Faulty mRNAs with a premature stop codon (PTC) are recognized and degraded by nonsense-mediated mRNA decay (NMD). Recognition of a nonsense mRNA depends on translation and on the presence of NMD-enhancing or the absence of NMD-inhibiting factors in the 3′-untranslated region. Our review summarizes our current understanding of the molecular function of the conserved NMD factors UPF3B and UPF1, and of the anti-NMD factor Poly(A)-binding protein, and their interactions with ribosomes translating PTC-containing mRNAs. Our recent discovery that UPF3B interferes with human translation termination and enhances ribosome dissociation in vitro, whereas UPF1 is inactive in these assays, suggests a re-interpretation of previous experiments and modification of prevalent NMD models. Moreover, we discuss recent work suggesting new functions of the key NMD factor UPF1 in ribosome recycling, inhibition of translation re-initiation and nascent chain ubiquitylation. These new findings suggest that the interplay of UPF proteins with the translation machinery is more intricate than previously appreciated, and that this interplay quality-controls the efficiency of termination, ribosome recycling and translation re-initiation.
Introduction
Nonsense-mediated mRNA decay (NMD) was first discovered in Saccharomyces cerevisiae and in men [1,2]. The pathway was initially described as an mRNA surveillance mechanism that recognizes and degrades transcripts containing a premature termination codon (PTC). Subsequent research showed that NMD is an important post-transcriptional regulator of eukaryotic gene expression and essential for cellular homeostasis, cell cycle progression, cellular stress response, development and differentiation, neural activity and immunity [3,4].
Recognition of a PTC-containing mRNA as an NMD substrate requires translation [5–7]. Translation termination occurring at a PTC, i.e. in a suboptimal environment, slows down the termination reaction, and possibly subsequent ribosome recycling and re-initiation, thus triggering NMD [8–10]. Typical PTC-containing NMD substrates are characterized by the presence of specific landmarks such as an exon junction complex (EJC) downstream from a stop codon or a long 3′ UTR (Figure 1). The EJC is deposited during splicing 20–24 nucleotides (nt) upstream of an exon–exon junction and removed during translation [11,12]. ‘Normal’ stop codons trigger efficient termination, ribosome recycling and translation re-initiation. These stop codons are typically positioned in the last exon (Figure 1A), and thus, the corresponding mRNAs are EJC-free. Recently, a higher rate of out-of-frame translation or low codon optimality in cellular transcripts was also linked to NMD in yeast [13]. How exactly a PTC is recognized by the NMD machinery is still enigmatic despite many years of research.
Hallmarks of normal and premature stop codons and their 3′-UTR context.
(A) Normal stop codons are positioned in the last exon in proximity of the poly(A) tail. (B) Premature stop codons are characterized by an EJC positioned at least 50–55 nt downstream from the PTC (the effect of the EJC is distance-dependent), accumulation of UPF1 next to the PTC and/or in 3′-UTRs which are longer than usual (see the main text for details). (C) Transcripts with a PTC close to the start codon can evade NMD by translation re-initiation, despite the presence of downstream EJCs.
(A) Normal stop codons are positioned in the last exon in proximity of the poly(A) tail. (B) Premature stop codons are characterized by an EJC positioned at least 50–55 nt downstream from the PTC (the effect of the EJC is distance-dependent), accumulation of UPF1 next to the PTC and/or in 3′-UTRs which are longer than usual (see the main text for details). (C) Transcripts with a PTC close to the start codon can evade NMD by translation re-initiation, despite the presence of downstream EJCs.
Human NMD is mediated by the eukaryotic Release Factors eRF1 and eRF3a, the conserved UP-Frameshift proteins UPF1, UPF2 and UPF3B, the kinase SMG1 and SMG5–SMG9 (Suppressor with Morphological effect on Genitalia). Prevailing NMD models suggest that the ATP-dependent RNA helicase UPF1 is the key factor to recognize the terminating ribosome at the PTC and to nucleate the NMD machinery [14–17]. UPF1 is suggested to interfere with translation termination [16] (see below) and to recruit the SMG1–8–9 kinase complex (SMG1c) [18]. UPF2 and UPF3B, which both are associated with a downstream EJC or recruited by an unknown mechanism, are required to activate UPF1 phosphorylation by SMG1c [18] as well as UPF1's ATPase and helicase activities [19,20], helped by the RNA helicase DHX34 [21]. Phospho-UPF1 then serves as a platform to recruit the endonuclease SMG6 and the SMG5 : 7–PP2A complex for deadenylation of the mRNA [22–24], thus triggering mRNA decay.
