TGFβ-like DAF-7 acts as a systemic signal for autophagy regulation in C. elegans

Autophagy involves the sequestration of cytoplasmic materials in a double-membrane autophagosome and its delivery to lysosomes for degradation (Feng et al., 2014; Mizushima et al., 2011; Zhao and Zhang, 2018). Under a variety of stress conditions, autophagy provides energy for the survival of cells. Autophagy also removes potentially toxic materials such as protein aggregates and damaged organelles to maintain cellular homeostasis.

During multicellular organism development, autophagy participates in diverse processes such as stress resistance, cell fate determination and tissue remodeling (Mizushima and Komatsu, 2011; Yang and Zhang, 2014). Studies of yeast and cultured cells have identified numerous factors that integrate various stressors with the autophagic machinery to modulate autophagy activity (Russell et al., 2014). In multicellular organisms, the stress response is coordinately controlled between different tissues/organs to ensure the maintenance of cellular homeostasis at an organismal level. Autophagy itself participates in a systemic starvation response by controlling the generation and release of cytokines, hormones, ATP, and other molecules to mediate the cross-talk between tissues/organs in energy metabolism and metabolic adaptive responses (Kaushik et al., 2011; Fenouille et al., 2017). Autophagy activity is also systemically coordinated to ameliorate deleterious effects elicited by locally imposed stresses such as nutrient restriction, and also to maintain cell, tissue, and organism homeostasis. Up-regulation of the autophagy protein Atg1 or AMP-activated protein kinase (AMPK) in flies induces autophagy in the target tissue and also elicits a systemic autophagy response in other tissues (Ulgherait et al., 2014). Malignant tumors in fly eyes trigger autophagy in the tumor microenvironment and also in distant tissues (Katheder et al., 2017). In Caenorhabditis elegans, amino acids modulate the activity of the metabotropic glutamate receptor homologues MGL-1 and MGL-2 in the AIY and AIB interneurons, respectively, through a yet-to-be-identified signal to regulate systemic autophagy responses (Kang and Avery, 2009). Several signaling pathways have been identified to act nonautonomously to influence autophagy in neighboring cells or in a short-range environment. For example, the fly complement orthologue Mcr (macroglobulin complement-related) functions through the immune receptor Draper in a cell-nonautonomous manner to regulate autophagy in neighboring cells (McPhee et al., 2010; Lin et al., 2017). Systemic intercellular signaling that elicits the long-range autophagy response in distant cells/tissues/organs remain largely unknown. Glucagon and glucagon-like peptide 1 have been shown to regulate autophagy in liver and β cells (Kanasaki et al., 2019). Insulin-like peptides also modulate autophagy during animal development. In C. elegans, insulin-like signaling regulates entry into dauer diapause. Under harsh environmental conditions, the insulin family member DAF-28 is down-regulated, accompanied by autophagy induction (Li, et al., 2003; ,Meléndez et al., 2003). The systemic autophagy response to tissue-specific AMPK/Atg1 induction is also associated with reduced insulin-like peptide levels in the brain (Ulgherait et al., 2014). It remains unclear whether hormones and growth factors are directly involved in systemic autophagy regulation, especially in response to environmental changes.

C. elegans is enclosed within a cuticle structure, which is essential for protection against environmental damage and pathogens, and also for body morphology and integrity. The outer layer of the cuticle contains alternating parallel circumferential bands, known as annuli and annular furrows, which comprise two discrete interacting groups of collagens (McMahon et al., 2003). Loss of function of annular furrow collagen genes, including dpy-2, -7, and -10, manifests as cuticle damage and elicits innate immune response and osmotic resistance phenotypes (Lamitina et al., 2006; Wheeler and Thomas, 2006; Pujol et al., 2008a,b). Here we demonstrated that loss of annular furrow collagens triggers systemic induction of autophagy. The cilia of sensory neurons, whose endings are embedded in the subcuticle or exposed to the external environment via openings in the cuticle (Inglis et al., 2007), are required for autophagy induction. The TGFβ molecule DAF-7, which is secreted from ciliated neurons, acts as a systemic factor to activate autophagy in distal cells. Our study revealed a circuit for sensing and coordinating autophagy regulation in different tissues in response to cuticle damage during worm development.

