NMD is an mRNA surveillance process through which mRNAs carrying a premature termination codon (PTC) are rapidly detected and degraded. A PTC is a stop codon in phase with the translational start codon and located anywhere upstream of the physiological stop codon, itself in phase with the translational start codon and PTC. In mammalian cells, the position of a PTC in the open reading frame (ORF) is crucial for NMD activation. Two patterns of NMD activation presumably coexist in the cell. The first relates the position of the PTC to the presence of at least one downstream splicing event at a distance of at least 50–55 nucleotides ^[1][13]. This model involves deposition of the EJC 20–24 nucleotides upstream of exon-exon junctions as a consequence of intron splicing ^[2][14]. The EJC then recruits the NMD core factor UPF3X (also called UPF3B) and then UPF2 before UPF1 is recruited. While little is known about recruitment of UPF3X and UPF2 to the EJC, that of UPF1 has been much more studied. During the first/pioneer round of translation, the interaction of UPF1 with the CBP80 protein placed on the 5’ cap facilitates its recruitment, together with the SMG1/SMG8/SMG9 proteins and the eRF1 and eRF3 release factors ^[3][15]. This interaction then facilitates recruitment of UPF1 by the other UPF proteins (UPF2 and UPF3X) (located downstream of the PTC) to the EJC, to form the decay-inducing (DECID) complex. Interestingly, it is UPF3X which participates in recruitment of eRF3, positioning it at the A site of the ribosome ^[4][16]. UPF1 is then phosphorylated by SMG1, thus losing its affinity for SMG1/SMG8/SMG9 and the eRFs and causing their departure and recycling of the ribosome ^[5][17]. Phosphorylated UPF1, stimulated notably by its interaction with UPF2, exerts helicase activity so as to remove the proteins located downstream of the PTC, thus preparing this end for degradation by ribonucleases ^[6][18]. Phosphorylation of UPF1 increases its affinity for a protein complex composed of SMG5, SMG7, SMG6, and protein phosphatase 2A. This complex then dephosphorylates the UPF1 protein. This UPF1 phosphorylation-dephosphorylation cycle leads to activation of NMD of the mRNA, with recruitment of enzymes involved in degrading the 5 ‘and 3’ ends, such as the DCP2 decapping protein and the exoribonuclease XRN1, the deadenylation proteins PARN and CCR4, and the exosome ^[7][8][8,19]. In addition the SMG6 protein, thanks to its endoribonuclease activity, induces cleavage in the vicinity of the PTC, generating free, accessible 5 ’and 3′ ends for exoribonucleases ^[9][7].

The second model of NMD activation is related to the distance between the PTC and the PABPC1 protein located on the polyA tail ^[10][11][10,20]. The greater this distance, the greater the probability of NMD induction. In this model, the UPF proteins accumulating on the 3′UTR, particularly UPF1 ^[12][13][14][21,22,23], compete with PABPC1 located on the polyA tail for recruitment of the translation termination complex. If PABPC1 recruits this complex, termination of translation does not lead to NMD activation but leads, instead, to initiation of new translation cycles. If UPF proteins recruit the complex, NMD of this mRNA is activated and initiation of new translation cycles is blocked.

Upon identification of the first NMD factors in humans, expression levels of the corresponding genes were shown to differ from tissue to tissue ^[15][24]. The mRNA encoding UPF3X, for example, is abundant in the testes and fetal brain and practically absent in the fetal liver and mammary glands. These differences suggest that the NMD efficiency may vary from tissue to tissue. This has been clearly shown in the context of Schmid meta-physeal chondrodysplasia linked to a nonsense mutation in the COL10A1 gene encoding collagen X. In patients with this disease, the mRNA encoding collagen X is efficiently degraded through NMD in cartilage cells but poorly or not at all in lymphoblasts and bone cells ^[16][25]. In this specific case, however, the reason for the difference in NMD efficiency from one cell type to another remains to be clarified.

