The nonsense-mediated decay pathway maintains synapse architecture and synaptic vesicle cycle efficacy

Post-transcriptional regulation of gene expression has a crucial role in the development, maintenance and plasticity of neuronal synapses. At the mRNA level, translation control in soma as well as remotely in dendrites and axonal growth cones (Bassell and Warren, 2008; Dictenberg et al., 2008; Giorgi et al., 2007; Lin and Holt, 2007; Piper and Holt, 2004) regulates pathfinding, synaptogenesis and synaptic function (Hu et al., 2002; Schacher and Wu, 2002; Sebeo et al., 2009; Sherff and Carew, 1999; Sherff and Carew, 2002; Yan et al., 2009). Neuronal mRNA transport and translation is often mediated by interaction between the 3′ untranslated region (3′ UTR) ‘zipcode’ and mRNA-binding proteins such as zip-code-binding protein (ZBP) and cytoplasmic polyadenylation-element-binding protein (CPEB) (Brittis et al., 2002; Lin and Holt, 2008). Other mRNA-binding proteins, such as fragile-X mental retardation protein (FMRP), similarly have crucial roles in the regulation of synaptic mRNA stability, trafficking and translation (Gatto and Broadie, 2008; Pan and Broadie, 2007; Pan et al., 2004; Pan et al., 2008; Repicky and Broadie, 2009; Tessier and Broadie, 2008; Zhang et al., 2001; Zhang et al., 2005). Moreover, local protein degradation via the ubiquitin proteasome system (UPS) has also recently been established as a key mechanism that shapes synaptic structural development, neurotransmission strength and synaptic plasticity (Haas and Broadie, 2008; Haas et al., 2007a; Haas et al., 2007b; Speese et al., 2003).

The Caenorhabditis elegans and Drosophila genetic systems have provided vital insights into the mechanisms of post-transcriptional regulation in sculpting synaptic properties. Drosophila FMRP is involved in mRNA trafficking and stability (Pan and Broadie, 2007; Pan et al., 2008; Pan et al., 2004; Tessier and Broadie, 2008; Zhang et al., 2001; Zhang et al., 2005), and the activity-dependent regulation of mRNA translation (Gatto and Broadie, 2008; Tessier and Broadie, 2008). At the synapse, FMRP has several functions, including the control of axonal and dendritic arbor size, bouton number and distribution, transmission strength, postsynaptic glutamate receptor trafficking and the regulation of presynaptic vesicle pools (Gatto and Broadie, 2008; Pan and Broadie, 2007; Pan et al., 2008; Pan et al., 2004; Repicky and Broadie, 2009; Tessier and Broadie, 2008; Zhang et al., 2001; Zhang et al., 2005). In balance with translation regulation, UPS-mediated degradation has dynamic functions that control synapse architecture, neurotransmission strength and synaptic protein abundance, including postsynaptic glutamate receptors (Haas and Broadie, 2008; Haas et al., 2007a; Speese et al., 2003). The ubiquitin ligase highwire acts to restrict synaptic overgrowth by down-regulating the MAPKKK–Wallenda pathway, where mutants exhibit increased neuromuscular junction (NMJ) branch and bouton numbers (Wan et al., 2000; Wu et al., 2007). Loss of C. elegans rpm-1 similarly causes disruption of synapse architecture and function (Nakata et al., 2005). RPM-1 negatively regulates the MAP kinase pathway of MAPKKK DLK-1, MAPKK MKK-4 and p38 MAPK PMK-3. Recently, this pathway was shown to also regulate trafficking of the AMPA-type glutamate receptor GLR-1 (Park et al., 2009). Destabilization of enhancer binding protein CEBP-1 mRNA has also recently been shown to alter local translation of MAPKKK components in distal axons to disrupt axon morphology and synapse formation in C. elegans (Yan et al., 2009). Other recent work has shown the importance of microRNA pathways (Cheever and Ceman, 2009a; Cheever and Ceman, 2009b; Kiebler et al., 2006; Xu et al., 2008) and processing bodies (P-bodies) (Barbee et al., 2006; Gibbings et al., 2009; Miyoshi et al., 2009; Pillai et al., 2005).

Nonsense-mediated decay (NMD), an mRNA surveillance system that ensures message integrity by degrading transcripts containing nonsense mutations (Bramham et al., 2008; Giorgi et al., 2007), represents a relatively unexplored mechanism for regulating mRNA stability in neurons. Importantly, NMD is also proposed to regulate normal transcript expression, and so provide an alternative mechanism of control. Seven NMD pathway ‘suppressor with morphogenetic effect on genitalia’ (Smg) genes were originally identified in C. elegans (Hodgkin et al., 1989), with six gene homologs subsequently identified in Drosophila (Gatfield and Izaurralde, 2004; Metzstein and Krasnow, 2006). It was recently demonstrated that eIF4AIII, a core exon junction complex (EJC) component, is associated with neuronal mRNA granules (Barbee et al., 2006; Giorgi et al., 2007; Shibuya et al., 2004; Shibuya et al., 2006). In mammals, the EJC is thought to have a crucial role in directing transcripts to the NMD pathway. This system controls glutamate receptor expression and long-term potentiation at the synapse via regulation of Arc protein expression (Bramham et al., 2008; Giorgi et al., 2007; Waung et al., 2008). Predicted NMD pathway targets include a number of synaptic genes; however, the functional requirement of NMD in maintaining the synapse has not before been directly tested in vivo.