In humans, the study of NMD mechanisms is complicated by the existence of different branches of NMD, occurring independent of the presence of UPF2, UPF3B or the EJC [25–27]. Experimentally, the main obstacle, however, for the molecular dissection of the NMD pathway and its branches is the lack of an in vitro NMD assay. This is possibly due to the fact that not all factors required for NMD are known; and thus, not all required components for NMD may be present in in vitro reactions. Accordingly, current NMD models are mostly based on genetic data in cells, as well as in vivo and in vitro protein–protein interaction assays. Consequently, the molecular events during translation termination at a PTC, which leads to assembly and activation of the NMD machinery, are still poorly understood.
Several lines of evidence suggest that translation termination at a PTC is mechanistically different and less efficient than normal translation termination: in cells, normal translation termination is too efficient to be followed experimentally, e.g. in primer extension experiments (toe-print assays). Termination at a PTC, however, is slower leading to a toe-print signal of the terminating ribosome [8,28]. Secondly, stop codon read-through assays indicate that PTCs are more susceptible to stop codon suppression, i.e. termination is less efficient [29,30]. Third, UPF1 knockdown reduces stop codon read-through, i.e. UPF1 interferes with translation termination in humans [16]. In contrast, yeast translation experiments indicate that Upf1p is important for termination at a PTC [8,31]. Finally, co-immunoprecipitations indicated that a ‘SURF’ complex assembles at PTC-stalled ribosomes, comprising SMG1c, UPF1, eRF1 and eRF3a [15,18].
Here, we discuss recent work revealing new functional interactions of NMD factors with the human translation machinery and characteristics of human NMD substrates.
UPF1 accumulates in the 3′-UTR of PTC-containing mRNAs
In human cells, UPF1 is bound to RNAs in a length-dependent, rather than a sequence- or translation-dependent manner [32,33]. UPF1 has been found to be enriched in the 3′-UTR of mRNAs as the result of displacement from the coding region by the translating ribosome [34] (Figure 1). NMD substrate discrimination was suggested to be achieved by ATPase-dependent dissociation of UPF1 from non-target mRNAs, leading to an enrichment of UPF1 on PTC-containing mRNAs next to the stop codon and near the 3′-end of the mRNA (Figure 1B) [35]. However, the accumulation of UPF1 in the 3′-UTR of NMD substrates is not sufficient to commit the mRNA to degradation and a second commitment step is required.
Exon junction complexes enhance NMD
Multiple observations suggest that the EJC — which consists of ATP-bound RNA helicase EIF4A3, Barentsz (MLN51/CASC3) and the heterodimer MAGOH-RBM8A (RNA-binding protein 8A, also known as Y14) as the core components [11,36], is the most important NMD-activating factor: mammalian NMD is enhanced if the PTC is located at least 50–55 nt upstream of an exon–exon junction (Figure 1B) [6,37]. NMD can be artificially triggered by tethering of RBM8A downstream from a stop codon [25,38,39]. Vice versa, PTC-containing mRNAs can be stabilized in vivo by the elimination of the EJC core component EIF4A3 [40,41]. Co-immunoprecipitations indicated that UPF2 and UPF3B, which are bound to the EJC [11], are part of a larger, decay-inducing complex which also comprises SMG1c, UPF1 and the terminating ribosome [18,21].
Notably, recent bioinformatics studies analyzing genome and transcriptome data corroborated that a downstream EJC is the most important predictor of human NMD [42,43]. Comparison of the matched exome (i.e. the expressed genome) and transcriptome (mRNA) from ∼10 000 human tumours allowed NMD efficiency of nonsense mutations to be determined through measuring the change in expression levels of the mutant mRNA compared with the average ‘wild-type’ mRNA from the same cancer subtype [43]. NMD efficiency was shown do decrease with the distance between the PTC and the EJC in exceptionally long exons. Moreover, transcripts with a PTC close to the start codon could evade NMD by translation re-initiation (Figure 1C), despite the presence of downstream EJCs [43], as previously reported [28,44,45].
Taken together, all findings agree with a direct cross-talk between the PTC-bound terminating ribosome and EJC-bound factors to induce NMD.