During C. elegans development, EPG-7 acts as a scaffold protein to facilitate autophagic degradation of the p62 homologue SQST-1 (Fig. 1, A, A′, B, and H; Tian et al., 2010; Lin et al., 2013). We performed genetic screens and identified a mutation, bp1237, that caused a dramatic reduction in the number of SQST-1::GFP aggregates in epg-7 mutants (Fig. 1, C and H). Simultaneous depletion of lgg-1 restored the accumulation of SQST-1 aggregates in epg-7; bp1237 mutants (Fig. 1, D and H). Levels of SQST-1::GFP protein were also decreased in epg-7; bp1237 mutants (Figs. 1 I and S1 U), while sqst-1 mRNA levels remained unchanged (Fig. S1 A). bp1237 exhibited a shorter and stout dumpy phenotype, known as Dpy, that is similar to animals lacking cuticle collagen genes (Fig. S1, B and C). Genetic mapping and transformation rescue experiments demonstrated that bp1237 is an allele of dpy-10 (Fig. 1, E and H; and Fig. S1, D and E). Accumulation of SQST-1::GFP aggregates in atg-18(bp594) hypomorphic mutants and in rpl-43(bp399) mutants, but not in null mutants of autophagy genes required for autophagosome formation, was also suppressed by loss of dpy-10 activity (Fig. 1, F–H; and Fig. S1, F–I, N, and O).

The level of the lipidated C. elegans Atg8 homologue LGG-1 (LGG-1-II) has been widely used to monitor autophagy activity (Zhang et al., 2015a). In dpy-10 mutants, both LGG-1-I (unlipidated) and LGG-1-II levels were reduced (Figs. 1 J and S1 V), which could be explained by elevated autophagy activity or impaired autophagy at the induction step. To distinguish between these two possibilities, we determined the levels of LGG-1 in epg-5 dpy-10 double mutants. In epg-5 mutants, nondegradative autolysosomes accumulate, resulting in accumulation of LGG-1 puncta and elevated levels of LGG-1-II (Fig. 1, K, L, N, and O; and Fig. S1 W; Tian et al., 2010). Accumulation of LGG-1 puncta in epg-5 mutants is ameliorated by a block in autophagy induction and aggravated by an elevation of autophagy induction (Guo et al., 2014; Wang et al., 2016). The number of GFP::LGG-1 puncta and levels of LGG-1-II in epg-5 mutants were greatly enhanced by simultaneous loss of dpy-10 (Fig. 1, M–O; and Fig. S1 W). mRNA levels of six autophagy genes analyzed were not up-regulated in dpy-10 mutants (Fig. S1 Q). These results indicate that loss of dpy-10 activity elevates autophagy activity.

According to their function in the formation of annular furrows, Dpy genes are classified into two groups (McMahon et al., 2003). Annuli are absent in mutants of dpy-10 family genes (i.e., dpy-2, dpy-3, dpy-7, dpy-8, and dpy-10), but not in mutants of dpy-13 family genes (i.e., dpy-5 and dpy-13; McMahon et al., 2003; Page and Johnstone, 2007). An allele of dpy-2, bp1242, was also isolated in the aforementioned screen (Fig. S1, J–K and P). Loss of function of other dpy-10 family genes, but not dpy-5, suppressed accumulation of SQST-1::GFP aggregates in epg-7 mutants (Fig. S1, L–M′ and P).

We characterized autophagy activity in tissues that are in direct contact with or separate from the cuticle. To facilitate the analysis, transgenes expressing SQST-1::GFP in specific tissues were used, including hypodermal cells (bpIs267; Fig. 2, A, B, and B′), intestinal cells (bpIs262; Fig. 2, A, E, and E′), and body wall muscles (bpIs287; Fig. 2, A, H, and H′). SQST-1::GFP aggregates accumulated in epg-7 mutants carrying bpIs267 or bpIs262 (Fig. 2, C, F, J, K, M, and N; and Fig. S1, X and Y), and this accumulation was dramatically reduced by simultaneous loss of function of dpy-10, but not dpy-5 (Fig. 2, D, G, J, K, M, and N; and Fig. S1, R, S, X, Y, and A1–B1′). Animals carrying bpIs287 accumulated SQST-1::GFP aggregates in muscle cells (Fig. 2, H, H′, L, and O; and Fig. S1 Z), and this phenotype was further enhanced in epg-7 mutants (Fig. S1, D1 and D1′). dpy-10(bp1237), but not dpy-5, suppressed accumulation of SQST-1::GFP aggregates in bpIs287 and also in epg-7; bpIs287 animals (Fig. 2, I, L, and O; and Fig. S1, T, Z, C1–E1, and J1). Accumulation of SQST-1::GFP aggregates in the hypodermis and intestine in epg-5 mutants failed to be suppressed by dpy-10 inactivation (Fig. S1, F1–I1, K1, and L1). Thus, autophagic activity is activated in multiple tissues by inactivation of annular furrow–related collagen genes.