Classical Drosophila forward genetic screens to generate mutants with defective adult eye electroretinograms resulted in isolation of a pool of mutants with profoundly disrupted photoreceptor synaptic transmission. These genes were targeted to identify novel mechanisms required for synapse formation and function. Cloning and genomic rescue of the no-on-and-off transient C (nonC) mutant from this pool showed it to be the phosphatidylinositol 3-kinase-like kinase (PIKK) Smg1 gene, the regulative kinase of the NMD pathway. Detailed analyses of the NMJ synapse and a range of Smg1 alleles showed defects in morphological architecture and impaired synaptic function owing to disrupted synaptic vesicle cycling. Similarly, mutation of other key NMD pathway components, including Upf2 (Smg3) and Smg6 mutants that abrogate NMD and cause accumulation of PTC-containing transcripts (Metzstein and Krasnow, 2006), comparably impairs presynaptic function. These results demonstrate that NMD mRNA regulation is crucial for the proper development of synaptic architecture and for the maintenance of synaptic transmission efficacy.

The ERG consists of a corneal-negative (downward) sustained component of summed photoreceptor phototransduction, and two transient components at lights-on and lights-off arising from neurotransmission in the lamina (Fig. 1A). The on-transient, in particular, corresponds directly to responses of laminar neurons to synaptic input from photoreceptors R1–R6 (for a review, see Pak, 1995). In wild-type animals, the peak amplitude of the sustained component increased with the strength of the light stimulus in a semi-log-dependent manner over a 3 log unit range of stimulus intensity (Fig. 1A). Synaptic transient amplitudes were lower only at the weakest light stimulus intensity, and then increased to a constant level over an increase in light intensity of three orders of magnitude (Fig. 1A). The peak amplitude of the phototransduction component and the synaptic on- and off-transient amplitudes were quantified and plotted as a function of the stimulus intensity (Fig. 1B).

In the nonC mutant isolated in the genetic screen for photoreception mutants (nonC^P37), robust phototransduction persisted, but the synaptic transients were not detectable (Fig. 1A). Neither on-transients (Fig. 1Ba) nor off-transients (Fig. 1Bc) of any significant amplitude could be elicited from the nonC^P37 mutant at any of the four light intensities tested. An independently isolated nonC mutant allele, nonC^MC45, showed a similar, selective synaptic impairment (Fig. 1A). This mutant also had no detectable photoreceptor synaptic on-transients (Fig. 1Ba), and displayed off-transients of reduced amplitude (Fig. 1Bc). The peak phototransduction responses of photoreceptors were slightly depressed in nonC^P37 and comparable with the wild type in nonC^MC45 (Fig. 1Bb). The loss of photoreceptor synaptic transmission in both nonC mutant alleles was totally rescued by introduction of the wild-type candidate gene (Fig. 1A, right; Fig. 1B), demonstrating that the synaptic defect is due solely to the loss of this single gene function (see below).

Candidates for the nonC gene were identified by a microarray-based approach screening for statistically significant alterations in mRNA levels of genes within the chromosomally mapped mutant interval (Fig. 2) (Leung et al., 2008; Long et al., 2008). Deficiency mapping using ERGs to examine resulting photoreception phenotypes to determine whether complementation had occurred placed nonC between the right breakpoint of deficiency Df(1)ED6849 (X:6698829) and right breakpoint of duplication Dp(1,Y)dx49 (6D8) (Fig. 2A). To accommodate possible mismatch between reported chromosomal breakpoints and gene locations, microarray data were examined in a wider region than that defined by mapping; namely 6C13-6E2, the 120 kb region between coordinates X:6660000 and X:6780000, which contains 17 genes. Microarray data showed that four genes showed statistically significant changes in mRNA levels in nonC mutants: shf, Smg1, CG4557 and CG4558 (Fig. 2A). Thus, these genes were identified as candidates.

All genes were fully sequenced in three different nonC mutants; nonC^P37, nonC^MC45 and nonC^MC47. No mutations were found in the shf, CG4557 or CG4558 genes in any nonC mutants. By contrast, the Smg1 gene carried a T to C transition in nucleotide 2552, resulting in the isoleucine to threonine change in residue 851 in both nonC^MC45 and nonC^MC47 (Fig. 1B). Four other genes in the Smg1 vicinity (6D4-7 region; Fig. 2A) and two genes in the 6E4 region were also sequenced. No mutations were detected in any of these genes in nonC^P37, nonC^MC45 or nonC^MC47. The genes sequenced included CG12796, which resides within the fourth intron of Smg1. These results suggested that Smg1 was the most likely candidate for the nonC gene. Following this identification, previously described Smg1 alleles were obtained, including Smg1^rst2, a premature stop codon resulting at residue 704 AAA to TAA (Chen et al., 2005), and Smg1^32AP, a genetic null as a result of a point mutation in the Smg1 kinase domain at residue 1651 (Metzstein and Krasnow, 2006) (Fig. 2B).

Three approaches were used to test the validity of the Smg1 gene identification: (1) assay of Smg1 mRNA levels in nonC mutants by quantitative PCR, (2) transient induction of RNA interference (RNAi) by direct injection of double-stranded Smg1 mRNA fragments, and (3) introduction of a wild-type Smg1 genomic construct in nonC mutants to assess phenotype rescue. Results of quantitative real-time PCR showed that the Smg1 mRNA levels were reduced 20–50%, with greater loss in nonC^P37 than in nonC^MC45. This allele dependence paralleled the relative severity of the ERG phenotype (Fig. 1). With the second validation approach, siRNA knockdown of Smg1 expression in the wild type accurately phenocopied nonC defects in synaptic transmission (see below). Thus, the first two approaches validate the identification of the Smg1 gene.

The most definitive proof of gene identity is to introduce the candidate wild-type gene into the mutant and assay rescue of mutant phenotype. We used a genomic Smg1 construct driven by the native promoter to ensure proper spatiotemporal expression. The Smg1 gene is ~13.6 kb and the predicted promoter region is ~3.2 kb, generating a sequence of nearly 17 kb and making the use of P-element-mediated transformation questionable. We therefore used P[acman]-mediated recombination (Venken et al., 2006) and ΦC31-mediated genome integration (Groth et al., 2004). Insertion of the genomic construct in several Smg1 mutant backgrounds was confirmed by PCR. Two copies of wild-type Smg1 completely rescued the nonC synaptic phenotype in the adult visual system (Fig. 1) and larval NMJ (see below; Fig. 3) in both the nonC^P37 and nonC^MC45 mutant alleles. These results together strongly support the conclusion that nonC and Smg1 are the same gene. Drosophila Smg1 encodes PIKK, with 34.1% protein sequence identity between CG32743 and human SMG1 (Fig. 2C) (Chen et al., 2005).

