The progress in C9orf72 research: ALS/FTD pathogenesis, functions and structure

ABSTRACT

The hexanucleotide repeat (GGGGCC) expansion in C9orf72 is accounted for a large proportion of the genetic amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD). The hypotheses of how the massive G4C2 repeats in C9orf72 destroy the neurons and lead to ALS/FTD are raised and improving. As a multirole player, C9orf72 exerts critical roles in many cellular processes, including autophagy, membrane trafficking, immune response, and so on. Notably, the partners of C9orf72, through which C9orf72 participates in the cell activities, have been identified. Notably, the structures of the C9orf72-SMCR8-WDR41 complex shed light on its activity as GTPase activating proteins (GAP). In this manuscript, we reviewed the latest research progress in the C9orf72-mediated ALS/FTD, the physiological functions of C9orf72, and the putative function models of C9orf72/C9orf72-containing complex.

KEYWORDS:

• DENN domain
• autophagy
• membrane trafficking
• neurodegenerative
• GAP
• GEF
• C9orf72
• SMCR8
• WDR41
• lysosome
• mTOR

Introduction

The massive GGGGCC (G4C2) repeat in the first intron of C9orf72 has been demonstrated to be one of the genetic hallmarks of amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD) [1–3], two lethal and untreatable neurodegenerative diseases. As a potential diagnostic and therapeutic target for genetic ALS and FTD, C9orf72 has been extensively researched, and significant progress has been achieved in the past decade.

First, a notable headway in discerning the pathogenic mechanism of C9orf72-mediated ALS and FTD has been made. Two main hypotheses, the toxic gain-of-function and loss-of-function of C9orf72, have been suggested to elucidate the pathogenic mechanism of how the G4C2 repeats in C9orf72 damage the neurons [4–19]. A recent study suggested that C9orf72 loss-of-function exacerbates the toxic gain-of-function phenotype [3].

In addition, the signalling pathways and physiological processes in which C9orf72 participates have been gradually deciphered. Dozens of reports have demonstrated that C9orf72 participates in the regulation of macroautophagy (hereafter referred to as autophagy) through different mechanisms, e.g., membrane trafficking and transcription [20–24]. Interestingly, the lysosome is a critical organelle that mediates C9orf72 function and is highly correlated with the C9orf72-ALS/FTD [25–27]. Intriguingly, substantial evidence indicates that the massive G4C2 repeat expansion of C9orf72 and thus the C9orf72 protein impacts immune activity [27–29]. Nevertheless, the specific function of C9orf72 is not clear.

During the exploration of the function of C9orf72, several protein partners of C9orf72 have been identified, including but not limited to SMCR8, WDR41, the ULK1 complex, and RAB39B [21,22,25,26]. Notably, C9orf72, SMCR8, and WDR41 can form a stable complex that exerts a regulatory function on lysosomes and autophagy pathways [21,24–27]. Intriguingly, both C9orf72 and SMCR8 belong to Differentially Expressed in Normal and Neoplastic cells (DENN) domain family. The most well-defined function of DENN domain family proteins is as guanine nucleotide exchange factors (GEF) for small GTPases [30–33]. Bioinformatic analysis indicates that C9orf72, SMCR8, FNIP1/2 (folliculin-interacting protein), and FLCN (folliculin) are subbranches of the DENN family [33–35]. However, the FNIP1/2-FLCN complex functions as a GTPase-activating protein (GAP) of RRAGC/D, which raises a new question: Does C9orf72 functions only as a GEF? A structural study of both the C9orf72-SMCR8 complex and FNIP2-FLCN complex demonstrated that these two complexes share a similar overall structure [36–40]. Remarkably, the C9orf72-SMCR8 complex has been proved to be a GAP for RAB8A, RAB11A, and ARF1³⁶ [38–40].

In this review, we summarize the relationship between C9orf72 and genetic ALS, present hypotheses of the relationship between abnormal C9orf72 and pathology, discuss the function of C9orf72, and explain the progress in exploring the structural biology of C9orf72 and related protein complexes.

C9orf72 and ALS disease

ALS and ALS-linked genes

Amyotrophic lateral sclerosis (ALS) is a devastating and fatal neurodegenerative disease characterized by progressive degeneration of both upper and lower motor neurons in the spinal cord and brain, leading to hyperreflexia, spasticity, muscular atrophy, paralysis, and ultimately death [41,42]. ALS typically starts focally, either in an upper limb or a lower limb or the bulbar region, and then spreads to other regions with time [41]. The disease usually progresses rapidly, with most patients dying of respiratory failure within 3–5 years of symptom onset [41]. Although there are currently no effective cures, three treatments for ALS have been approved, and many clinical trials are ongoing, including trials of antisense oligonucleotides (ASOs) targeting C9orf72 sense HRE-containing RNAs (ASO BIIB078, ClinicalTrials.gov identifier: NCT03626012) and other ALS-linked genes [43]. The three approved pharmacological treatments for ALS are Riluzole, Dextromethorphan, and Edaravone [44]. Riluzole is an antagonist of glutamate release, and it extends patient survival for approximately 3–19 months [45–47]. Dextromethorphan was approved to ameliorate ALS patients’ emotional lability and enhance speaking and swallowing function in patients [48,49]. Edaravone, a free radical scavenger, was recently shown to slow the decline in functional scores after six months of treatment in a phase III clinical study. Thus, it was approved for the treatment of early-stage ALS cases [50]. Unfortunately, these currently approved molecules only showed moderate improvement in ALS patients. Therefore, significantly improved treatments are greatly needed.

ALS is traditionally classified into familial and sporadic subtypes based on ALS presence in family members, whereas sporadic ALS comprises the most ALS patients, ~ 90% [41]. More than 20 genes have been implicated in familial ALS, sporadic ALS, or both, with the most common genes including SOD1, FUS, TARDBP/TDP-43, C9orf72, PFN1, and UBQLN2 [41,51–54].

In 2006, genetic studies, including the conventional linkage and genome-wide association studies, first linked ALS and frontotemporal dementia (FTD) to the chromosome 9 open reading frame 72 (C9orf72) gene [55,56], showing that a mutation in the noncoding region of C9orf72 is the most common genetic cause of ALS and FTD in a specific population. The C9orf72 gene, consisting of 11 exons, has three main alternatively spliced transcript variants and produces two protein isoforms (Figure 1). It was recently found that the expression of the long C9orf72 isoform, but not the short isoform, can partially rescue the dendritic arborization phenotype in hippocampal neurons from C9orf72-knockout mice, suggesting that the two C9orf72 isoforms might function differently [57]. To date, little is known about the functions of the short C9orf72 isoform; hence, in this review, C9orf72 refers to the long isoform.