In C. elegans, the cilia of sensory neurons are involved in sensing the chemical and physical extracellular environments in C. elegans (Inglis et al., 2007). The formation and function of cilia depends on continuous trafficking of structural, regulatory, and signaling molecules by intraflagellar transport (IFT), which is mediated by a multiprotein complex. The cilia of amphid and phasmid neurons, labeled with a reporter for the IFT protein OSM-6, were normal in dpy-10 mutants (Fig. S2, A–D). Loss of function of che-3, which encodes a dynein heavy chain 1b isoform involved in retrograde IFT, caused no defect in degradation of SQST-1 and did not affect accumulation of SQST-1 aggregates in epg-7 mutants (Fig. 3, A, B, E, F, I, and I′). Loss of che-3 activity restored the accumulation of SQST-1::GFP aggregates in hypodermal and intestinal cells in dpy-10; epg-7 mutants (Fig. 3, C, G, L, and M). SQST-1::GFP aggregates also persisted in the muscle in che-3; dpy-10 mutants (Fig. 3, J and N). Loss of function of osm-3, which encodes the IFT motor kinesin-II homologue, also abolished the suppression effect of dpy-10(bp1237) (Fig. S2, E–M). These results suggest that cilia are involved in sensing the cuticle damage and activating autophagy in distal tissues in dpy-10 mutants.

We next sought the signal that is transduced from ciliated neurons to trigger the long-range autophagy response. Loss of function of genes involved in synaptic transmission or in generation and release of neuropeptides, including unc-13, unc-31, egl-3, and egl-21 (Richmond et al., 1999; Li and Kim, 2008), had no effect on the suppression effect of dpy-10 inactivation (Fig. S2, N–K1). In response to changes in environmental conditions, ciliated sensory neurons secrete DAF-7, a TGFβ-related ligand for the dauer TGFβ pathway. Once secreted, DAF-7 activates a canonical TGFβ signaling pathway in target tissues (Fig. 3 O; Hu, 2007). Loss of function of daf-7 had no effect on SQST-1::GFP removal and did not affect the defective degradation of SQST-1 in epg-7 mutants (Fig. 3, P, Q, X, Y, D1, and D1′). We found that loss of daf-7 activity restored the accumulation of SQST-1::GFP aggregates in the hypodermis and intestine of dpy-10; epg-7 mutants (Fig. 3, R, Z, I1, and J1) and in the muscle of dpy-10 mutants (Fig. 3, E1 and K1). The receptor encoding gene daf-1 was also required for the suppressing effect of dpy-10 inactivation (Fig. S3, A–I). In contrast, simultaneous depletion of the antagonistic coSMAD DAF-3 suppressed the accumulation of SQST-1::GFP aggregates in the hypodermis and intestine in dpy-10; daf-7; epg-7 mutants and in the muscle in dpy-10; daf-7 mutants (Fig. 3, T, B1, G1, and I1–K1). Thus, dauer TGFβ signaling is required for systemic autophagy activation in dpy-10 mutants.