The NMJ was used to systematically characterize the role of Smg1 at the synapse. To first assess function, neurotransmission recordings were done on eight genotypes; an allelic series of four independent Smg1 mutants, two transheterozygous mutant combinations, a wild-type control and a control of a mutant containing the wild-type Smg1 genomic construct (rescue). Fig. 3 shows representative EJC responses to single (Fig. 3A) and repetitive (Fig. 3B) nerve stimulation, and the quantified results on EJC amplitude (Fig. 3C) and fidelity (Fig. 3D).

With basal stimulation (0.5 Hz), control synapses exhibited robust, high-fidelity transmission, whereas all Smg1 mutants showed a similar ~50% decrease in EJC amplitude (Fig. 3A), with significant loss of transmission fidelity (Fig. 3B). Controls displayed a mean EJC amplitude of 72.5±4.1 nA and mutants exhibited mean amplitudes of 39.0±7.1 nA (Smg1^nonCMC45), 41.2±5.0 nA (Smg1^nonCP37), 38.4±5.7 nA (Smg1^rst2) and 43.1±4.9 nA (Smg1^32AP) (Fig. 3C). EJC amplitude in all mutants was significantly decreased compared with the control (P<0.003; n≥8 animals for each genotype), but did not differ significantly between mutant alleles (P>0.5). The heteroallelic mutant combination Smg1^nonCP37/Smg1^rst2 showed a similarly decreased EJC amplitude (38.4±6.2 nA) compared with the heteroallelic Smg1^nonCP37/+ control (74.8±4.5 nA; P<0.003). Wild-type Smg1 in the Smg1^nonCP37 homozygous background (rescue) restored EJC amplitude to slightly above control level (85.1±7.5 nA; Fig. 3A–C). These results indicate that loss of Smg1 alone impairs basal synaptic transmission.

Presynaptic or postsynaptic defects can impair neurotransmission. One method by which mechanistic defects can be differentiated is to assay spontaneous synaptic vesicle fusion, or miniature EJC (mEJC) events, occurring in the absence of action potentials (Gatto and Broadie, 2008; Long et al., 2008; Trotta et al., 2004). In Smg1 mutants, there were no detectable changes in mEJC amplitude or frequency compared with controls. The mean mEJC frequencies were: wild-type control, 1.07±0.2 Hz; Smg1^nonCP37, 1.2±0.3 Hz; Smg1^nonCMC45, 1.2±0.3 Hz; Smg1^rst2, 1.1±0.3 Hz; Smg1^32AP, 1.1±0.3 Hz (P>0.5 in all cases; n≥8 animals per genotype). These results suggest that there is no significant alteration in the probability of spontaneous synaptic vesicle fusion at active zones. The mean mEJC amplitudes were: control, 0.7±0.2 nA; Smg1^nonCP37, 0.8±0.3 nA; Smg1^nonCMC45, 0.7±0.2 nA; Smg1^rst2, 0.8±0.3 nA; and Smg1^32AP, 0.7±0.3 nA (P>0.5 in all cases; n≥8 animals per genotype). These results suggest that there is no significant change in glutamate receptor density or channel function at individual postsynaptic sites. Together, these data indicate a specific impairment in evoked, Ca²⁺-influx-dependent presynaptic neurotransmitter release.

Smg1 mutants displayed impaired transmission fidelity with repetitive stimulation (Fig. 3B), which is consistent with presynaptic dysfunction. Amplitude variation was calculated by dividing the standard deviation by the mean EJC amplitude for each recording, and then averaging the variation values of animals within the same genotype. For mutants, these values were: Smg1^nonCP37, 0.57±0.28 (P<0.004); Smg1^nonCMC45, 0.44±0.27 (P<0.03); and Smg1^rst2, 0.41±0.21 (P<0.04) compared with the genomic rescue line 0.18±0.08 (P<0.04 compared with control; Fig. 3D; n≥8 animals for each genotype). The Smg1^32AP allele 0.33±0.15 was significantly more variable than the genomic-rescue control, but not the wild type (P=0.4). For the transheterozygote Smg1^nonCP37/Smg1^rst2, variance was calculated at 0.43±0.04 compared with the transheterozygous control at 0.28±0.15 (P<0.04). Taken together, these results indicate that loss of Smg1 function impairs basal synaptic efficacy and reduces the fidelity of neurotransmission.

Defective synapse function might be caused by perturbed development. We assayed NMJ structure using anti-HRP, a presynaptic membrane marker that reveals the terminal area, branching pattern and bouton deposition. In parallel, we assayed key molecular components of the synapse, including anti-bruchpilot (NC82) in presynaptic active zones, and anti-discs large (DLG) and glutamate receptor subunit IIA (GluR) in the apposing postsynaptic domains. Two independent mutant alleles Smg1^nonCP37 and Smg1^nonCMC45, and a third mutant transheterozygous condition, were compared with the wild type and genomic-rescue controls at the muscle 6–7 NMJ in segment A3. The summarized results are shown in Fig. 4.