Figure 1. Gene products of C9orf72.

The three RNA transcripts of the C9orf72 gene eventually lead to two protein isoforms.

Roles of C9orf72 in ALS

G4C2 hexanucleotide repeat expansion (HRE) in the first intron of the C9orf72 gene may play a critical role in the process of ALS disease. Mechanisms of C9orf72 HRE-induced neurodegenerative changes have been proposed (Figure 2), which mainly consist of loss-of-function and gain-of-toxicity mechanisms, but their relative contribution to pathogenicity is still debated.

Figure 2. The hypothesized assaults caused by G4C2 repeat expansion in C9orf72.

The consequences of C9orf72 mutation include loss of C9orf72 function and gain-of-function with either RNAs that capture RNA binding proteins (RBPs) or dipeptides that disturb cellular homoeostasis.

Most evidence suggests that C9orf72 HRE-induced neurodegeneration depends mainly on a gain-of-toxicity mechanism [58–60], including RNA toxicity and dipeptide repeat protein (DPR) toxicity [29,61]. Because of G4C2 repeat expansion in the C9orf72 gene, higher-order DNA structure G-quadruplexes are formed, which can generate truncated RNA transcripts [62]. These aberrant RNA transcripts form G-quadruplexes and RNA hairpin structures. Additionally, HRE-containing RNA can hybridize with HRE-containing DNA to form R-loops, which are incredibly stable. During cellular processes, including DNA replication, transcription, translation, and telomere function, R-loops are transiently formed [63]. However, the persistent R-loops lead to adverse effects, including genome instability and DNA damage [63]. The G4C2 repeat sense and antisense RNAs transcribed from HREs can form nuclear RNA foci capable of sequestering vital RNA-binding proteins (RBPs) and thus disturb the normal functions of theses proteins [64,65]. On the other hand, repeat-associated non-ATG (RAN) translation generates five different dipeptide repeat proteins (DPRs): glycine-alanine (GA), glycine-arginine (GR), proline-alanine (PA), proline-arginine (PR), and glycine-proline (GP). Furthermore, these DPRs exhibit distinct biochemical properties, with arginine-rich DPRs apparently the most toxic, primarily impeding primarily nucleocytoplasmic transport and RNA processing [5,11,66].

Further studies observed overt motor neuron degeneration when the C9orf72 level was reduced in C.elegans and zebrafish. In addition, C9orf72-KO mice exhibited no neuron degeneration or motor deficit, suggesting that C9orf72 haploinsufficiency is insufficient to cause neurodegeneration in mammals [67–71].

The determination of whether gain-of-function or loss-of-function mechanisms induce the predominate G4C2-expanded C9orf72 effect was realized in a recent study [4]. In human-induced motor neurons (iMNs), the repeated expansion reduced C9orf72 expression, which caused neurodegeneration through the accumulation of glutamate receptors resulting from C9orf72 haploinsufficiency and blocked the degradation of neurotoxic DPR proteins resulting from the G4C2 expansion [4]. The neurodegeneration caused by cooperating gain-of-function and loss-of-function mechanisms was attenuated by the expression of constitutively active GTPase RAB5⁴, indicating that dysfunctional vesicle trafficking is involved in C9orf72 ALS. This finding is in line with that the function of C9orf72 as a GAP to regulate GTPase activity, which is discussed in the subsequent sections. Indeed, recent studies have shown that a reduction in C9orf72 protein levels with a long HRE can enhance the early accumulation of DPRs, suggesting that the loss of C9orf72 has a synergistic effect with the repeat-dependent gain-of-toxicity [3].

Physiological functions of C9orf72 in model organisms

TDP-43 mutations and mouse models

The studies on TARDBP (gene encoding TDP-43) provided informative references for C9orf72 research. Mutations in TARDBP (gene encoding TDP-43) account for approximately 4% of familial ALS cases and a smaller percentage of sporadic ALS cases [72–75]. Tremendous research has been performed to elucidate the molecular genetic underpinnings of TDP-43 related ALS in mouse models [76–79], and the results may serve references for studying C9orf72. TDP-43 is a nucleic acid-binding protein that shuttles between the nucleus and the cytoplasm [80]. In ALS patients’ motor neurons, TDP-43 is mislocalized to the cytoplasm and self-aggregates after posttranslational modification [81]. Additionally, various mutations in TDP-43 have been discovered to be associated with ALS [41]. More than 50 mutations in TDP-43, including G298S, A315T, M337V, G348C, and A382T, have been demonstrated to cause or be linked with ALS [75,82,83].

To study the mechanism of TDP-43 toxicity, various model organisms have been generated, including yeast [84,85], Drosophila [85], Caenorhabditis elegans [86,87], zebrafish [88,89], and many mouse models [79,90,91]. Although it is impossible to clarify the unique features of TDP-43-induced ALS based on a phenotype-genotype correlation found in each mouse model of TDP-43 mutations, some certain clinical characteristics of human TDP-43 mutations can be reproduced in mouse models [79,90,91]. Since genetic knockout of TARDBP in mice causes embryonic lethality [92–94], transgenic mice overexpressing human TDP-43 (WT or mutated) are mostly used [79,90,91]. Correlations between mutations and phenotypes in mouse models can be determined based on protein expression levels and phenotypes [79,90,91]. Low hTDP-43 expression causes age-related motor dysfunction and cytosolic inclusions. Nevertheless, mice with low hTDP-43 expression were not paralysed and died in the normal age range, and those with high hTDP-43 expression showed early-onset and fast disease progression but no clear motor neuron loss or hTDP-43 inclusion bodies [90,95,96]. In mice, hTDP-43-WT overexpression only partially recapitulated ALS features with rapid onset, low survival, motor neuron loss, and muscle atrophy [90,97–99]. However, the double knock-in mice expressing hTDP-43-WT and hTDP-43-Q331K developed rapid progressive limb paralysis, died early, and exhibited p62-, ubiquitin-, and TDP-43-positive inclusion bodies [100]. hTDP-43A315T and hTDP-43G348C transgenic mice developed age-related progressive motor deficits and cognitive impairments with TDP-43 nuclear and cytosolic inclusion bodies in neurons [95,99,101,102]. hTDP-43 Q331K knock-in mice developed behavioural abnormalities but presented with no ALS motor phenotypes [103]. hTDP-43 M337V knock-in mice developed mutant dose-dependent ALS motor phenotypes and neuromuscular junction abnormalities [91,104]. In addition to mice harbouring the aforementioned coding mutations, mouse models expressing a cytoplasmic form of TDP-43 without a nuclear targeting sequence (TDP-43-ΔNLS) recapitulated multiple features of TDP-43 ALS, including progressive neurodegeneration, gliosis, and abnormalities in motor, cognitive and social domains [105–107].

