Mutant copper-zinc superoxide dismutase associated with amyotrophic lateral sclerosis binds to adenine/uridine-rich stability elements in the vascular endothelial growth factor 3′-untranslated region

Abstract

Vascular endothelial growth factor (VEGF) is a neurotrophic factor essential for maintenance of motor neurons. Loss of this factor produces a phenotype similar to amyotrophic lateral sclerosis (ALS). We recently showed that ALS-producing mutations of Cu/Zn-superoxide dismutase (SOD1) disrupt post-transcriptional regulation of VEGF mRNA, leading to significant loss of expression. Mutant SOD1 was present in the ribonucleoprotein complex associated with adenine/uridine-rich elements (ARE) of the VEGF 3′-untranslated region (UTR). Here, we show by electrophoretic mobility shift assay that mutant SOD1 bound directly to the VEGF 3′-UTR with a predilection for AREs similar to the RNA stabilizer HuR. SOD1 mutants A4V and G37R showed higher affinity for the ARE than L38V or G93A. Wild-type SOD1 bound very weakly with an apparent K_d 11- to 72-fold higher than mutant forms. Mutant SOD1 showed an additional lower shift with VEGF ARE that was accentuated in the metal-free state. A similar pattern of binding was observed with AREs of tumor necrosis factor-α and interleukin-8, except only a single shift predominated. Using an ELISA-based assay, we demonstrated that mutant SOD1 competes with HuR and neuronal HuC for VEGF 3′-UTR binding. To define potential RNA-binding domains, we truncated G37R, G93A and wild-type SOD1 and found that peptides from the N-terminal portion of the protein that included amino acids 32-49 could recapitulate the binding pattern of full-length protein. Thus, the strong RNA-binding affinity conferred by ALS-associated mutations of SOD1 may contribute to the post-transcriptional dysregulation of VEGF mRNA.

Amyotrophic lateral sclerosis (ALS) is a relentless disease of motor neurons that leads to progressive paralysis of muscles and ultimately death. Approximately one out of ten ALS cases is familial, of which 20% are associated with mutations of Cu/Zn-superoxide dismutase (SOD1) (Rosen et al. 1993; Pasinelli and Brown 2006). Identification of this gene locus prompted the notion that motor neuron degeneration was related to toxicity from inadequate removal of free radicals by mutant SOD1 (Rosen et al. 1993). Further studies of SOD1 mutations in humans and mouse models, however, dispelled this hypothesis. For example, deletion of SOD1 from the mouse genome did not produce an ALS phenotype (Reaume et al. 1996). Moreover, some disease-associated mutants had preserved SOD1 enzyme activity (Cleveland and Rothstein 2001). These findings helped to forge the prevailing hypothesis that mutant SOD1 acquires a toxic function related to protein misfolding and aggregate formation (Cleveland and Rothstein 2001; Pasinelli and Brown 2006). These aggregates can coprecipitate and inactivate other cellular factors or impair normal processes such as chaperoning or proteasomal degradation (Cleveland and Rothstein 2001). We recently found that mutant SOD1 impaired post-transcriptional processing of vascular endothelial growth factor (VEGF) mRNA leading to a significant decline in expression (Lu et al. 2007). VEGF has gained recognition as an important neuroprotective factor for motor neurons and other neuronal subsets in the nervous system (Jin et al. 2000; Oosthuyse et al. 2001; Ogunshola et al. 2002; Brockington et al. 2004; Storkebaum et al. 2004; Tolosa et al. 2008). A direct link between VEGF and ALS was made in a landmark study where depletion of VEGF produced an ALS phenotype in mice (Oosthuyse et al. 2001). In our previous study, we found that impaired VEGF mRNA processing in vivo was mediated by adenine/uridine-rich elements (ARE) in the 3′-untranslated region (UTR) and that mutant SOD1 aberrantly colocalized to the ARE-associated protein complex (Lu et al. 2007). The ARE can positively regulate mRNA stability and translation, often in response to stressors such as hypoxia and cytokine exposure, through interaction with RNA stabilizers that bind specifically to the ARE (Ross 1995; Brennan and Steitz 2001; Barreau et al. 2006). The best-characterized RNA stabilizers are members of the embryonic lethal abnormal visual (ELAV) family which include the ubiquitous HuR and neuronal-specific Hel-N1, HuC, and HuD (Brennan and Steitz 2001). Here, we show that mutant SOD1 acquires a high binding affinity for VEGF and two other class II AREs, interleukin (IL)-8 and tumor necrosis factor (TNF)-α, in a pattern similar to HuR. We further show that mutant SOD1 can directly compete with HuR and HuC for VEGF mRNA binding. Truncation analysis of mutant SOD1 indicated that RNA-binding activity resides in the N-terminal half of the protein. Taken together, these findings suggest that post-transcriptional dysregulation of VEGF mRNA may stem from direct binding of mutant SOD1.