Expression of daf-7 is mainly limited to the ASI pair of ciliated neurons, and the daf-7 expression level in these cells is modulated by the availability of food and pheromones, and also by infection (Fig. 3 L1; Hu, 2007; Meisel et al., 2014). The Pgpa-4::daf-7 transgene, in which daf-7 is specifically expressed in the ASI neurons, was sufficient to rescue the effect of daf-7(ok3125) on SQST-1::GFP aggregate accumulation in dpy-10; daf-7; epg-7 mutants as well as in dpy-10; daf-7 mutants (Fig. 3, S, A1, F1, and I1–K1). The fluorescence signal of a Pdaf-7::gfp transcriptional reporter was stronger in ASI neurons in dpy-10 mutants than in wild-type animals (Fig. 3, M1–O1). Expression of daf-3 specifically in the ASI neurons or in the muscle, however, failed to rescue the phenotype in the hypodermis in dpy-10; daf-7; daf-3; epg-7 mutants (Fig. 3, V, W, and I1). Accumulation of SQST-1::GFP aggregates in the hypodermis and intestine in dpy-10; daf-7; daf-3 epg-7 mutants and in the muscle of dpy-10; daf-7; daf-3 mutants was restored by expression of daf-3 driven by hypodermis-, intestine-, and muscle-specific promoters, respectively (Fig. 3, U, C1, and H1–K1). These results indicate that DAF-3 functions cell-autonomously in regulating autophagy in dpy-10 mutants. Simultaneous depletion of dpy-10 suppressed accumulation of SQST-1::GFP aggregates in the hypodermis and intestine in che-3; daf-3; epg-7 triple mutants and in the muscle in che-3; dpy-10; daf-3 mutants (Fig. 3, D, H, and K–N). Thus, TGFβ signaling acts downstream of cilia in activating autophagy in distal tissues in dpy-10 mutants.

We examined the role of the C. elegans AMPK catalytic subunit (AMPKα) orthologue, AAK-2, in autophagy activation in dpy-10 mutants. The expression of the functional AAK-2 reporter, AAK-2(aa1–321)::Tomato, was increased in dpy-10 mutants compared with control animals (Fig. 4, A–C). Phosphorylation of Thr172 in the activation loop of human AMPKα is required for AMPK activation, and the residues surrounding the phosphorylated Thr residue are conserved in AAK-2 (Mihaylova and Shaw, 2011). In dpy-10 mutants, levels of phosphorylated AAK-2, detected by the antibody recognizing Thr172-phosphorylated AMPK, were increased (Figs. 4 D and S3 J). Loss of function of aak-2 had no effect on SQST-1::GFP removal and did not affect SQST-1::GFP accumulation in epg-7 mutants (Fig. 4, E and F; and Fig. S3, K, K′, M, and M′). SQST-1::GFP aggregates accumulated in the hypodermis in dpy-10; epg-7; aak-2 mutants (Fig. 4, G and U). Expression of AAK-2 specifically in the hypodermis rescued this effect of aak-2 inactivation, suggesting that AAK-2 acts cell-autonomously to activate autophagy (Fig. 4, H and U). In dpy-10; daf-3 epg-7 aak-2 mutants, SQST-1::GFP aggregates accumulated in the hypodermis (Fig. 4, I and U), indicating that AAK-2 functions downstream of or in parallel to TGFβ signaling. SQST-1::GFP aggregates were still absent in the intestine of dpy-10; epg-7 aak-2 mutants and in the muscle of dpy-10; aak-2 mutants (Fig. S3, L, N, U, and V). Thus, AAK-2 is required for elevated autophagy activity in the hypodermis, but is dispensable in the intestine and muscle of dpy-10 mutants.

PMK-1 p38 MAPK signaling and the STAT-like protein STA-2 are involved in triggering the innate immune response in the hypodermis and intestine in response to environmental stresses (Pujol et al., 2008a,b; Dierking et al., 2011; Zhang et al., 2015b). Loss of function of pmk-1 and sta-2 had no effect on degradation of SQST-1 aggregates (Fig. 4, J, J′, K, N, N′, O, R, and R′; and Fig. S3, O, O′, Q, Q′, S, and S′). Loss of dpy-10 activity suppressed accumulation of SQST-1::GFP aggregates in the hypodermis in pmk-1; epg-7 mutants and sta-2; epg-7 mutants (Fig. S3, P, T, and W). In the intestine of dpy-10; epg-7 mutants, accumulation of SQST-1::GFP aggregates was restored by loss of activity of pmk-1 or sta-2 (Fig. 4, L, P, V, and W). Simultaneous daf-3 inactivation suppressed the accumulation of SQST-1::GFP aggregates in the intestine in dpy-10; pmk-1; epg-7 mutants and dpy-10; sta-2; epg-7 mutants (Fig. 4, M, Q, V, and W), suggesting that PMK-1 and STA-2 function upstream of or in parallel to DAF-3 in regulating autophagy in dpy-10 mutants. Accumulation of SQST-1::GFP aggregates was suppressed in muscle cells in dpy-10; pmk-1 mutants (Fig. S3, Q, Q′, R, and V). SQST-1::GFP aggregates persisted in the muscle in dpy-10; sta-2 mutants but were absent in dpy-10; sta-2; daf-3 mutants (Fig. 4, S, T, and X). Therefore, in dpy-10 mutants, PMK-1 is dispensable for autophagy activation in the hypodermis and muscle, while STA-2 is dispensable in the hypodermis but required in the intestine and muscle.