Physiological functions of C9orf72 in model organisms

The in vivo physiological functions of C9orf72 have also been well studied in model organisms. Various animal models have been generated to elucidate the pathological mechanisms of C9orf72-ALS for decades, including yeast [66,108,109], worms [68,110–112], fruit flies [11,12,113–115], zebrafish [67,116–119], mice [7,9,120–124], and induced pluripotent stem neuron cells (iPSNs) derived from the ALS patients [113,125–130].

The nematode Caenorhabditis elegans has also been used as an ALS animal model to study the mechanisms and phenotypes of C9orf72 loss-of-function [110]. Deletion of the C. elegans C9orf72 orthologue, F18A1.6/alfa-1, caused severe paralysis, impaired cytoplasm-nuclear transportation, neurodegenerative motive defects, and early death [68]. The alfa-1 deletion also caused dysregulation in endosomal-lysosomal homoeostasis, which was partially rescued by the expression of human C9orf72 [111]. Interestingly, C. elegans alfa-1 cooperatively functions with smcr8 in regulating the function and re-formation of lysosomes [111]. In addition, deletion of alfa-1 in C. elegans caused enhanced nuclear translocation of transcription factor hlh-30 (an ortholog of human TFEB), leading to activated lipolysis and premature death under starvation conditions [112]. Thus, this study revealed novel functions of C9orf72/alfa-1 in nutrient sensing and metabolic pathways, which may indicate potential mechanisms by which C9orf72-loss disturbs cellular homoeostasis in neurons.

The C9orf72 ortholog in zebrafish is very similar to human C9orf72[67], and three different studies have been performed by either knocking down or deleting C9orf72 in zebrafish [116]. Antisense morpholino oligonucleotide (MOs) knockdown of zebrafish C9orf72 caused motor neuron axonopathy and shortened axons [67]. Disruption of only the functional DENN domain of C9orf72 in zebrafish without manipulation of the full-length gene, caused an altered neuronal network, deficient axon formation, and motor activities [117]. Interestingly, dramatically reduced GTPase activity was observed in this zebrafish model [117]. This finding is in line with the molecular function of C9orf72 as a GAP to regulate GTPase activity, which is discussed in the subsequent sections. Recently, a microRNA-based C9orf72-knockdown model was developed with zebrafish, and it featured phenotypes of motor neuron defects, muscle atrophy, and mislocalized TDP-43[118-119]. The observations of theses zebrafish models suggest that C9orf72 loss-of-function can contribute to ALS pathogenesis, although it is not an exclusive mechanism.

With a short generation time and a relatively less complex brain, Drosophila models offer great advantages for the study of neurodegenerative diseases [131]. Numerous studies using fly models of C9orf72-ALS have been established mostly by the overexpression of repeat sequences or DPRs, which is necessary since Drosophila lack a C9orf72 homolog [113]. However, C9orf72-knockout mice exhibited splenomegaly and lymphadenopathy but no the neurodegenerative phenotypes, as described above, although mouse C9orf72 is homologous to human C9orf72, and more phenotypes related to immunity and inflammation were observed [69,70,132]. Mice with deletion of one or two copies of the C9orf72 genes showed age-dependent lethality [132]. A deficiency of C9orf72 caused low mTOR activity and thereby upregulated levels of transcription factor EB (TFEB), which is repressed by mTOR and can promote the transcription of lysosomal and autophagy genes [132]. The deletion of C9orf72 or its partner SMCR8 caused obvious autoimmunity and inflammatory reactions [69,133,134]. Besides, deletion of the Toll-like receptors (TLRs) TLR3, TLR7, and TLR9 attenuated induced splenomegaly and lymphadenopathy, and activated circulating T cells in SMCR8 deleted mice [69]. Double-knockout (dKO of C9orf72 and SMCR8) mice showed more severe immune abnormalities than single-gene-knockout mice [70]. Further analysis found that the macrophages from the dKO mice exhibit blocked lysosomal degradation and exocytosis [70]. Interestingly, the dKO mice showed increased mTORC1 activity, and mTORC1 inhibition reversed macrophage dysfunction [70].

Synaptic dysfunction caused by C9orf72 deletion or mutations

Loss of normal communication between the presynaptic motor neurons and postsynaptic muscle is one of the hallmark features of ALS [135,136]. Synaptic dysfunction has been implicated in C9orf72-ALS [127,137]. The localization of C9orf72 in neurons provides clues for the link between its mutation and synaptic dysfunction [71]. It has been shown that C9orf72 is enriched at synapses [138,139] and localized to both the pre-and postsynaptic compartments [71]. Experiments with C9orf72-knockout mice suggested that C9orf72 is vital for maintaining postsynaptic receptor levels [71] and is required for synaptic plasticity and adult neurogenesis [120]. The overexpression of poly-GA peptides encoded by C9orf72 with G4C2 expansion in primary mouse cortical neurons led to decreased dendritic arborization, which supports the concept of synaptic dysfunction in C9orf72-ALS [121]. The overexpression of a 48 G4C2 repeat construct in rat spinal cord neurons reduced the number of primary dendritic branches [59]. In mice expressing 30 or more G4C2 repeats in C9orf72, a significant decrease in synaptic bouton number and synaptic quantal content, leading to attenuated evoked potentials, was observed [12,126]. A transgenic mouse line expressing poly-PR of C9orf72 with G4C2 expansion specifically in neurons, develops partial neuropathology of C9orf72-ALS and shows differential expression of genes related to synaptic transmission [17]. Poly-GA in the neurites of mouse models exhibited altered Ca [2]+ influx and synaptic vesicle release, causing a reduction in synaptic vesicle-associated protein 2 (SV2), which was also seen in cortical and motor neurons derived from patient induced cell lines [137].