Loss of Smg1 resulted in structurally under-elaborated NMJ with reduced terminal area, fewer synaptic branches and boutons, and a synapse that was generally confined closer to the muscle nerve entry site (Fig. 4A–C). Overall, mutant synaptic area was significantly reduced (P<0.03; n≥8 animals for each genotype). Wild-type controls had a mean area of 2414±210 μm² compared with 1502.8±142 μm² in Smg1^nonCP37, and 1607±157 μm² in Smg1^nonCMC45 (Fig. 4D). This growth defect could be effectively rescued with genomic Smg1 (2561.0±135.3 μm²). The transheterozygous mutant Smg1^nonCP37/Smg1^rst2 exhibited an area of 1524.8±119.0 μm² compared with the transheterozygous control Smg1^nonCP37/+ of 2315.2±172.1 μm². In parallel, there was a more highly significant (P<0.003; n≥8 animals for each genotype) reduction in the number of synaptic terminal branches in mutants (Fig. 4E). A branch was defined as axonal arbor process having two or more clearly defined boutons. The wild type had an average of 12.2±3.1 branches, approximately twice as many as Smg1^nonCP37 (5.2±0.4), Smg1^nonCMC45 (5.8±1.5) and Smg1^rst2/p37 (6.2±2.1). This phenotype could also be rescued with wild-type Smg1 (11.8±3.2; Fig. 4E). Similarly, the number of synaptic boutons, defined as any clear axonal swelling that was greater than 1 μm in diameter, was significantly reduced in Smg1 mutants (39.8±7.1, Smg1^nonCP37; 47.5±6.5, Smg1^nonCMC45; 52.6±6.8, Smg1^rst2/p37) compared with controls (wild type, 87.5±11.8; rescue, 93.1±4.9; P<0.003, n≥8; Fig. 4F). Thus, loss of Smg1 causes impaired synaptic structural development.

Synapse molecular components were next examined to determine whether individual synaptic bouton differentiation was compromised by the loss of NMD function. Fluorescence intensity and the number of fluorescent punctae was measured in muscle 6–7 NMJ boutons in both right and left bilateral A3 hemisegments, and then averaged to generate a single data point for each larvae. Eight animals were assayed in each of five genotypes. First, the presynaptic active zone protein Bruchpilot, labeled with anti-NC82, was used to examine synaptic vesicle release sites (Fig. 4A). Counting of the number of anti-NC82 puncta within individual boutons revealed a noticeable but statistically insignificant reduction in Smg1 mutants (Smg1^nonCp37, 84.7±8.7; Smg1^nonCMC45, 79.3±7.8; Smg1^rst2/p37 89.5±6.7) compared with the wild-type (125.2±11.7) and genomic rescue (119.3±10.2) controls. Second, components of the postsynaptic domain were similarly analyzed, including glutamate receptors (GluRIIA; Fig. 4B) and membrane scaffold Discs Large (DLG; Fig. 4C). No significant changes were evident in either GluRIIA (control, 67.9±1.2; Smg1^nonCp37, 62.8±1.0; Smg1^nonCMC45, 58.2±1.2; Smg1^nonCrst2/p37, 65.3±1.3; genomic rescue, 70.5±1.8) or DLG (control, 74.9±5.8; Smg1^nonCp37, 63.8±4.2; Smg1^nonCMC45, 71.6±6.1; Smg1^rst2/p37, 59.7± 8.2; genomic rescue, 78.2±9.1) fluorescence intensities (Fig. 4B,C). Together, these data show that Smg1 facilitates gross NMJ synaptic morphological development, but does not appear to be required for maintenance of pre- and postsynaptic molecular specializations in individual boutons.

Smg1 mutant disruption of NMJ morphology might be sufficient to explain impaired neurotransmission. The reduction in synaptic bouton number without strong compensation in individual boutons should cause functional impairment. However, synaptic structure and function are not well correlated, and indeed are separately regulated. Therefore, it is possible that other synaptic defects occur when NMD function is disrupted. To test these possibilities, the NMJ was first examined by transmission electron microscopy. Representative synaptic bouton images and results are shown in Fig. 5.

Quantified features included synaptic bouton cross-sectional area, mitochondria number and area, synaptic vesicle number and density, active zone number and density, number of clustered vesicles at active zones (<250 nm from t-bar), number of membrane-docked vesicles at active zones (<20 nm from t-bar), and the internal pool of internal synaptic vesicles (>250 nm and <500 nm from t-bar). Many of these features were indistinguishable between wild-type and Smg1 mutants, but several changes were found. Bouton size was significantly larger in Smg1 mutants (Smg1^nonCMC45 5.4±0.6 μm²; Smg1^nonCP37 5.3±0.5 μm²) compared with the control (3.1±0.2 μm²) (P<0.001; n=57 control, 50 Smg1^nonCMC45 and 69 Smg1^nonCP37 boutons). Consistently, mutants contained significantly (P=0.01) more synaptic vesicles (Smg1^nonCMC45, 317±22 vs control, 257±14), although mean vesicle density was still significantly (P<0.009) reduced (e.g. Smg1^nonCMC45, 82±7 vs control, 104±6), owing to the enlarged bouton area (Fig. 5A). In addition, vesicle pool distributions were slightly altered in the absence of Smg1 function. The number of docked vesicles was significantly (P<0.0001) elevated in mutants (e.g. Smg1^nonCMC45, 2.5±0.1) compared with the wild type (1.9±0.1) (Fig. 5B, top). The number of vesicles clustered around the active zone was also significantly increased (P<0.04) in mutants (e.g. Smg1^nonCP37, 19.0±1.7) compared with the control (15.0±1.3) (Fig. 5B, middle). By contrast, there was a highly significant (P<0.0001) reduction in the number of internal synaptic vesicles in mutants (e.g. Smg1^nonCMC45, 25.2±2.0) compared with the control (36.0±2.0) (Fig. 5B, bottom). Thus, Smg1 is involved in regulation of the trafficking underlying synaptic vesicle pools.