UPF1 and UPF2 are inactive in translation termination in vitro
Scheme illustrating the interactome of translation and NMD factors.
A scheme for each factor UPF1, UPF2, UPF3B, eRF1, eRF3a and PABPC1 is represented in the central circle. Graduations indicate 50 amino acids. Domains specific to each factor are mapped to the scheme. Coloured ribbons are indicating known interactions: green for structural data, grey for in vitro pulldowns using purified proteins and red for in vivo co-immunoprecipitations from cell extracts (adapted from ref. [83]). N: N-terminal domain; M: middle domain; C: C-terminal domain; 2/3: 2 and 3 domains; G: GTPase domain; RRM: RNA-recognition motif; PABC: Poly(A)-binding protein C-terminus. SQ: serine–glutamine-rich domain; CH: cysteine–histidine-rich domain; helicase domain: RecA1 and RecA2; MIF4G: middle domain of eIF4G; U1B: UPF1-binding domain; EBM: exon junction complex-binding motif.
A scheme for each factor UPF1, UPF2, UPF3B, eRF1, eRF3a and PABPC1 is represented in the central circle. Graduations indicate 50 amino acids. Domains specific to each factor are mapped to the scheme. Coloured ribbons are indicating known interactions: green for structural data, grey for in vitro pulldowns using purified proteins and red for in vivo co-immunoprecipitations from cell extracts (adapted from ref. [83]). N: N-terminal domain; M: middle domain; C: C-terminal domain; 2/3: 2 and 3 domains; G: GTPase domain; RRM: RNA-recognition motif; PABC: Poly(A)-binding protein C-terminus. SQ: serine–glutamine-rich domain; CH: cysteine–histidine-rich domain; helicase domain: RecA1 and RecA2; MIF4G: middle domain of eIF4G; U1B: UPF1-binding domain; EBM: exon junction complex-binding motif.
More recently, a fully reconstituted human translation system was used to probe the impact of the three human UPF proteins on translation termination [47]. This translation system faithfully monitors each step of mammalian translation, including translation termination, ribosome recycling and re-initiation on an upstream or downstream start codon [48–50]. Using this in vitro system, translating ribosomes stalled at a stop codon can be purified. The subsequent addition of release factors followed by toe-printing assays allows observation of stop codon recognition by eRF1–eRF3a [48]. Using limiting amounts of release factors, the system was adapted to test the impact of each NMD factor individually and in combination [47]. Surprisingly, UPF1 and a large fragment of UPF2 had no impact on the efficiency of translation termination. UPF1 was inactive, irrespective of its phosphorylation by the kinase SMG1 or of UPF1's ATPase activity [47]. This suggests that UPF1 does not play a role in translation termination, a finding which contradicts the prevailing NMD models. In agreement with these findings, recent work from Rachel Green's laboratory suggests that yeast Upf1 has no role in translation elongation, termination or ribosome recycling in vitro [51].
UPF3B interferes with termination in vitro
In contrast, the NMD factor UPF3B, which is a peripheral EJC subunit, interfered with stop codon recognition and peptide release from the ribosome [47]. In agreement with a role of UPF3B in translation termination, purified UPF3B was shown to interact directly with eRF3a and with the ribosome (Figure 2). Moreover, a direct interaction of UPF3 and UPF1 was discovered (Figure 2) [47]. Previous co-immunoprecipitations were repeated and it was shown that UPF3B is part of the complex between UPF1 and eRF3a, likely mediating this interaction. For UPF3B's inhibitory effect on translation termination, the RNA-recognition motif (RRM) and the middle domain are required, but not the EJC-binding domain, consistent with an association of UPF3B with the EJC in the 3′-UTR of NMD substrates (Figure 1B). The fact that UPF3B binds RNA may be relevant for the EJC-independent branch of NMD [27,47,52]. UPF3B bound to the mRNA 3′-UTR could interact with the PTC-bound ribosome and eRF3a and slow-down translation termination (Figure 3). A transient interaction of UPF3B and UPF1 on the ribosome could contribute to the assembly of the NMD machinery at the PTC, for instance by recruiting UPF2 and SMG1–8–9.
Modified model for initiation of NMD, adapted from ref. [83].