Immune dysfunction caused by C9orf72 deletion or mutation

Multiple studies have verified that C9orf72 is related to immune system function; for example, immune system dysregulation was observed in homozygous C9orf72-knockout mice, including hyperplasia in the spleen and cervical lymph node, severe autoimmunity, T-cell activation, and increased expression of inflammatory cytokines and chemokines [133,134,140–142]. These C9orf72-null mice exhibited increased immune infiltration in lymphoid organs and elevated cytokines in plasma [133,134]. At the cellular level, bone marrow-derived macrophages (BMDMs) and microglia from these mice exhibited a proinflammatory state, and bone marrow transplantation (BMT) experiments were performed, indicating that the efficacy of C9orf72 in immune function [134,140]. Besides, a similar phenotype was reported in a recent retrospective study. Analysis of the cerebrospinal fluid (CSF) and plasma of 248 patients with motor neuron disease (MND) showed that 164 ALS patients carry mutations in SOD1 (n = 24), C9orf72 (n = 19), and in other ALS related genes (n = 119), which revealed that the differences in survival between patients with different ALS subtypes are correlated with cytokine levels, and more specifically, C9orf72-ALS patients had higher CSF interferon-alpha levels than the patients with one of the other aforementioned ALS subtypes, providing some progressive insights into the pathological mechanism of ALS [134].

A recent study showed that selectively ablating C9orf72 in mouse dendritic cells (DCs) caused marked early-activation of the type I interferon response [142]. In addition, C9orf72 deficiency in myeloid cells disrupted autophagy-mediated degradation of STING, which led to persistent activation of type I interferon, following enhanced autoimmunity, adaptive immune activation, and antitumor immunity [142]. Although these results collectively related to immune dysfunction, the inconsistency in the severity of the persistent autoimmunity outcomes reported by different groups was notable, with some reporting higher premature mortality [60,134,141] and others reporting no effect on survival [140], indicating that environmental contribution might occur in the field. Indeed, researchers supported a hypothesis that the environment, primarily the gut microbiota, plays a significant role in modulating autoimmunity, neuroinflammation, motor deficits, and survival of C9orf72-null mice [140]. Numerous studies have committed to discovering the connection among autophagy, immunity, and neurodegeneration through C9orf72 function and advancing our understanding of the pathological mechanism of C9orf72-induced ALS.

Mechanism of C9orf72 function in autophagy

Regulation of autophagy by C9orf72 and the related complex

The autophagic function of C9orf72 was implied in C9orf72, or SMCR8 deleted mice, which show dysregulation of mTOR or TFEB, which are closely related to autophagy [69,70,132]. The interaction between C9orf72 and SMCR8 was initially proposed due to their structural similarity to FNIP and FLCN, respectively, which are two DENN-containing proteins that are able to form a stable complex [34,35]. In 2016, following affinity purification and mass spectrometric (MS) analysis of anti-C9orf72 immunoprecipitants from different cell lines, three studies separately confirmed that the C9orf72 protein can form a tripartite complex with SMCR8 and WDR41 at both the endogenous and exogenous levels [21,22,25,26,132]. Additionally, the interaction between C9orf72 and SMCR8 was proven to be mediated by the DENN domain in C9orf72 [21]. C9orf72 interacted tightly with SMCR8, and the C9orf72-SMCR8 complex was found to be strong and even resistant to high amounts of salt and detergent [25]. Furthermore, the C9orf72 protein level declined when the expression of SMCR8 was suppressed through small interfering RNA (siRNA) [25,27], while neither cellular C9orf72 nor SMCR8 knockout led to notable downregulation of WDR41 protein levels [132]. Similarly, a recent study demonstrated that loss of SMCR8 and WDR41 significantly diminished C9orf72 protein levels, and SMCR8 levels were also decreased after C9orf72 and WDR41 loss [27,143]. Taken together, these results suggest that both SMCR8 and C9orf72 play a key role in maintaining the stabilization of each other, while WDR41 might stabilize the C9orf72-SMCR8 heterodimer.

A systematic proteomic analysis of the human autophagy system initially revealed SMCR8 as an interacting protein of FIP200 [144], a component of the ULK1 complex. FIP200, ATG13, ULK1, and ATG101 form the ULK1 kinase complex [145,146], the most upstream of the autophagy machinery regulating autophagy in response to nutrient deprivation. C9orf72 exhibited colocalization with both autophagosomes and lysosomes in N2a cells, indicating that C9orf72 might directly affect the autophagy machinery [20]. Accordingly, the evidence suggested a potential connection between C9orf72 and autophagy initiation. The interaction between the C9orf72-SMCR8-WDR41 complex and all the ULK1 complex subunits was then determined [21,23,24,26]. Besides, the interaction between C9orf72 and the ULK1 complex was enhanced under amino acid starvation conditions [21]. Additionally, the interaction between SMCR8 and FIP200 was much weaker than that of the C9orf72-SMCR8-WDR41 complex [21]. In summary, these results suggest that the C9orf72-SMCR8-WDR41 complex likely regulates autophagy through its interaction with the ULK1 complex (Figure 3).

Figure 3. Regulation of the ULK1 complex by C9orf72-SMCR8-WDR41.

C9orf72-SMCR8-WDR41 binds to the ULK1 complex in an unclarified interaction mode, which leads to increased activity of the ULK1 complex and followed autophagy initiation.

Indeed, experiments showed that the decrease in either C9orf72 or SMCR8 caused impaired autophagy induction [21,24,147], which was the same circumstance in both the C9orf72 and SMCR8 knockdown cells [21,24,147]. Additionally, both ULK1 protein levels and phosphorylation at serine 318 of the ULK1 substrate ATG13 were markedly increased upon SMCR8 knockout. In contrast, an opposite effect was observed in the same study resulting from the depletion of C9orf72 or WDR41[21-24]. A recent study found inhibited autophagy and reduced ULK1 protein levels in neurons from C9orf72 depleted mice [147], supporting the notion that C9orf72 and SMCR8 might act against each other to regulate the autophagy initiation complex. Nevertheless, it was later discovered that the increased abundance of ULK1 protein was on due to elevated ULK1 mRNA levels caused by SMCR8 loss. This study also revealed that SMCR8 can regulate the expression of several autophagy genes, including ULK1 and WIPI2 [24]. Since the native functions of these proteins are multifaceted, the exact mechanism by which C9orf72-SMCR8-WDR41 regulates the ULK1 complex remains unclear (Figure 3).