Materials and methods

Plasmids, Purification of recombinant SOD1 proteins

SOD1 cDNAs (wild-type, A4V, G93A, L38V, and G37R) sequences were amplified by standard PCR and were inserted into the BamHI/XhoI sites of pGEX-5X-1 (GE Healthcare, Piscataway, NJ, USA) to generate glutathione S-transferase (GST)-SOD1 fusion proteins. All primers are summarized in the Table 1. Truncated peptides were generated by the same strategy, using G37R, G93A, or wild-type SOD1 cDNA as PCR templates. All PCR products were verified by sequencing. To make GST-SOD1 fusion proteins, all constructs were transformed into the BL21 strain of Escherichia coli (Stratagene, La Jolla, CA, USA). After induction with 1 mM Isopropyl-β-D-thiogalactopyranoside bacterial lysates were prepared in B-Per buffer (Pierce, Rockford, IL, USA) and then sonicated. GST-SOD1 fusion proteins were isolated by glutathione sepharose 4B and eluted by glutathione elution buffer (GE Healthcare). Glutathione was removed by extensive dialysis in protein storage buffer (100 mM KCl, 1 mM MgCl₂, 5% sucrose, and 40 mM Tris, pH 7.5). Metallation of GST-SOD1 fusion proteins was based on a protocol described elsewhere (Ezzi et al. 2007). Briefly, proteins were incubated overnight with 0.2 mM ZnCl₂ followed by 3 h incubation with 0.2 mM CuCl₂. Proteins were then dialyzed in protein storage buffer overnight. Metallated wild-type (WT) and G93A SOD1 proteins without the GST tag were kindly provided by Dr J. Julien (Laval University, QC, Canada) (Ezzi et al. 2007). Biosynthesis and purification of HuC and HuR proteins, and TIA-1 related protein (TIAR) are described elsewhere (King et al. 1999; Nabors et al. 2001; Suswam et al. 2005). Concentrations of proteins were determined using the bicinchoninic acid protein assay kit (Pierce). For Coomassie blue staining, equimolar amounts of each protein were resolved by 12% sodium dodecyl sulfate-polyacrylamide gel electrophoresis. Gels were stained in GelCode blue (Pierce) for 1 h and then destained in water for 2 h. For Western blot analysis of SOD1 proteins, 300 ng of each protein were resolved by sodium dodecyl sulfate-polyacrylamide gel electrophoresis. The blot was probed with an anti-SOD1 antibody (Santa Cruz Biotechnology, Santa Cruz, CA, USA) at a dilution of 1 : 500 in 5% milk.

Table 1.

Primer sequences for generating constructs and probes

Construct/probe	Primer sequence
SOD1 (1-153)	Fwd: 5′-ATATGGATCCCCGCGACGAAGGCCGTGT-3′
	Rev: 5′-ATATCTCGAGTTATTGGGCGATCCCAATT-3′
A4V (1-153)	Fwd: 5′-ATATGGATCCCCGCGACGAAGGTCGTGTGCGTGCTGAAGGGCGA-3′ a
WT (1-31)	Rev: 5′-ATATATCTCGAGTTACACCTTCACTGGTCCATT-3′b
G37R (1-49)	Rev 5′-ATATATCTCGAGTTACTCATGAACATGGAATCCA-3′b
G37R/WT (1-85)	Rev: 5′-ATATATCTCGAGTTAGCCCAAGTCTCCAACAT-3′b
G37R/WT (32-98)	Fwd: 5′-ATATATGGATCCCCGTGTGGGGAAGCATTAAA-3′
	Rev: 5′-ATATATCTCGAGTTAAGACACATCGGCCACAC-3′
G93A (49-153)	Fwd: 5′-ATATATGGATCCCCGAGTTTGGAGATAATACAGC-3′a
WT (102-153)	Fwd 5′-ATATATGGATCCCCTCTGTGATCTCACTCTCA-3′a
VEGF FL	Rev: 5′-AGATCAGAATTAAATTCTT-3′
VEGF T1	Rev: 5′-AAAATAAATATGTACTACGGAATATCTCGAAA-3′
VEGF T2	Rev: 5′-AATATCTTTTCCCCACAATTATTACGGATAAAC-3′
VEGF T3	Fwd 5′-GAATCCTAATACGACTCACTATAGGGAGCATCACGTCTTTGTCTCT-3′
	Rev: 5′-GTTGTTTAAAAATAAATATGTAC-3′
VEGF T4	Fwd: 5′-GAATCCTAATACGACTCACTATAGGGAGGACAAAGAAATACAGATA-3′
	Rev: 5′-AGATCAGAATTAAATTCTT-3′

SOD, copper/zinc superoxide dismutase; WT, wild-type; VEGF, vascular endothelial growth factor.