To investigate synaptic vesicle pool size and cycling dynamics, the lipophilic fluorescent dye FM1-43 was used to visualize vesicular turnover (Fig. 6). Acutely dissected NMJ preparations were exposed to FM1-43 (10 μM) during a 1 minute 20 Hz stimulation of the motor nerve. Preparations were then washed in Ca²⁺-free saline to halt vesicle cycling and remove external dye, and then imaged to visualize dye uptake (Fig. 6A). The Smg1 mutants were clearly impaired in the cycling rate of synaptic vesicles compared with controls, with immediately obvious reduced dye incorporation. Mean fluorescent intensities were quantified within individual boutons averaged in n≥8 animals for each genotype (Fig. 6B). Wild-type boutons had a mean fluorescent intensity of 130.87±13.6 compared with 45.07±7.8 in Smg1^nonCP37 and 49.9±16.6 in Smg1^nonCMC45. The transheterozygous mutant Smg1^nonCP37/Smg1^rst2 exhibited an intensity of 42.5±14.1 compared with 118.3±8.7 in the Smg1^nonCP37/+ transheterozygous control. The decreases in the mutants were all highly significant (P<0.003) compared with control levels. Introduction of wild-type Smg1 resulted in an elevated level of dye loading (145.0±23.2), which was not significantly different from the wild type (Fig. 6B).

The fluorescent FM1-43 signal was photoconverted to an electron-dense signal for EM comparisons (Vijayakrishnan et al., 2009). The converted signal was clearly detected at the light microscopy level to again reveal impaired FM dye loading in Smg1 mutants (Fig. 7A; compare with Fig. 6A). At the EM level, control boutons loaded dye much more readily within synaptic vesicles, filling the majority of vesicles in any given terminal under these conditions (Fig. 7B). By contrast, mutant boutons contained relatively few labeled vesicles. Loading was quantified by counting the number of electron-dense vesicles per bouton cross-section. Controls exhibited an average of 172.0±23.4 loaded vesicles compared with 35.3±3.4 in Smg1^nonCMC45 (P<0.002; sample size n=7 animals and 57 boutons for control, and 7 animals and 55 boutons for Smg1^nonCMC45; Fig. 7C). Together, these results show impaired Smg1 presynaptic function at the level of whole animal and single synaptic boutons.

The reduced population of cycling synaptic vesicles in Smg1 mutants predicted a more severe reduction in mutant synaptic transmission amplitude than documented above (see Fig. 3). A likely explanation is that the high-frequency stimulation (HFS) required for the FM1-43 dye loading experiments revealed a greater level of functional impairment under conditions of elevated use. To test this hypothesis, the same 20 Hz HFS paradigm was used in physiological recordings (Fig. 8). Following basal EJC stimulation at 0.2 Hz for 50 seconds (ten stimuli), EJC amplitudes were assessed during a 60 second, 20 Hz HFS train (1200 stimuli), followed by a 50 second, post-HFS recovery at 0.2 Hz stimulation (10 stimuli). Fig. 8 shows representative EJC responses (Fig. 8A) and normalized EJC amplitude results (Fig. 8B).

At 0.5 mM external [Ca²⁺], wild-type NMJs maintained a nearly constant level of transmission, with no significant tendency to facilitate or fatigue (Fig. 8A, top; Fig. 8B). Transheterozygote controls (e.g. Smg1^nonCP37/+) were indistinguishable from the wild type, similarly maintaining normalized values near 1.0 and never decaying below 0.94 of basal values. By contrast, Smg1 homozygous and transheterozygous mutants started out impaired by ~50% (see Fig. 3) and then experienced further significant fatigue during the HFS train (Fig. 8A, bottom). Initially, normalized EJC amplitudes demonstrated a small amount of facilitation (<15 seconds), but this facilitation was not maintained (Fig. 8B). However, from ~20 seconds, Smg1 mutants showed an overall steady decline in transmission amplitudes for the remainder of the stimulus train (Fig. 8B). Upon return to basal (0.2 Hz) stimulation, the Smg1 mutant EJC amplitudes gradually recovered towards their initial basal level, with a time-course mirroring (30–40 seconds) the rate of fatigue during the HFS train (data not shown). Thus, under the same conditions used for FM1-43 dye loading, functional transmission in Smg1 was reduced to ~20% of control levels.

We next examined synaptic ultrastructure with the HFS train (20 Hz, 1 minute). Fig. 8C shows representative active zone images from the wild-type control and Smg1^nonCMC45 mutant with no stimulation (rest) and with HFS (stim). Synaptic vesicle and membrane trafficking parameters were quantified (n=7 animals, n=20 for each genotype and each condition) with remarkably different consequences in Smg1 mutants compared with the control. With regard to bouton size, HFS increased the mean area from 3.1±0.2 μm² to 4.8±0.6 μm² in the control, but decreased the area from 5.4±0.6 μm² to 3.8±0.4 μm² in the Smg1 mutant (P<0.01; Fig. 8D). Although control boutons showed no significant change in the number of enlarged (>60 nm) membrane cisternae (mean 3.7±1.4 following HFS), stimulated mutant boutons showed a highly significant elevation of these trafficking organelles (mean 6.3±1.2; P<0.01; Fig. 8E). With regard to vesicle pools, the total synaptic vesicle number following HFS for controls was 248±30 compared with 179±18 in the mutant (P<0.05), but with no significant difference remaining in vesicle density (P=0.3). Stimulated wild-type controls maintained both the clustered active zone and membrane-docked vesicle pools (Fig. 8F,G). By contrast, stimulated Smg1^nonCMC45 mutants showed a highly significant loss of clustered vesicles (from 12.0±1.0 with HFS vs 18.0±0.8 at rest), resulting in a significantly (P<0.05) smaller pool than the stimulated control. Similarly, Smg1^nonCMC45 mutants lost docked vesicles from 2.5±0.1 per active zone at rest to 1.3±0.2 in the stimulated condition (Fig. 8G). Together, these data show that loss of Smg1 prevents the synaptic vesicle cycle from maintaining the rate required during periods of demand for high-frequency transmission.