Loss of function of dpy-10 family members triggers the innate immune response, as shown by up-regulation of neuropeptide-like proteins, and also elevates expression of hyperosmotic stress response genes, such as gpdh-1, which encodes glycerol-3-phosphate dehydrogenase (Lamitina et al., 2006; Wheeler and Thomas, 2006; Pujol et al., 2008b). Loss of activity of osm-7, which activates the osmosensitive response, failed to suppress the accumulation of SQST-1::GFP aggregates in epg-7 mutants (Fig. S3, X–Y′). Up-regulation of Pnlp-29::GFP and Pgpdh-1::GFP in dpy-10 mutants remained unaffected by simultaneous depletion of che-3, daf-7, or aak-2 (Fig. S3, Z–N1). Therefore, functional cilia, dauer TGFβ signaling, and AMPK/AAK-2 are not involved in the innate immune response and hypertonic stress in dpy-10 mutants.

Here we revealed a circuit that senses and transduces the cuticle damage caused by loss of function of the cuticle annular furrow collagen genes to trigger systemic induction of autophagy (Fig. 4 Y). The nature of the signal in dpy-10 family gene mutants that is perceived by the ciliated sensory neurons has yet to be determined. DAF-7 secreted from the ASI pair of ciliated neurons acts as a systemic factor to activate a canonical TGFβ signaling pathway in distant tissues to activate autophagy in dpy-10 mutants. Different signaling pathways are involved in triggering expression of distinct stress response genes caused by the cuticle damage in annular furrow–related collagen gene mutants (Pujol et al., 2008a,b). Our study sheds new light on the systemic signals that elicit long-range autophagy responses to environmental stress for maintaining cell, tissue, and organism homeostasis.

The following strains were used in this work: aak-2(ok524), atg-18(bp594), che-3(cas495), daf-1(m40), daf-3(tm4698), daf-7(ok3125), dpy-2(bp1242), dpy-5(e61), dpy-10(bp1237), dpy-10(e128), egl-3(nr2090), egl-21(n476), epg-5(tm3425), epg-7(tm2508), osm-3(p802), pmk-1(km25), rpl-43(bp399), sta-2(ok1860), unc-13(e1091), unc-31(e169), bpIs151(Psqst-1::sqst-1::GFP+unc-76), bpIs267(Phyp7::sqst-1::gfp+unc-76), bpIs262(Pges-1::sqst-1::gfp+unc-76), bpIs287(Pmyo-3::sqst-1::gfp+unc-76), bpIs360(Prgef-1::sqst-1::gfp+Pord-1::rfp), bpIs345(Pgpdh-1::gfp+unc-76), adIs2122(lgg-1p::GFP::lgg-1+rol-6(su1006)), mnIs17(osm-6p::osm-6::gfp+unc-36), uthIs202(aak-2(intron 1)::aak-2(aa1-aa321)::Tomato::unc-54 3′UTR+rol-6(su1006)), and frIs7(nlp-29p::GFP+col-12p::dsRed).

epg-7 mutants carrying the bpIs151 transgene were used for ethyl methanesulfonate mutagenesis. bp1237 and bp1242 were isolated from a screen of ∼5,000 genomes for mutations that dramatically decreased the number of SQST-1::GFP aggregates compared with epg-7 single mutants. Both bp1237 and bp1242 were mapped to chromosome II by single nucleotide polymorphism mapping. A transgene containing fosmid WRM0636aD09 restored the accumulation of SQST-1::GFP in epg-7; bp1237 or epg-7; bp1242 mutants. bp1237 and bp1242 cause a Dpy phenotype. dpy-10 and dpy-2 are located in the region where bp1237 and bp1242 are mapped. A transgene expressing dpy-10 and dpy-2 rescued the phenotype in bp1237 and bp1242 mutants, respectively. The molecular lesions in bp1237 and bp1242 were determined by sequencing. bp1237 contains a deletion that removes 163 amino acids from DPY-10. bp1242 contains a glycine-to-arginine mutation at amino acid 126 in DPY-2.