The regulation of the autophagy inhibitor mTORC1 by C9orf72 and the related complexes

Intriguingly, C9orf72 and SMCR8 were reported to regulate mTORC1 activity (Figure 4), indicating that the ULK1 complex might be involved in a mTORC1-dependent regulatory mechanism of C9orf72 and SMCR8 [70,132,143]. Recent studies have shown that the levels of phosphorylated ribosomal protein S6 kinase 1 (S6K1) and mTORC1 are both increased in C9orf72- or SMCR8-deficient macrophages compared to WT controls [70,143]. In addition, several studies observed that C9orf72, SMCR8, and WDR41 colocalize at lysosomes, and this colocalization is enhanced upon starvation, such as amino acid deprivation [25,26,148,149]. It was further confirmed that WDR41 is indispensable for recruiting C9orf72 and SMCR8 to lysosomes during starvation [148]. More specifically, lysosomes were enlarged and compactly clustered in the perinuclear region in C9orf72 or SMCR8 deficient cells [148]. Furthermore, SMCR8-defective cells are generally accompanied by cellular hypertrophy [25,70,140,150]. Moreover, lysosomal defects in macrophages in C9orf72-deficient mice were observed, as indicated by the increased levels of lysosomal proteins such as progranulin (PGRN), prosaposin (PSAP), LAMP1, and cathepsin D (CathD) [26]. Similarly, SMCR8 deficiency caused reduced levels of LAMP1 [70,143]. Accordingly, we can conclude that C9orf72 and SMCR8 regulate mTORC1 activity and lysosomal biogenesis, despite in different ways, and raising the possibility that C9orf72-SMCR8-WDR41 may regulate the ULK1 kinase complex by mTORC1-dependent and mTORC1-independent mechanisms.

Figure 4. Regulation of the mTOCRC1 complex by C9orf72-SMCR8-WDR41.

The WDR41 in the complex confers the lysosome localization, where it interacts with and regulates the kinase activity of the mTORC1 complex.

The structure of C9orf72 and the related complexes

The structure of C9orf72

The DENN domain-containing family is an ancient and highly conserved protein branch, and its members have the N-terminal Longin domain and the C-terminal DENN domain.

Structure-based model building suggested that C9orf72, together with SMCR8, FLCN, and FNIP1/2, compose a subfamily of DENN family proteins (Figure 5a) [33,34,151]. In the C9orf72-SMCR8-WDR41 complex structure, C9orf72 folds into a classic DENN family protein with an N-terminal Longin domain (uDENN domain) and a C-terminal DENN domain that are highly conserved to those in DENN1B-S (Fig. 5b, Fig. 6a) [31,38,39]. The Longin domain is composed of five central antiparallel β-strands (β1- β5) sandwiched between helix α1 and helices α3 and α4. Similarly, the core of the DENN domain consists of a β-sheet (β1ʹ- β5ʹ), which is layered by α6-α9 and appended by a helical bundle (α10-α15) (Figure 5b). The order of β strands in the Longin domain is β3, β4, β5, β1, and β2, whereas in the DENN domain is β2, β3, β1, β4, and β5 (Figure 5b). Of note, the linker between the Longin domain and the DENN domain of C9orf72 is highly flexible. Thus, C9orf72 may adopt an overall structure similar to that of DENN1B-S (Figure 6a).

Figure 5. C9orf72 is a member of the DENN domain family.

a. The domain schematic of C9orf72, SMCR8, FNIP2, and FLCN.

b. The structural feature of the Longin domain and DENN domain. Upper: The topology of the Longin and DENN domain. Lower: The Longin domain (orange) and DENN domain (blue) of C9orf72 (PDB: 6LT0). All the structure models in this paper are generated using Pymol [182].

Figure 6. A putative model of the C9orf72-RAB complex.

a. The structure model of C9orf72 (PDB: 6LT0), Longin domain: orange, DENN domain: blue.

b. DENND1B-RAB35 (PDB: 3TW8), Longin domain: green, DENN domain: red, RAB35: grey.

c. Superimposition of the structure models of C9orf72 and DENND1B-RAB35. The colours are same as (A) and (B). The two structures will adopt a similar overall structure after that the DENN domain of C9orf72 rotates ~90° and moves ~45 Å towards the DENN of DENND1B.

d. The structure model of C9orf72 adopts similar folds to DENND1B.

e. The putative structure model of C9orf72-RAB is generated on the basis of the structure of DENND1B-RAB35.

The structure of the C9orf72-SMCR8 complex

C9orf72 and SMCR8 can form a stable complex that participates in the regulation of autophagy [21–27]. Biophysical analysis and cryo-EM analysis revealed that the C9orf72-SMCR8 complex is a heterodimeric dimer(Figure 7a, C) [38]; moreover, the monomeric structure of the C9orf72-SMCR8 complex was reported as well [39]. The discrepancy between the two studies might reflect three reasons: First, the C9orf72-SMCR8 dimer was overexpressed in insect cells [38], whereas the monomer was purified for HEK293 cells [39]. The posttranslational modifications in the two cell lines might be different, which might lead to two different statuses of the proteins. Second, the complex purified from the two systems might both be dimeric with most parts of the dimers broken during the process of cryo-EM sample preparation in work by Su et al [39]. The latter scenario is supported by observing monomeric particles of C9orf72-SMCR8 in the cryo-EM images in the paper by Tang et al [38]. Finally, and most possibly, C9orf72 and SMCR8 may bind to each other to form a heterodimer, and then the heterodimers dimerize to form a tetramer of C9orf72 and SMCR8 with a 2:2 stoichiometry. Thus, the C9orf72-SMCR8 complex might equilibrate between monomeric and dimeric states according to cellular activity. The two studies captured only the monomeric and dimeric status. Therefore, more work is needed to fully confirm the current data and gain understanding of the physiological significance of the dimerization of the C9orf72-SMCR8 protomer.