When a ribosome encounters a PTC, stop codon recognition can be delayed by the presence of UPF3B bound to the EJC, or by UPF3B bound to the mRNA in EJC-independent NMD [27]. Moreover, the long distance to PABP prevents efficient translation termination and re-initiation. At the PTC, the ribosome, release factors, UPF3B and likely also UPF1 can form a complex. Hydroxyl-radical probing suggests that yeast Upf1 binds to the ribosomal L1 stalk, near to the E-site [51]. UPF1 can also bind UPF3B [47]. UPF1 may ubiquitylate the PTC-encoded nascent chain and target the nascent polypeptide to degradation. After termination, UPF3B and UPF1 both could be involved in ribosome dissociation. UPF1 can recruit other NMD factors, including the SMG1–8–9 complex. UPF2 binding displaces UPF3B from the ribosome-release factor complex [47]. UPF2–UPF3B activates the kinase SMG1 leading to UPF1 phosphorylation. Phospho-UPF1 recruits decay factors such as the SMG6 endonuclease and the SMG5–7 heterodimer. SMG6 displaces UPF3B from the EJC and cuts the mRNA close to the PTC.
Modified model for initiation of NMD, adapted from ref. [83].
When a ribosome encounters a PTC, stop codon recognition can be delayed by the presence of UPF3B bound to the EJC, or by UPF3B bound to the mRNA in EJC-independent NMD [27]. Moreover, the long distance to PABP prevents efficient translation termination and re-initiation. At the PTC, the ribosome, release factors, UPF3B and likely also UPF1 can form a complex. Hydroxyl-radical probing suggests that yeast Upf1 binds to the ribosomal L1 stalk, near to the E-site [51]. UPF1 can also bind UPF3B [47]. UPF1 may ubiquitylate the PTC-encoded nascent chain and target the nascent polypeptide to degradation. After termination, UPF3B and UPF1 both could be involved in ribosome dissociation. UPF1 can recruit other NMD factors, including the SMG1–8–9 complex. UPF2 binding displaces UPF3B from the ribosome-release factor complex [47]. UPF2–UPF3B activates the kinase SMG1 leading to UPF1 phosphorylation. Phospho-UPF1 recruits decay factors such as the SMG6 endonuclease and the SMG5–7 heterodimer. SMG6 displaces UPF3B from the EJC and cuts the mRNA close to the PTC.
UPF2 addition to the termination reaction interferes with UPF3B's inhibition of translation termination [47]. Similarly, UPF3B was found to disrupt the interaction between UPF2 and eRF3a to form a more stable UPF2–UPF3B complex [46]. UPF2 and UPF3B are required for activation of SMG1 kinase as well as stimulation of UPF1's helicase and ATPase activities. Taken together, UPF2 binding could co-ordinate the switch from PTC-related activities of UPF3B (and of UPF1, see below) to UPF1 phosphorylation and activation of mRNA decay.
Ribosome dissociation by UPF3B
When release factors are added in excess to the translation termination reaction, UPF3B was observed to destabilize post-termination complexes (post-TCs), as evidenced by a diminution of the post-TC bands and an increased signal for the full-size mRNA/cDNA bands in toe-printing experiments [47]. UPF3B-induced ribosome dissociation was incomplete and only observed after GTP hydrolysis (and not in the presence of non-hydrolysable GTP analogues) and after nascent chain release [47]. This UPF3B activity is reminiscent to the activity of eIF1-1A-eIF3-eIF3j, which together promotes splitting of post-TCs into 60S subunits, tRNA, mRNA and 40S subunits [53].
Again, UPF2 inhibited UPF3B's capacity to dissolve post-TCs [47]. This could indicate either that UPF2 joins the factors on the PTC-stalled ribosome at a later stage, after translation termination and ribosome dissociation (Figure 3), or that UPF3B's effect on translation termination is specific to the UPF2-independent NMD branch [25]. In the former case, UPF1, UPF2 and UPF3B will be engaged in differently composed complexes at different stages of termination and NMD: UPF3B binds ribosome-release factors, interacts with UPF1 on the ribosome and dissociates post-TCs in the absence of UFP2 (Figure 3). Subsequent UPF2 joining would displace release factors from the complexes, in order to form UPF1–UPF2–UPF3B-containing complexes engaging in messenger ribonucleoprotein remodelling [19] and in the recruitment of decay enzymes triggering the decay phase of NMD [54].
Intriguingly, UPF3B has been reported to have a general effect on human translation and stimulate protein synthesis when tethered to the coding region of mRNA [55]. This yet uncharacterized effect depends on the presence of UPF3B's RRM-like domain and is independent of UPF2 and RBM8A (EJC) binding.