Even closely related cell lines can differ in this regard. In a 2007 study, for example, up to three-fold variations in NMD efficiency were found between different HeLa cell lines from different laboratories. This variability was not due to differential levels of core NMD factors (such as UPF1, UPF2, and UPF3X), but to the EJC component RNPS1 ^[17][26].

Interestingly, correlating NMD efficiency with levels of transcripts encoding the major NMD factors appears difficult. In a study on mice, for example, nonsense-mediated decay of a PTC-carrying MEN I mRNA appeared twice as efficient in the ovaries and testes as in the intestines and thymus ^[18][27]. Yet, whereas levels of transcripts encoding major NMD factors and EJC components were definitely elevated in the testes, they were quite comparable in the ovaries, intestines, and thymus. This lack of correlation might mean that one or more genes whose expression correlates with NMD efficiency were excluded from this study or that NMD factor expression is regulated at different levels, as described below.

As during cell differentiation, a new register of genes is expressed, one might expect quality control mechanisms to remain vigilant throughout this process. NMD efficiency has been studied during several cell differentiation processes, including development of the nervous system. First evidence indicates that mutations in NMD factors (including UPF3X and UPF2) or EJC proteins lead to mental retardation in humans ^{[19][20][21][22][23][24][25][26][27][28]}[28,29,30,31,32,33,34,35,36,37]. Regulation of NMD during development of the nervous system seems essential, as NMD is known to regulate axonal development ^[29][38]. During this process, overexpression of the micro-RNAs miR-9, miR-124, miR-125, and miR-128 leads to repressed expression of UPF1, CASC3 (also named MLN51), and SMG1 ^{[30][31][32][33]}[39,40,41,42]. NMD is thus most likely inhibited, presumably to allow synthesis of natural NMD substrates.

Modulation of NMD has also been observed during differentiation of C2C12 myoblasts to myotubes. During this process the UPF2 protein level drops drastically, and this results in strong inhibition of UPF2-dependent NMD. This allows the synthesis of certain natural NMD substrates, including the mRNA encoding myogenin, a protein required for this differentiation ^[34][43].

Alternative splicing is a highly regulated process generating different mRNA isoforms from a single transcript. These different mRNA species may code for proteins with different properties. The literature abounds in reviews describing this regulation ^[35][36][37][44,45,46]. The transcripts of several NMD factors are subject to alternative splicing, including those encoding the UPF3 (also named UPF3A) and UPF3X proteins ^[15][24]. Shorter proteins may result from the absence of one or more exons (notably exons 4 and 8) in the corresponding mRNA. The specific functions of these isoforms have not yet been thoroughly investigated, although the two isoforms of have been shown to differ in their ability to interact with specific partners. For example, the short isoform of UPF3 interacts with SMG7 and little or not with UPF2, whereas the long isoform interacts strongly with UPF2 and little or not with SMG7 ^[38][47]. The fact that this exon-skipping leads to synthesis of a truncated protein rather than to a PTC-generating frameshift mutation very strongly suggests that the truncated protein must be synthesized under specific conditions and must perform a specific function that remains to be clarified.

More recently, the pre-mRNA encoding UPF1 has also been shown to undergo alternative splicing. This affects the “regulatory loop” between the UPF1 RecA2 and 1B domains, which can be either 11 or 22 amino acids long ^[39][48]. The consequence of this difference is modulation of the affinity of UPF1 for RNA and of its sequence selectivity ^[40][49].

There exist links between NMD and the ER. This has notably been highlighted in a recent study showing that part of the NMD machinery localizes to the translocon, which sends newly synthesized proteins to the ER lumen. NMD thus ensures quality control of the mRNAs translated there ^[41][50].