Figure 7. The interface between DENN domains of the C9orf72 and SMCR8.

a. Left: the structure of the FLCN-FNIP2 complex (PDB: 6ULG). Longin domain: orange (FNIP2) and pink (FLCN), DENN domain: blue (FNIP2), and cyan (FLCN). The dash line oval indicates the interface between the FNIP2 DENN domain and the FLCN DENN domain. Right: the structure of C9orf72-SMCR8 protomer (PDB: 6LT0), Longin domain: orange (C9orf72), and pink (SMCR8), DENN domain: blue (C9orf72) and cyan (SMCR8). The dash line oval indicates the interface between the C9orf72 DENN domain and the SMCR8 DENN domain, which contributes to the formation of the C9orf72-SMCR8 protomer.

b. In the helix bundle on the C9orf72-DENN and the SMCR8-DENN interface, three different views are showed from left to right. This helix bundle shows a pseudo-centre symmetry.

c. The structure of the C9orf72-SMCR8 dimer complex. The colours are same as panel (A). The dash line ovals indicate the interface between the C9orf72 DENN domain and the SMCR8 DENN domain, while the dash line rectangles indicate the dimer interface between the C9orf72-SMCR8 protomers.

d. The schematic diagram of the structure of the C9orf72-SMCR8 dimer complex in panel (C).

Characterizations of DENN domains in C9orf72-SMCR8 complex

DENN family proteins are reported to be monomers in solution, e.g., DENN1A (PDB: 6EKK), and DENN1B [31]. However, in the cryo-EM structure model of the FLCN-FNIP2 complex, the DENN domains of FLCN and FNIP2 are dimerized with their C-terminal helices forming a helix bundle. This binding feature is confirmed in the structure of the C9orf72-SMCR8 complex (Figure 7a) [38,39], with structural homology to the FLCN-FNIP2 complex. The helix bundle of the C9orf72-SMCR8 interface consists of helices from α11 to α14 in C9orf72 and helices from α10 to α13 in SMCR8. (Figure 7a) [37,152]. Intriguingly, the eight helices show a pseudo-centre symmetry: α11 of C9orf72 to the α10 of SMCR8, the α12 of C9orf72 to the α11 of SMCR8, the α13 of C9orf72 to the α12 of SMCR8, the α14 of C9orf72 to α13 of SMCR8 (Figure 7a,B).

Unexpectedly, the dimerization of the two protomers in the C9orf72-SMCR8 complex is mediated by the most distal C-terminus motif of C9orf72 and the DENN domain of SMCR8 (Figure 7c, D) [38]. Although the electron density of this interface is obscure, the region of C9orf72 that binds to the SMCR8 DENN domain was identified by mutagenesis analysis and analytical ultracentrifugation [38]. This binding manner between DENN domains defines a novel organization mode of DENN domains [38,40]. Whether the interface mediated the dimerization of protomers is widespread among DENN family proteins has yet to be explored, and a bioinformatics analysis will be an effective way to make this determination.

C9orf72 is a regulator of the RAS superfamily

The RAS superfamily of small GTPases consists of more than150 members in humans and can be divided into six major subgroups based on the function and amino acid sequence: RAS, RHO, RAB, RAN, ARF, and RRAG [153,154]. The RAS members, sharing a conserved globular G-domain, function as binary molecular switches that are activated by binding GTP and inactivated when GTP is hydrolysed to GDP [153,155–157]. The GTP-GDP cycle of the RAS superfamily is controlled by GEFs, which facilitate the exchange of GDP to GTP, and by GAPs, which stimulate GTP hydrolysis [153,155–157]. RABs are the largest subfamily among the RAS superfamily and play a critical role in regulating membrane trafficking [155–158]. ARFs are ubiquitous in eukaryotic cells and vital regulators of vesicle biogenesis during intracellular trafficking [153]. RRAGs, the sixth branch of the RAS superfamily, are sensors of amino acids in mTOR signal pathway and function in heterodimers, e.g., RRAGA/B-RRAGC/D [37,152,154,159,160].

DENN-containing proteins have been shown to be GEF or GAP of RABs [20–23,26,34,35,38,40,151,161], ARFs [39,162], and RRAGs [37,152,154,159,160]. For example, DENN1B acts as a GEF for RAB35 by inducing conformational changes in the Switch I and Switch II motifs of RAB35 to facilitate the GDP dissociation [30,31,163]. The MON1-CCZ1 complex functions as a GEF for RAB7 by locking the conformation of the RAB7 Switch I motif, making it is incapable of binding with GDP [33,164].

As a DENN family protein member, C9orf72 is suggested to be a GEF for dozens of RAB GTPases [20–23,151]. Consistent with this hypothesis, C9orf72 was shown to interact and colocalize with a bunch of RAB proteins, including but not limited to RAB1A, RAB5A, RAB7A, RAB8A, RAB11A, and RAB39B [20–23,151]. These RABs are essential for cell migration, membrane receptor recycling, synaptic vesicle secretion, and cell polarization [158,165]. Hence, C9orf72 may plays a pivotal role in the RAB cascade and regulate these cellular processes via RABs. Intriguingly, C9orf72 was indicated as an effector protein of RAB1A because C9orf72 binds to activated RAB1A preferentially [23], and confirmation of this hypothesis would expand the spectrum of known functions of C9orf72 and the DENN family.

Further studies demonstrated that C9orf72 in complex with SMCR8 and WDR41 possesses GEF activity towards RAB8A and RAB39B, as shown in in vitro GDP/GTP exchange assays [21,22]. Moreover, the C9orf72-SMRC8 complex has been proven to be GAP for small GTPases [38–40], however, these studies do not rule out the possibility that C9orf72 alone might function as a GEF or other kind of regulator of small GTPases. As the study into C9orf72 advances, an increasing number of small GTPases will likely be identified as the partners of C9orf72.

The C9orf72-SMCR8 complex is a GAP for members of the RAS superfamily

Similar to C9orf72 alone, the C9orf72-SMCR8 complex functions as a GEF for several RABs in many studies [21–27,132]. Unexpectedly, the C9orf72-SMCR8 complex was also found to be able to accelerate the consumption rate of GTP by RAB8, RAB11, and ARF1, however, this complex cannot increase the nucleotide exchange rate of these small GTPases [38–40], The FNIP2 complex functions as a GAP for RRAGC/D on lysosomes [37,152,159,166], which indicates that the C9orf72-SMCR8 complex is not a GEF for these small GTPases. Similarly, the FLCN was wrongly suggested to be a GEF for RAB35 on the basis of an in vitro assay [167]. Of note, the structure of the C9orf72-SMCR8 protomer resembles that of the FLCN-FNIP2 complex. Strikingly, the structural superimposition suggests that C9orf72 is the structural homolog of FNIP2, while SMCR8 corresponds to FLCN [38–40]. Moreover, both bioinformatic analysis and structural comparisons show that the arginine finger of FLCN is conserved in SMCR8[38-40]. Mutagenesis analyses combined with biophysical analyses demonstrated that the C9orf72-SMCR8 complex acts as a GAP for RAB8A [38], RAB11A [38], and ARF1[39].

