70s did not alter the liquid-like, dynamic feature of TDP-43 NBs with a transient stress (250 μM arsenite, 30 min) (Fig. 1k), si-Hsp70s reduced the dynamics of TDP-43 NBs with prolonged stress especially at 120 min (Fig. 1l, m). Furthermore, Hsp70 KD made cells more vulnerable to TDP-43-mediated cytotoxicity, leading to decreased cell viability with prolonged stress (Fig. 1n). Together, these data indicated that the recruitment of Hsp70 to TDP-43 NBs not only promoted the assembly of TDP-43 NBs but also helped to maintain them in the highly dynamic, liquid-like state during cellular stress.

TDP-43 consists of an N-terminal domain (NTD), two RNA recognition motifs (RRMs), and a prion-like, low complexity domain (LCD). We then purified the truncated LCD and ∆LCD proteins of TDP-43 to determine which region mediated the co-LLPS with Hsp70. Hsp70 promoted the LLPS of TDP-43 LCD dramatically but only showed minimal effect on that of TDP-43 ∆LCD (Supplementary information, Fig. S8a–i). TDP-43 LCD was identified as the key region in mediating liquid-to-solid phase transition and pathological fibril formation of TDP-43.8 Indeed, TDP-43 LCD LDs underwent rapid maturation with drastically decreased dynamics (Supplementary information, Fig. S8j). Strikingly, with Hsp70, the dynamics of the TDP-43 LCD LDs was well maintained even after an hour of incubation. In contrast, Hsp70 could not change the gel-like state of TDP-43 ∆LCD droplets (Supplementary information, Fig. S8j, k). Furthermore, Hsp70 potently prevented TDP-43 LCD from forming amyloid fibril (Fig. 1o–q).

We next titrated Hsp70 to 15N-labeled TDP-43 LCD and performed solution NMR spectroscopy. The 2D 1H-15N HSQC spectra showed a significant signal attenuation upon Hsp70 titration. Strikingly, the residues with dramatic changes mainly clustered on the conserved region (CR, 320–340 aa) (Fig. 1r; Supplementary information, Fig. S9), which was previously shown to adopt a transient α-helical conformation and was essential for mediating the LLPS and amyloid aggregation of TDP-43.8,9,10 To rule out the possibility that the intensity attenuation of CR is due to CR–CR homotypical interaction which is crucial for mediating TDP-43 LLPS,9 we prepared A326P mutation previously identified to disrupt CR–CR interaction (α-helical conformation of CR) and abolish TDP-43 LLPS (Supplementary information, Fig. S10a–i).9 A326P TDP-43 LCD showed similar intensity attenuation pattern as wild-type (WT) TDP-43 LCD upon Hsp70 titration (Supplementary information, Fig. S10j), supporting the direct binding of Hsp70 to CR of TDP-43 LCD.

In addition, we demonstrated that Hsp70-NTD exhibited a stronger activity than C-terminal domain (CTD) of Hsp70 in binding the CR, promoting LLPS and preventing fibrillation of TDP-43 LCD (Supplementary information, Fig. S11). Since Hsp70-NTD is an adenosine triphosphate (ATP)-binding domain, we next examined the effect of ATP in this process. ATP did not markedly affect co-LLPS of TDP-43 and Hsp70 (Supplementary information, Fig. S12a), though the interaction between TDP-43 and Hsp70 was modestly reduced (Supplementary information, Fig. S12b, c). Furthermore, although ATP moderately diminished the chaperone activity of Hsp70 in suppressing TDP-43 LCD fibrillation, Hsp70 could still potently delay the maturation and solidification of TDP-43 LCD droplets (Supplementary information, Fig. S12d–g). Since AMPPNP, a non-hydrolysable analog of ATP, exhibited the similar effect as ATP (Supplementary information, Fig. S12d–g), we speculated that the effect of ATP here was more likely due to a competition for Hsp70-NTD binding with TDP-43 LCD rather than the ATPase activity of Hsp70-NTD. Together, our data indicated that Hsp70 directly interacts with the CR of TDP-43 LCD mainly via the NTD, which stabilizes TDP-43 in the liquid-like phase and prevents it from amyloid fibrillation.

To demonstrate that the direct interaction between Hsp70 and TDP-43 mediated the anti-aggregation effect, it would be ideal to test on TDP-43 mutants with disrupted Hsp70–TDP-43 interface. Unfortunately, mutation of the interface severely impaired the capability of TDP-43 to phase separate in vitro and in cells (Supplementary information, Fig. S13).9,10 Alternatively, we assessed the effect of Hsp70 OE against pathological aggregation of TDP-43 NBs. K181E of TDP-43 is an ALS/FTD-causal mutation that is unable to bind RNA and formed abundant aggregates.11 Indeed, cells expressing TDP-43-K181E formed small or large nuclear inclusions (NIs) in the absence of stress, and Hsp70 was co-localized with both small and large NIs. However, only large amyloid NIs were hyperphosphorylated (pTDP-43), a histopathological hallmark of ALS (Supplementary information, Fig. S14),12 and large TDP-43-K181E NIs recovered the most slowly in the FRAP assay. More importantly, OE of Hsp70s (HSPA1A and HSPA8 together) not only markedly increased the dynamics of the large NIs (Fig. 1s–u), but also significantly reduced the pTDP-43 levels of TDP-43-K181E in the nucleus (Fig. 1v–y; Supplementary information, Fig. S14).

Hsp70 OE suppressed the transition of TDP-43-K181E NIs to hyperphosphorylated, pathological aggregation without affecting the number of cells forming TDP-43 NIs (Fig. 1x, y), suggesting that Hsp70 plays a more prominent role in keeping TDP-43 NBs in the liquid-like, non-pathological state than trigging the assembly of TDP-43 NBs. Intriguingly, Hsp70 is decreased in ALS patients associated with TDP-43 pathology6 and OE of TDP-43 in both fly and mouse motor neurons also leads to reduced levels of Hsp70.13 Thus, decrease of Hsp70 may be both a consequence and a contributing factor of TDP-43 pathogenesis, which may further worsen the liquid-to-solid transition and aggregation of TDP-43 in ALS and related diseases.

Our recent study revealed that the long non-coding RNA NEAT1 preferentially binds to the RRM1, which provides a “nucleation core” and promotes the assembly of TDP-43 NBs.1 In the present work, we demonstrate that Hsp70 binds to the aggregation-prone region of TDP-43 LCD, which maintains TDP-43 NBs in the highly dynamic, liquid-like phase and prevents its liquid-to-solid phase transition (Supplementary information, Fig. S15). In line with our earlier1 and present studies, a recent paper reported that mutations of the RRMs cause spontaneous formation of intranuclear TDP-43 annuli associated with Hsp70.14 Collectively, our finding of Hsp70 being a potent “dynamics keeper” of TDP-43 NBs provides novel insight into the mechanism of how cells maintain aggregation-prone proteins such as TDP-43 in “condensed but highly dynamic” NBs during cellular stress.