Smg1 acts as a regulative kinase that triggers the formation of the mRNA surveillance complex during NMD. The above impairments were therefore hypothesized to result from loss of this NMD mechanism. However, it is possible that Smg1 acts independently of NMD in some unique role. To determine whether Smg1 acts as part of the NMD pathway, independent mutations in other components of the Smg complex were assayed. The Upf2 (Smg3) mutant was isolated in a genetic mosaic screen of EMS-induced mutations. The Upf2^25G allele acts as a genetic null that abrogates NMD (M.M.M., unpublished results). The Smg6 mutant was isolated in a genetic screen for mutations that affect NMD using an NMD-sensitive reporter construct. The Smg6²⁹² allele acts as a genetic null with greatly reduced NMD activity (M.M.M., unpublished results). Fig. 9 shows synaptic functional assays in these two mutants with both TEVC electrophysiology and FM1-43 dye imaging.

Evoked neurotransmission was assayed with 0.5 Hz nerve stimulation in 0.5 mM Ca²⁺ external saline (Fig. 9). Representative EJC traces for the wild-type control and the two null mutants are shown in Fig. 9A. The controls displayed a mean EJC amplitude of 75.2±7.89 nA, whereas Upf2 (Smg3) mutants exhibited average amplitudes of 36.1±4.78 nA and Smg6 mutants had average amplitudes of 41.3±5.98 nA (P<0.004; n=8 animals for all three genotypes; Fig. 9B). The severity of the neurotransmission deficit for both of these NMD pathway components was comparable with the Smg1 mutants (see Fig. 3), indicating that NMD has an important role in the regulation of synaptic function, and that loss of Smg1 is equivalent to loss of NMD function in this requirement.

The synaptic vesicle cycle was next assayed by FM1-43 dye incorporation following a 20 Hz, 1 minute stimulation of the motor nerve in 0.5 mM Ca²⁺ saline. Representative NMJ images of the wild-type control and the two null mutants are shown in Fig. 9C. The control loaded much more dye than either NMD pathway mutant, with a qualitatively clear difference in fluorescence intensity. Controls showed a mean fluorescent intensity of 148.1±16.17 compared with Upf2 (Smg3) mutants at 32.8±8.51 and Smg6 mutants at 39.9±9.68 (both P<0.002 compared with control; n=8 animals per genotype; Fig. 9D). These defects are similar too, albeit slightly more severe than Smg1 mutant phenotypes (compare with Fig. 6). Taken together, these data suggest that Smg1 operates in the same mechanism as Upf2 (Smg3) and Smg6, and therefore that the NMD pathway has a crucial role in the functional differentiation of the synapse.

As part of a long-term, unbiased, forward genetic screen for synaptic dysfunction mutants in the Drosophila visual system, we have identified mutants in Smg1, which is a key regulatory kinase in the NMD pathway. NMD is an mRNA surveillance system that is critical for maintenance of transcript integrity. The Smg proteins (Smg1–Smg7) act together to degrade mRNA transcripts containing nonsense mutations that would produce truncated proteins with potentially deleterious activities. This study reveals that Smg1 functions to protect the development of synaptic morphological architecture and maintain high-fidelity neurotransmission function, especially under conditions of heightened activity. Among isolated Smg mutants, only Smg1 nulls are viable as adults; Smg5 mutants die as first instars, and Upf2 (Smg3) and Smg6 die as pupae. These lethal periods have restricted our adult analyses to Smg1 mutants, and our larval analyses to Smg1, Smg3 and Smg6 mutants, and might indicate that Smg1 is less essential for NMD than the other complex components. Together, our studies indicate that NMD activity is required in at least two disparate synaptic classes, retinal photoreceptor synapses and neuromuscular junction, suggesting that an NMD-mediated protective mechanism is broadly important in synapses throughout the nervous system.

We, and others, have exerted a great deal of effort exploring the roles of mRNA regulation in synaptic mechanisms (Antar and Bassell, 2003; Gatto and Broadie, 2009; Pfeiffer and Huber, 2009; Tessier and Broadie, 2009; Yan et al., 2009). In particular, mechanisms of mRNA stabilization, trafficking, RISC-mediated degradation and localized translation have all been recently determined to occur in proximity to synapses and/or with important roles in synapse regulation (Giorgi et al., 2007; Hengst and Jaffrey, 2007; Lin and Holt, 2007; Sebeo et al., 2009; Yan et al., 2009). These insights have raised the question of possible NMD involvement (Giorgi et al., 2007). In this study, we report that disruption of NMD-mediated mRNA regulation results in impaired morphological NMJ development, loss of functional transmission at central (histaminergic) synapses and peripheral (glutamatergic) synapses, and impaired synaptic vesicle cycling at the NMJ, especially under conditions of elevated demand. These results indicate that NMD has an important role in maintaining synapse architecture and high-fidelity synaptic efficacy.

Synaptic morphological defects can be readily observed with the loss of appropriate mRNA regulation of synaptic components. This requirement has been characterized for a number of mRNA-binding proteins, prominently including the fragile-X mental retardation protein (FMRP) (Gatto and Broadie, 2008; Pan and Broadie, 2007; Pan et al., 2008; Tessier and Broadie, 2008; Tessier and Broadie, 2009). Here, we show that loss of mRNA regulation via disruption of the NMD pathway also impairs synaptic architecture at the well-characterized Drosophila NMJ. In Smg1 mutants, NMJ synaptic arbors appear underdeveloped and underelaborated, with limited differentiation outside the immediate vicinity of the initial point of muscle innervation. The range of morphological phenotypes includes overall reduction in synaptic terminal area, decrease in synaptic arbor branching by ~50% and decrease in synaptic bouton number by ~50%.