Unfortunately, the mechanisms by which the C9orf72-SMCR8 complex binds to small GTPases and the role of the SMCR8 arginine finger in facilitating the hydrolysis of GTP are unknown.

Proposed structural model of the C9orf72-RAB complex and C9orf72-SMCR8-RAB complex

The existing binding region in DENN family proteins

A comparative analysis indicated that three regions in the Longin domain, the A region, B region, and C region, are critical for both intra- and intermolecular binding [168]. The A region, formed by β3, α1 and part of β4/5, primarily mediates the dimerization of Longin domains and binding with the effectors (Figure 8a) [168]. The B region, consisting of α1 and the β1-β2 hairpin, is mainly involved in the recognition of small GTPases (Figure 8a) [168]. The function of the C region is the most variable and usually involves the α2, α3, and sometimes α4; however, this region also participates in binding with the partners (Figure 8a) [168]. The three regions partially overlap, and sometimes the binding of DENN family proteins with their partners requires cooperation between the A and C region; e.g., MON1-CCZ1 must form a complex to function as the GEF for RAB7 [164]; GATOR1 requires NPRL2 and NPRL3, both containing the Longin domain, to regulate the RRAGA/B GTPases as GAP [160]; and the FLCN-FNIP2 complex functions as a GAP of RRAGC/D GTPases via its Longin pair [37,152]. Both C9orf72 and the Longin domain of SMCR8 are necessary for the GAP activity of RAB8A/11A³⁸[39]. In general, the DENN family proteins interact with their partners in multiple manners, bestowing regulatory ability of C9orf72 and related complex to modulate the RAS superfamily through different mechanisms.

Figure 8. The existing binding regions on DENN family proteins.

a. The A region, Bregion, and C region, which are responsible for both intra- and inter-molecular binding, are coloured in red, yellow, and orange, respectively. Two views are showed here.

Structural model of the C9orf72-RAB complex

Radioactive assays showed that C9orf72 alone is able to bind to several RABs [21,23]. Hence, one of the remaining questions is how C9orf72 alone recognizes specific RABs.

Although this question cannot be addressed clearly before the structure of C9orf72-RAB is solved, based on the structure of DENN1B-RAB35, a speculative structural model of C9orf72-RAB was generated (Figure 6c-E). In this model, C9orf72 adopts a conformation that is similar to that of DENN1B-S, and the binding of C9orf72 and RAB is dependent on both the Longin domain and the DENN domain of C9orf72. However, the binding mode between C9orf72 and RABs is unclear, whether C9orf72 alone is a GEF for RABs is unknown, and the determination of both issues requires more functional, biochemical, and structural studies.

Structural model of the C9orf72-SMCR8-RAB complex

The structural model of the C9orf72-SMCR8-RAB complex has not yet been reported; however, we can obtain some clues regarding the model of the C9orf72-SMCR8-RAB complex by studying the existing structural model of GAP-RAB complexes or GEF-RAB complexes, e.g., the structure of the FLCN-FNIP2-RRAGA-RRAGC complex [37,152], the structure of the NPRL2-NPRL3 subcomplex [160], and the structure of the MON1-CCZ1-RAB7 complex [164]. Notably, in the structure of the MON1-CCZ1-RAB7 complex, MON1 and CCZ1 form a Longin pair, which shows a highly similar overall fold to that of the C9orf72-SMCR8 complex (Figure 9a, B). Attractively, the complex of C9orf72 and the Longin domain of SMCR8 shows almost full GAP activity for RAB8A and RAB11A [38]. Although the MON1-CCZ1 complex is a GEF for RAB7 and the C9orf72-SMCR8 complex is a GAP for RAB8/11, it is possible that the Longin pairs of the two complexes recognize RABs in a similar manner. Therefore, a putative structural model of the C9orf72-SMCR8-RAB complex is proposed here based on the structures of the MON1-CCZ1-RAB7 complex and the C9orf72-SMCR8 complex (Figure 9c-E). Since both RABs and ARFs belong to the RAS superfamily and share a highly conserved G domain, the C9orf72-SMCR8 complex probably binds to ARF1 in a manner similar to that shown in the structural model of the C9orf72-SMCR8-RAB complex. Nevertheless, verification of the true mechanism by which the C9orf72-SMCR8 complex recognizes its target RABs or ARFs requires data support.

Figure 9. A putative model of the C9orf72-SMCR8-RAB complex.

a. The structure of the C9orf72-SMCR8 complex (PDB: 6LT0), Longin domain: orange (C9orf72), and pink (SMCR8), DENN domain: blue (C9orf72) and cyan (SMCR8).

b. The structure of the MON1-CCZ1-RAB7 complex (PDB: 5LDD). MON1-Longin: wheat, CCZ1-Longin: pale green, RAB7: grey.

c. And d. Superimposition of C9orf72-SMCR8 complex and MON1-CCZ1-RAB7 complex. The two Longin pairs will adopt a similar overall structure after the SMCR8-Longin moves ~20 Å towards the RAB7.

e. The putative structure model of the C9orf72-SMCR8-RAB complex.

Cellular localization of the C9orf72-SMCR8 complex is mediated by WDR41

WDR41, a WD40 repeat-containing protein, is one subunit of the C9orf72-SMCR8-WDR41 complex with its C-terminal helix embedded into the hydrophobic groove of the DENN domain of SMCR8 without directly contacting with C9orf72[38-40].

WDR41 expression is widespread in cells, including the cytoplasm, ER, Golgi, and lysosome [22,25,26]. Recent studies have shown that WDR41 localizes to the lysosomes upon interacting with SLC66A1/PQLC2 and therefore anchors the C9orf72-SMCR8 complex to the lysosomes, which modulates mTOC1 signalling [148,149]. The results from the Hurley lab indicate that the docking of the C9orf72-SMCR8-WDR41 complex to lysosomes sequestrates the GAP activity for ARF1[39]; moreover, the C9orf72-SMCR8 complex also functions as a GAP for RABs [38,40]. The latest results from Ferguson’s lab’s suggested that WDR41 binds to the inwards facing SLC66A1/PQLC2 through the 7 CD loop [169]. SLC66A1/PQLC2 might function as both a receptor of WDR41 and a transporter of cationic amino acids, which links the lysosome homoeostasis, the cationic amino acid level, and the function of the C9orf72-SMCR8-WDR41 complex [169].