Efficient translation termination and re-initiation prevents NMD
Unusually, long 3′ UTRs have been shown to trigger NMD in yeast, Drosophila and Caenorhabditis elegans [8,56–58] (Figure 1B). NMD can be prevented by tethering Poly(A)-binding protein (PABP) downstream from a PTC or by introducing a secondary structure into the 3′ UTR mRNA to bring PABP closer to the stop codon [17,56,59]. A direct stimulation of translation termination by PABP was demonstrated using a reconstituted mammalian translation system [60]. PABP was shown to interact with the N-terminal part of eRF3a [61] and with the ribosome (Figure 2), and it was suggested to promote the recruitment of eRFs to the ribosome thus facilitating stop codon recognition [60]. Thus, human UPF3B and PABP could compete at the terminating ribosome for the binding of eRF3a (Figure 3), leading to opposing effects on translation termination efficiency [47,60].
Cytoplasmic PABP 1 (PABPC1) also interacts with the initiation factor eIF4G (Figure 2), a subunit of the eIF4F complex which binds the 5′-cap of mRNA. The PABPC1–eIF4G interaction circularizes the mRNA and juxtaposes the stop codon and the start codon, facilitating translation re-initiation on the same mRNA after translation termination. Efficient translation re-initiation was suggested to inhibit NMD [28,44,62]. Consistently, NMD can be prevented by tethering eIF4G to the 3′-UTR downstream from a PTC, suggesting that initiation factors are also anti-NMD factors [9,63]. Similarly, eIF3 is required to prevent NMD of mRNAs with a PTC close to a start codon (Figure 1C), and knockdown of eIF3 subunits renders such mRNAs NMD-sensitive [9,28]. This is explained by the finding that eIF3 can remain bound to elongating ribosomes after initiation [64]. Thus, for relatively short ORFs, PABPC1 could still be close to the ribosome when it reaches the stop codon, thereby preventing NMD [62]. This highlights that at a normal stop codon, a tight link between translation termination, ribosome recycling and translation re-initiation exists, ensuring optimal translation of mRNA.
Interestingly, eIF3 subunits have been shown to interact with UPF1 (Figure 2), which then prevents the formation of elongation-competent 80S initiation complexes and thus represses translation and favours NMD [65–67]. Similarly, in yeast, in vitro re-initiation following termination was shown to be less efficient at a PTC compared with a normal stop codon [68]. Moreover, Upf2p has been found to interact with eIF4A (Figure 2), which is part of the cap-binding complex eIF4F [69], establishing a second potential link between the translation machinery and NMD factors to suppress translation of nonsense mRNAs. Finally, dissociation of post-TCs by UPF3B in the absence of initiation factors [47] could prevent translation re-initiation and favour the formation of decay-inducing complexes.
NMD avoidance mechanisms
Long 3′-UTRs could stimulate NMD by length-dependent accumulation of UPF1 (see above) [32]. While model systems established a link between 3′-UTR length and NMD susceptibility [8,17,70,71], a genome-wide correlation of 3′-UTR length and NMD is complicated by the fact that many mRNAs with long 3′-UTRs are protected from NMD mostly by unknown mechanisms [43]. One such evasion mechanism includes an RNA-stability element (RSE) located within 200 nt downstream from an NMD-triggering PTC. This RNA motif recruits the polypyrimidine-tract-binding protein 1 (PTBP1) [72] and together they form a 3′-mRNP structure that limits UPF1 association with the 3′-UTR, possibly by preventing initial UPF1 binding or by blocking 5′–3′ translocation of UPF1 along the mRNA [72].
Ribosome recycling and nascent chain degradation at a prematurely terminating ribosome
Little is known about how the PTC-bound ribosome is recycled after translation termination. UPF3B has been shown to dissociate post-termination ribosome complexes [47]. Yeast experiments indicate that PTC-bound ribosomes require Upf1p to be removed from the mRNA [73]. ATPase-deficient Upf1p was shown to lead to an accumulation of mRNA fragments that comprise the PTC-bound ribosome and the downstream 3′-UTR, implicating Upf1p in ribosome release [73]. These fragments were generated by XRN1, a 5′-to-3′ exonuclease, which is stalled by the presence of ribosomes. Thus, the mRNA with the PTC-bound ribosome and the 3′-UTR fragment remained intact. In fact, Upf1p's ATP-binding and RNA-binding activities, as well as Upf2p and Upf3p, were required for efficient removal of the PTC-bound ribosome from the mRNA. This implicates that Upf1p plays a role in termination or ribosome recycling at the PTC after complex formation with Upf2p and Upf3p. In agreement, the deletion of any of the three Upf-encoding genes in yeast causes ribosome-release defects [68].