Cells undergo a wide variety of stresses, such as hypoxia, stresses induced by extracellular agents (pathogens, drugs), as well as cell-homeostasis-disrupting ER stress. If a stress is too strong or prolonged and cannot be resolved, then apoptosis is initiated. One-way cells regulate the stress response is by controlling synthesis of the factors involved in it. NMD participates importantly in this regulation, particularly in regulating levels of ER stress response factors. Very interestingly, ER stress back-regulates NMD: when ER stress becomes too great, NMD becomes unable to inhibit the stress response and this response causes NMD inhibition. This inhibition involves activation of the protein kinase RNA-like endoplasmic reticulum kinase (PERK), which phosphorylates the eIF2α protein, a subunit of the translation initiation factor eIF2 and participant in the pioneer round of translation ^[42][43][5,51]. Once phosphorylated, eIF2α causes inhibition of translation and hence of NMD. This might be rationalized as follows: if a stress is too strong to be coped with, the solution is to go towards cell death. This involves inhibition of NMD, leading notably to expression of apoptotic genes such as GADD45b and p21 ^[44][11].

The ability of cardiac glycosides to inhibit NMD was discovered by screening the Pharmakon library, containing approximately 1,600 drug candidates. Cardiac glycosides bind to and inhibit the Na+/K+ATPase, this leading to an increase in the level of intracellular calcium ^[45][52]. Confirmation that a high cytoplasmic calcium level inhibits NMD was obtained by using an ionophore (A23187 or thapsigargin) to induce calcium release from the ER into the cytoplasm in human U2OS cells. How calcium interferes with the NMD machinery remains to be understood. One hypothesis is that an increase in cytoplasmic calcium activates caspase 3 ^[46][53], known to cleave the NMD core factors UPF1 and UPF2 during apoptosis ^[47][48][54,55]. It should be noted that this regulation of NMD by the intracellular calcium level might be cell type dependent: A study done on mouse N2a cells failed to demonstrate any regulation of NMD by calcium ^[49][56]. It is therefore likely that depending on how sensitive a cell type is to changes in intracellular calcium, NMD might be more or less affected. The exact mechanism that might link intracellular calcium to the NMD efficiency thus remains to be elucidated.

One role of quality control mechanisms in the cell is to check the conformity of an mRNA to ensure correct transmission of the message encoded in its gene. As this checking obviously costs energy, it is legitimate to wonder whether a cell engaged in a cell death program continues to devote resources to checking the integrity of mRNAs to be translated. For NMD, two teams answered this question in 2015. They found NMD to be inhibited during apoptosis, following activation of caspases 3 and 7, which cleave the two central NMD factors UPF1 and UPF2 ^[47][48][54,55]. Interestingly, as the fragments thus generated are themselves apoptotic, an apoptosis-amplifying loop is established. Inhibition of NMD during apoptosis thus plays both a passive and an active role: ensuring that the cell does not commit energy resources to this quality control mechanism and contributing towards the cell death objective via production of UPF1- and UPF2-derived apoptotic fragments and through increased expression of NMD-repressed apoptotic genes, such as GADD45b and p21 ^[44][11].

It is very common for a mechanism to self-regulate. For example, the expression of transcription factors is often regulated at the transcriptional level. Likewise, many splicing factors are regulated by alternative splicing. It is therefore not surprising that several central factors of NMD are regulated by NMD. Thus, levels of the proteins UPF1, UPF2, UPF3X, as well as SMG1, SMG5, SMG6, and SMG7 rise when NMD is inhibited ^[50][51][59,61], so as to compensate for decreased NMD efficiency. The corresponding transcripts are natural substrates of NMD, and the underlying mechanism involves their 3’UTR, whose length exceeding 1kb puts them in the ‘long 3’UTR’ category. In addition to a long 3′UTR, the transcripts encoding UPF2, SMG1, SMG5, SMG6, and SMG7 have one or more small upstream ORFs providing another means of NMD activation ^[51][61]. The presence of several elements that make NMD-factor-encoding mRNAs natural substrates of NMD suggests a complex autoregulation which certainly must respond to specific changes in the intra- and/or extracellular environment.