Defects in NMD mutants are presumably caused by the production of truncated proteins with dominant-negative or deleterious gain-of-function activities that damage the machinery of synaptic development. We propose that Smg mutants are unable to prevent translation of damaged transcripts, and thus the production of ‘toxic proteins’ results in smaller, less-developed and functionally compromised synapses. Where and when the NMD machinery performs this essential activity is uncertain; it could be wholly in the neuronal soma, or locally at synapses for transcripts that might undergo local translation. We note, however, that the Smg1 mutant defects revealed here appear largely restricted to the presynaptic domain; postsynaptic scaffolding and glutamate receptor expression is not detectably altered, and there is no detectable change in postsynaptic physiological function (DiAntonio et al., 1999). Local mRNA translation in growth cones has been demonstrated to be crucial for axon pathfinding and synapse maturation (Brittis et al., 2002; Hu et al., 2002; Hu et al., 2004; Lin and Holt, 2008; Sebeo et al., 2009; Yan et al., 2009). The defects reported here could occur in the motor neuron growth cone or during later stages of synaptic maturation.

The cyclic events that orchestrate the release of neurotransmitter at the synapse consist of a rapidly repeating series of vesicular trafficking, exocytosis and endocytosis steps. There are many molecular stages in this pathway that are susceptible to disruption, which when impaired cause similar defects in vesicle cycling. Consistent with a protective mechanism for the NMD pathway in this mechanism, experimental introduction of peptide fragments of synaptic vesicle cycle proteins is a well-established means of investigating molecular requirements (Gitler et al., 2008; Gitler et al., 2004; Hilfiker et al., 2005; Morgan et al., 2003). The multiple Smg gene mutants analyzed here all show profound disruptions in the endocytosis phase of the synaptic vesicle cycle, revealing an impairment in the ability to retrieve vesicular membrane and associated proteins from the plasma membrane after vesicle fusion. Moreover, disruption of the NMD pathway causes a significantly reduced functional readily releasable pool of vesicles, although the morphological releasable pool appears relatively intact in Smg1 mutants after periods of both basal and intense synaptic activity. This suggests that the loss of neurotransmission in Smg mutants is due not only to impaired vesicle biogenesis, but perhaps also to the interruption of molecular functions underlying evoked vesicle exocytosis.

A long list of proteins has demonstrated roles in synaptic vesicle endocytosis and exocytosis (Owald and Sigrist, 2009; Sudhof, 2000; Sudhof, 2004; Sudhof and Rothman, 2009). Loss of NMD probably generates ‘toxic protein fragments’ that interfere with many of these proteins, both individually and by preventing the formation of presynaptic molecular complexes. One of the most predicted consequences of such fragmentary proteins is to act as dominant negatives by interfering with (or out-competing) appropriate protein–protein interactions during complex formation. The Smg mutants show a more-severe impairment of presynaptic function under conditions of high-frequency stimulation. During such periods of elevated demand, the stress placed on the synaptic vesicle machinery is obviously greater, demanding that proteins function more rapidly and that molecular complexes form and function at a faster rate. Our model predicts that mutant proteins produced in the absence of the NMD pathway would therefore have a greater deleterious impact under conditions of high-frequency stimulation. The inhibition caused by truncated synaptic vesicle proteins presumably accounts for the very small amount of vesicle cycling evident at the Smg mutant synapse during periods of high demand.

It was recently shown that loss of the NMD-initiating factor and exon-junction complex component eIF4AIII leads to synaptic functional impairments (Bramham et al., 2008; Giorgi et al., 2007). It is also known that localized mRNAs encode receptors, cytoskeletal proteins and regulatory proteins that are critical for synaptic development, and that the localized translational regulation of these mRNAs is crucial for synaptic stability and function (Sutton and Schuman, 2009; Ule and Darnell, 2006). Our future work will focus on the identification of the mRNA species encoding synaptic proteins that accumulate in the absence of the NMD pathway, mRNAs locally translated in the late-maturing axonal growth cone and potentially presynaptic terminal, and study of how the production of aberrant proteins synthesized in the absence of NMD activity impairs synaptic architecture and presynaptic function.

The first nonC mutant (nonC^P37) was isolated by ethyl methanesulfonate (EMS) mutagenesis of Oregon-R (OR) base stock (Pak et al., 1969). The nonC^MC45 and nonC^MC47 mutants were generated independently by P-element hybrid dysgenesis (Engels et al., 1989). The method of microarray-based identification of genes with mutations responsible for ERG defects has been described previously (Leung et al., 2008; Long et al., 2008). The Smg1^rst2 allele is a premature stop (AAA to TAA) at amino acid 704 (Chen et al., 2005). The Smg1^32AP allele is a genetic null isolated in a mosaic screen of EMS-induced X-chromosome mutations (Metzstein and Krasnow, 2006). The Smg6²⁹² mutant, isolated in a screen using an NMD-sensitive reporter construct, is a point mutation (GT to AT) disrupting the sixth intron splice site, truncating Smg6 at amino acid 558 to generate a null condition (K. Frizzell, S. Rynearson and M.M.M., unpublished results). The Upf2^25G (Smg3) mutant was isolated in a genetic mosaic screen for EMS-induced X chromosome mutations, resulting in near complete loss of NMD activity (Metzstein and Krasnow, 2006). The Smg1 genomic rescue construct was generated from BAC clone RP98-5L2 containing the Smg1 gene (X:6718671 to 6732283; 13612 bp) and 5′-flanking region (X:6715452 to 6718670; 3218 bp) by recombineering (Groth et al., 2004; Venken et al., 2006). The parental stocks were confirmed by genotyping and mutant phenotype assay. In the rescue combination, the rescue construct and homozygous mutant background were confirmed by genetic markers and PCR.