The mRNA degradation phase in NMD is initiated by UPF1 phosphorylation in higher eukaryotes. Phospho-UPF1 recruits SMG6, which cleaves the mRNA close to the PTC, SMG5-7, which recruits CCR4-Not deadenylase and the decapping enzyme Dcp2 (reviewed in ref. [74]) (Figure 2). However, how the C-terminally truncated proteins, which are encoded by nonsense mRNA and which are potentially harmful for the cell, are recognized and efficiently removed is enigmatic. The Not4 subunit of the CCR–Not complex has E3 ubiquitin ligase activity and has been implicated in co-translational quality control, targeting proteins encoded by defective mRNA to the proteasome [75]. Notably, UPF1's cysteine- and histidine-rich (CH) domain (Figure 2) structurally resembles RING domains frequently found in E3 ubiquitin ligases [76].
Yeast Upf1p has been reported to self-ubiquitylate in an Upf3p-dependent manner in vitro [77]. Moreover, Upf1p-stimulated and ubiquitin-dependent degradation of proteins encoded by PTC-containing mRNA (PTC product) has been shown in yeast cells [78–80]. Similarly, depletion of E3 ubiquitin ligase Ubr1 also stabilized PTC products in yeast, suggesting that Upf1p and/or Ubr1's E3 ligase activities contribute to the degradation of PTC-induced truncated peptides [78]. In agreement, proteomics identified interactions of UPF1 with ubiquitin–proteasome components [67,81]. Recently and independent of its NMD activity, human UPF1 was shown to act as an E3 ubiquitin ligase promoting the degradation of the transcription factor MYOD which regulates myogenesis [82]. It is therefore tempting to speculate that UPF3B and UPF1 bound to the PTC-stalled ribosome could promote ubiquitylation of the nascent chain and its subsequent degradation by the proteasome (Figure 3).
Conclusions
Taken together, the interplay of UPF proteins and the translation machinery appear to be much more intricate than previously appreciated. The new findings suggest that the NMD factors quality control the efficiency of translation termination, ribosome recycling and translation re-initiation: both, UPF1 and UPF3B are found in the 3′-UTR of ‘normal’ mRNAs [32,34,35,52], and in conjunction with an EJC, they strongly enhance NMD (Figure 3). If the ribosome fails to terminate and re-initiate translation efficiently on an mRNA, the assembly of NMD complexes and their activation is favoured. Consistently, UPF3B interferes with translation termination in vitro [47], and UPF1 is suggested to inhibit translation re-initiation by binding eukaryotic initiation factor 3 [65,66]. While the PTC-stalled ribosome is likely the assembly platform for the NMD machinery [18] or the trigger for NMD machinery activation, the order of factor recruitment, the molecular function of the UPF factors and the regulation of NMD still require further clarification.
Abbreviations
- CH
Cysteine–Histidine-rich domain
- EJC
Exon Junction Complex
- eRF
eukaryotic Release Factor
- NMD
Nonsense-Mediated mRNA Decay
- nt
nucleotides
- PABP
Poly(A)-Binding Protein
- PABPC1
Cytoplasmic PABP 1
- Post-TC
Post-Termination Complex
- PTC
Premature Termination Codon
- RRM
RNA-Recognition Motif
- SMG
Suppressor with Morphological effect on Genitalia
- SMG1c
complex comprising the kinase SMG1, SMG8 and SMG9
- SURF
SMG1c, UPF1, eRF1
- UPF
UP-Frameshift protein
- 3′-UTR
3′-Un-Translated Region
Funding
We gratefully acknowledge support from the European Research Council [ComplexNMD, 281331] and funding from EMBL and RFBR (Russian Foundation for Basic Research) for a collaborative project.
Acknowledgments
We are grateful to Dr Gabriele Neu-Yilik for critically reading the manuscript. We thank Drs Miao Tian and Miao Wei for help with Figure 2. We also wish to apologize to colleagues whose work we could not cite due to space constraints.
Competing Interests
The Authors declare that there are no competing interests associated with the manuscript.