RNAi bacterial clones were cultured on nematode growth medium agar plates containing 1 mM IPTG. L1 larval animals were plated onto RNAi feeding plates, and the F1 progeny were examined for the corresponding phenotype. For RNAi injection, single-strand RNA was synthesized by Large Scale RNA Production System T7 and SP6 kits (Promega, P1300 and P1280). Single-strand RNAs were then mixed and injected into animals. The F1 progeny were analyzed.

To generate tissue-specific daf-3 expression constructs, the y37a1.b5 promoter (2.9 kb; also known as the hyp7 promoter), hlh-1 promoter (2.9 kb), ges-1 promoter (3.3 kb), and gpa-4 promoter (3.0 kb) were amplified by PCR from genomic DNA. The PCR-amplified promoter region was inserted between the SphI and XmaI sites of the pPD49.26 vector, and the full-length daf-3 cDNA was then inserted between the KpnI and SacI sites. The constructs were confirmed by sequencing. The constructs were injected into animals at a concentration of 20 ng/µl with a comarker, Pord-1::rfp (50–100 ng/µl). At least three independent transgenic lines were analyzed for each rescue experiment.

To construct Pgpa-4::daf-7, 3.0-kb gpa-4 promoter was cloned into pPD49.26 at the PstI/XmaI sites, and 1.9-kb daf-7 genomic DNA was then inserted at the KpnI/SacI sites. To generate Pdaf-7::GFP reporter, 3.0-kb daf-7 promoter was cloned into pPD49.26-GFP at the PstI site.

Total RNA was extracted from synchronized animals using Trizol reagent. Quantitative PCR reactions were performed using UltraSYBR Mixture RT-PCR kits (CWBIO, CW2602M) and a PCR biosystems QuantStudio 7 Flex. Three independent experiments were performed. The results were analyzed with GraphPad Prism 5.

For quantification of the number of SQST-1::GFP aggregates, images were captured with a microscope (Zeiss M2) at 400× magnification. Images were captured at similar regions in the animals at the same developmental stage using the same exposure time. The number of SQST-1::GFP aggregates was quantified using ZEN lite 2011 and analyzed with GraphPad Prism 5. Five animals for each strain were used for quantification. For measurement of the fluorescence intensity of GFP::LGG-1 puncta, images were acquired at 630× magnification using a Zeiss LSM880 microscope. For quantification of the fluorescence intensity of Paak-2::AAK-2::Tomato, Pnlp-29::GFP, and Pgpdh-1::GFP, images were acquired at 100× magnification using a Zeiss M2 microscope. Images were quantified using ImageJ (National Institutes of Health) and analyzed with GraphPad Prism 5. At least five animals were analyzed. Immunoblotting results are representative of three independent experiments. The intensity of Western blot signals was quantified using ImageJ. Statistical comparisons were made using the two-tailed unpaired Student’s t test, and the results are shown as mean ± SEM; ns, no significant difference, P > 0.05; *, P < 0.05; **, P < 0.01; ***, P < 0.001.

Fig. S1 shows elevated autophagy activity in dpy-10 mutants. Fig. S2 shows the requirement of functional cilia in systemic autophagy induction in dpy-10 mutants. Fig. S3 shows the requirement of TGFβ signaling in systemic autophagy activation in dpy-10 mutants. Functional cilia, dauer TGFβ signaling, and AMPK/AAK-2 are not involved in the innate immune response and hypertonic stress response in dpy-10 mutants.

We are grateful to Dr. Isabel Hanson for editing work and Dr. Malene Hansen (Sanford Burnham Prebys Medical Discovery Institute, La Jolla, CA) for the AAK-2(aa1-321)::Tomato strain.

This work was supported by the following grants to H. Zhang: Beijing Municipal Science and Technology Commission (Z181100001318003), Chinese Ministry of Science and Technology 017YFA0503401), National Natural Science Foundation of China (31421002, 31561143001, 31630048, and 31790403), Strategic Priority Research Program of the Chinese Academy of Sciences (grant XDB19000000), and Key Research Program of Frontier Sciences, Chinese Academy of Sciences (grant QYZDY-SSW-SMC006).

The authors declare no competing financial interests.

Author contributions: H. Zhang designed the experiments. Y.J. Zhang and L.X. Qi performed the experiments. H. Zhang wrote the manuscript.