Wandering third instars were dissected and fixed in Bouin's fixative (10 minutes) or ice-cold methanol (5 minutes), washed in PBS containing Triton X-100 (30 minutes) and then incubated with primary antibodies at room temperature (4 hours). Presynaptic NMJ terminals were visualized with anti-horseradish peroxidase (HRP) [Texas-Red conjugated, 1:200 (Invitrogen) or anti-rabbit, 1:250 (Sigma)]. Presynaptic active zones were visualized using anti-bruchpilot [NC82, anti-mouse, 1:100; University of Iowa Developmental Studies Hybridoma Bank (DSHB)]. The postsynaptic domain was visualized with anti-Discs Large (DLG, anti-mouse, 1:100; DSHB) and anti-GluRIIA (8B4D2, anti-mouse, 1:10; DSHB). Secondary antibodies were Alexa Fluor 488 (1:200, anti-mouse; Invitrogen) and Alexa Fluor 546 (1:200, anti-rabbit; Invitrogen) incubated at room temperature (2 hours). Imaging was done on a Zeiss 510-Meta confocal microscope (Jena, Germany). Branch and bouton number were quantified from HRP-labeled NMJs, with branches and boutons averaged between A3 hemisegments in each animal (for each n=1). Fluorescence intensities were measured in entire NMJs or individual boutons using Metamorph software (MDS Analytical Technologies).

Electroretinogram (ERG) recordings were made as previously described (Long et al., 2008). In brief, with recording electrode through the cornea in the photoreceptor layer and ground electrode in the head, animals were dark-adapted (2 minutes) and then a 300 W halogen lamp used to generate light stimuli (CS2-73 filter) via a light guide. Light intensity was attenuated in log-unit steps using neutral density filters from an unattenuated intensity of 830 μW/cm². Signals were sampled at 2 kHz with an analog-to-digital converter, and data analyzed with Axoscope software (Molecular Devices, Sunnyvale, CA). Two-electrode voltage-clamp (TEVC) recordings were made at the wandering third instar NMJ as previously described (Long et al., 2008; Rohrbough et al., 1999). In brief, muscle 6 in segment A3 was voltage-clamped (V_hold=−60 mV) with current recordings made on an Axoclamp 200B amplifier (Molecular Devices) and analyzed with Clampex 7.0 software (Axon Instruments). Recording saline contained 128 mM NaCl, 2 mM KCl, 4 mM MgCl₂, 0.5 mM CaCl₂, 5 mM trehalose, 70 mM sucrose and 5 mM HEPES. Miniature excitatory junctional currents (mEJCs) were recorded continuously in gap-free recording mode (n=1 represents 120 seconds) in a low-pass setting (500 Hz). For evoked EJCs, cut motor nerves were stimulated with glass suction electrodes at suprathreshold voltage (0.5 mseconds). Currents were filtered at 1000 Hz and analyzed with PClamp 7.0 software.

NMJ ultrastructural analyses were performed as reported previously (Long et al., 2008; Vijayakrishnan et al., 2009). In brief, two approaches were used: (1) dissected third instars were immediately fixed (resting condition), or (2) the cut segmental nerve (segment A3) was stimulated with a glass suction electrode at 20 Hz for 1 minute immediately followed by fixation (stimulated condition). Preparations were fixed in 2% glutaraldehyde (1 hour), washed in PBS (10 minutes) and then transferred into 1% OsO₄ in PBS (2 hours). Segment A3 muscle 6–7 was dissected free for all further processing. Preparations were stained en bloc in 1% aqueous uranyl acetate (1 hour), dehydrated in an ethanol series then propylene oxide (30 minutes) and embedded in araldite. Ultrathin (<60 nm) sections were cut (Leica Ultracut UCT 54 ultramicrotome) and transferred to formvar-coated slot grids on synaptiek grid-sticks for post-staining with lead nitrate and uranyl acetate. Sections were imaged using a Phillips CM10 Transmission Electron Microscope at 80 kV, with Images collected on a 2 megapixel AMT CCD camera.

NMJ FM1-43 dye loading using electrical motor nerve stimulation was done as previously described (Dermaut et al., 2005). In brief, 20 Hz suprathreshold motor nerve stimulation was applied for 60 seconds in 0.5 mM Ca²⁺ saline containing 10 μM FM1-43. Preparations were immediately washed in 0 mM Ca²⁺ saline for 5 minutes, and fluorescent images of NMJ 6–7 terminals acquired on a Zeiss confocal microscope. Fluorescence intensities were determined using Metamorph software (MDS Analytical Technologies) using the hand-select line-drawing tool to trace around each bouton. The fluorescent signal was also photoconverted to an electron-dense signal for ultrastructural analyses, as previously described (Vijayakrishnan et al., 2009). In brief, FM1-43-loaded preparations were fixed in 1.6% paraformaldehyde, 2% glutaraldehyde (5 minutes), washed in Tris-buffered saline (TBS, pH 7.5; 20 minutes), and then incubated in 0.15% 3,3′-diaminobenzidine (DAB, DEKO; 5 minutes). The DAB solution was refreshed and samples illuminated using a 100 W mercury lamp, a 63× 0.95W objective and a standard FITC filter for 20 minutes. Photoconverted preparations (see Fig. 7B) were post-fixed overnight in 2% glutaraldehyde at 4°C, placed in 1% osmium (OsO₄) at room temperature (1 hour) and then placed in uranyl acetate at room temperature (1 hour). Samples were then processed for electron microscopy using the standard protocol, as described above.

ANOVA parametric statistics with Dunnett's analyses were used to determine statistical differences between matched mutants and controls in all studies.

We are grateful to the Bloomington Drosophila stock center and University of Iowa Developmental Studies Hybridoma Bank for Drosophila strains and antibodies, respectively. We are grateful for the help of Phillip SanMiguel and Richard Westerman (Purdue Genomics Core). We thank Peter Coombe in the Heisenberg laboratory (Wurzburg, Germany) for the gift of mutant alleles. This work was supported by NIH grants GM54544 to K.B., and MH07504 to K.B., R.W.D. and W.L.P. M.M.M. was supported by March of Dimes research grant 5-FY07-664 and NIH 1R01GM084011. Deposited in PMC for release after 12 months.