Upon depletion of cationic amino acids, the C9orf72-SMCR8-WDR41 complex is recruited to a lysosome by SLC66A1/PQLC2 [148, 149, 169]. At this time point, the following scenarios might be possible: First, the C9orf72-SMCR8-WDR41 complex cannot inactivate the targeted small GTPases, including ARF1. Inactivation of ARF1 leads to the redistribution of the Golgi apparatus to the endoplasmic reticulum (ER) [170,171]. The ER-Golgi intermediate is critical to the biogenesis of autophagosomes [172]. Together, the recruitment of the C9orf72-SMCR8-WDR41 complex to lysosomes might be important for maintaining the activation state of ARFs, which is critical to the maintenance of ER-Golgi intermediates and the initiation of autophagy (Figure 10A). Second, there might be some unidentified effector proteins of the C9orf72-SMCR8-WDR41 complex on the lysosome. Interestingly, many studies indicate that C9orf72 participates in nutrient-sensing in lysosome that is independent of mTOR signalling [25,132,149,173]. Accordingly, the recruitment of the C9orf72-SMCR8-WDR41 complex to lysosomes might be part of a lysosome-associated pathway that participates in the cellular response to nutrient variation [148,169]. Of course, the recruitment of C9orf72 and the related complex to the lysosome also facilitates the interplay with mTOR [173,174]. Noteworthily, these two hypotheses may reflect functions that parallel each other and might represent only small portions of the C9orf72-related network.

Figure 10. The proposed function models of the C9orf72-SMCR8-WDR41 complex.

a. The C9orf72-SMCR8-WDR41 complex acts as the GAP for ARF1 or other small GTPases in the cytoplasm, whereas the C9orf72-SMCR8-WDR41 complex cannot function as the GAP of ARF1 when it is sequestered by SLC66A1/PQLC2, which can sense the level of cationic amino acids and binds to WDR41 at the inwards open conformation. On the other hand, recruitment of the CSW complex to the lysosome is essential to the cross-talk between the CSW complex and its partners on the lysosome. The structure model of SLC66A1/PQLC2 is generated using human SLC26A9 (PDB:7CH1) as a template [183]. Grey: ARF1, orange, and teal: SLC66A1/PQLC2. Blue: C9orf72, pink: SMCR8, green: WDR41, grey: RABs. The arrows with a solid line indicate reported signal pathways, whereas the arrows with a dashed line indicate proposed connections.

b. Model of the CSW complex localizing on ER or Golgi apparatus: the CSW complex localizes on ER/Golgi via binding to receptors or lipidation. The labels and colours are the same as the panel(A). The brown lines with a circle head denote the lipidation.

C. The C9orf72-SMCR8-WDR41 complex tether two endosomes/vesicles and regulates the fusion through its GAP activity. The C9orf72-SMCR8-WDR41 complex is coloured as a panel(A).

The mechanism of WDR41 localization to the ER and Golgi remains unclear. Many receptors and transporters that are similar to SLC66A1/PQLC2 localize to the ER/Golgi as well and play a pivotal role in the sensing and regulation of nutrients and metabolites, e.g., SLC16A13[175] SLC30A7[175] and the SLC35 subfamily [176,177]. Based on the binding mode of WDR41 and SLC66A1/PQLC2, WDR41 might dock on the ER/Golgi by interacting with a specific receptor or transporter under certain conditions (Figure 10B). It would not be surprising if receptors for WDR41 were found on the ER/Golgi in future.

The localization of the C9orf72-SMCR8-WDR41 complex to the ER/Golgi might also be facilitated by lipidation modifications, such as prenylation [157,178], palmitoylation [179,180], and myristoylation [181]. Intriguingly, the localization of RABs that function on the ER/Golgi is regulated by prenylation [178]. The lipidation of the C9orf72-SMCR8-WDR41 complex might synergize with the prenylation of RABs and facilitate the colocalization of the C9orf72-SMCR8-WDR41 complex and RABs to the ER/Golgi, where the C9orf72-SMCR8-WDR41 complex takes part in the regulation of membrane trafficking as a GAP (Figure 10B).

Moreover, C9orf72 is able to regulate the fusion of endosomes via RABs [70,151,161]. The C9orf72-SMCR8-WDR41 heterotrimer complex can form dimers; therefore, the C9orf72-SMCR8-WDR41 complex can tether two endosomes close to each other using WDR41 as a membrane anchor and regulate the membrane fusion via its GAP activity for RABs or ARF1(Figure 10C). [38,40].

Perspectives

The G4C2 repeat expansion of C9orf72 accounts for almost 50% of genetic ALS cases and ~ 30% cases of genetic FTD cases. This historic finding highlights the future of C9orf72 as a therapeutic target for genetic ALS and FTD. The pathogenic mechanism has been revealed, and the major hypotheses converge. The final answer to this riddle awaits the discovery of more details of C9orf72 related cellular processes.

As a versatile gene, C9orf72 is critical to homoeostasis since its produce participates in the regulation of many signalling pathways, e.g., autophagy, membrane trafficking, lysosome function, and immune activity. Dozens of converging points that C9orf72 regulates the aforementioned pathways have been identified; however, we believe that they represent only a small part of the C9orf72-related network.

Despite many unanswered questions, the physiological function of the C9orf72-SMCR8-WDR41 complex is emerging, which deepens our understanding of the function of the C9orf72-SMCR8-WDR41 complex and sheds light on the pathogenic mechanism of C9orf72 abnormalities in the C9-ALS/FTD. To uncoil the remaining entangled details related to C9orf72, persistent multidisciplinary efforts and collaboration are required.

Acknowledgments

This work was supported by the National Key R&D Program of China grant 2017YFA0506300 (L. K.), 2018YFC1004601 (S.Q.), and NSFC grants 32071214 (Q. S.), 31770820 (L. K.).

Disclosure statement

The authors declare that they have no conflicts of interest with the contents of this article.

Additional information

Funding

This work was supported by the National Key R&D Program of China [2017YFA0506300]; NSFC [31770820]; NSFC [32071214]; National Key R&D Program of China [2018YFC1004601].