^aReverse primer same as SOD1 (1-153) Rev

^bForward primer same as SOD1(1-153) Fwd.

RNA probe synthesis

The VEGF 3′-UTR probe template was derived from the VEGF regulatory segment-3′ and radiolabeled as previously described using T7 RNA polymerase (Levy et al. 1998; King 2000; Nabors et al. 2001). Truncated probe templates (T1-4) were generated by PCR from the original clone (King 2000). IL-8 and TNF-α 3′-UTR templates and synthesis of riboprobes are described elsewhere (Nabors et al. 2001, 2003). The control RNA oligonucleotide was derived from the TNF-α 3′-UTR: 5′-CGGAGCCCAGCCCUCCCCAUGGAGCCAGCUCCCUCU-3′ (Sigma-Proligo, The Woodlands, TX, USA). The oligonucleotide was radiolabeled as previously described (Suswam et al. 2005).

RNA-binding assays

Equimolar amounts of protein were incubated with RNA probes (2.5 fmol/reaction) in binding buffer (10 mM HEPES, pH 7.5, 3 mM MgCl₂, 14 mM KCl, 1 mM dithiothreitol (DTT), 5% glycerol, and 0.2% nonidet P-40 (Sigma-Aldrich, St Louis, MO, USA) at 25 °C. Low molecular weight heparin (5 mg/mL) and yeast tRNA (0.2 μg/mL) were added to block non-specific binding. The total reaction volume was 20 μL. After 20 min, the reaction was digested with 50 U of Rnase T1 (Ambion, Austin, TX, USA) for 15 min at 37 °C. Samples were separated on a pre-electrophoresed 6% native polyacrylamide gel. Gels were dried and analyzed by phosphorimaging (GE Healthcare, Piscataway, NJ, USA). For cold probe competition, labeled VEGF probe (2.5 fmol) was incubated with SOD1 or GST-HuR proteins in the presence of excess molar amounts of unlabeled VEGF probe or the following ribopolymers: poly(A), poly(U), poly(AU), poly(C), or poly(GC) (Pharmacia, Piscataway, NJ, USA). For binding affinity experiments, a limiting concentration of labeled VEGF probe was used (0.25 fmol per reaction). Band densities of free and bound probe were quantitated with PDQuest software (Bio-Rad, Hercules, CA, USA). Binding curves were calculated using GraphPad Prism Software v. 5.0 (GraphPad Software, San Diego, CA, USA). Apparent K_d values were estimated from the curves and represented the protein concentration at which 50% of the RNA probe was bound. For the ELISA assay, synthesis of biotinylated VEGF RNA probe and RNA binding were carried out as previously described (King 2000), with the following binding buffer: 10 mM HEPES, pH 7.5, 3 mM MgCl₂, 14 mM KCl, 1 mM DTT, 5% glycerol, 0.2% nonidet P-40, and 0.2 μg/mL tRNA. A Dunnett’s multiple comparison test was used for statistical comparisons between WT and mutant proteins.

Results

SOD1 binds to the VEGF 3′-UTR in a pattern similar to HuR

To assess RNA binding, we performed an electrophoretic mobility shift assay (EMSA) with recombinant G93A and WT SOD1 using a probe from the VEGF regulatory segment within the 3′-UTR (Fig. 1a) (Levy et al. 1998; King 2000). This segment contains AREs that modulate VEGF mRNA stability and up-regulate expression in response to a number of stressors such as hypoxia, cytokine stimulation, and glucose deprivation (Levy et al. 1996, 1998; Nabors et al. 2003; Yun et al. 2005). The SOD1 proteins lacked any fusion tag, were metallated, and immunoreactive with an anti-SOD1 antibody by Western blot (Fig. 1b). Using equimolar amounts of protein, a weak shift was observed with WT SOD1 whereas two prominent shifts were detected with G93A (Fig. 1c). The upper shift was similar in mobility to WT SOD1, but the other shift migrated substantially faster in the gel. HuR, which is known to bind to AREs in the VEGF 3′-UTR (Levy et al. 1998; Nabors et al. 2001, 2003) produced three band shifts, two of which were similar to SOD1. The shifts for SOD1 and HuR were competed away with the addition of unlabeled VEGF probe. A control RNA probe lacking AU-rich sequence failed to compete away the shifts for mutant SOD1. To determine whether mutant SOD1 could bind to different loci within the VEGF 3′-UTR, we produced four truncated probes, T1-4, and tested them by EMSA (Fig. 1d). Each probe contained one or more AREs. T1 and T2 both recapitulated the binding pattern of full-length probe whereas T3 and T4 truncations produced only the upper shift. Thus, the first ARE, which is common to T1 and T2, likely produced the lower shift. This region has more extensive AU-rich sequence and tandem AUUUA pentamers. The overall pattern suggested that G93A can bind to multiple discrete ARE loci within the 3′-UTR. As AREs are not unique to VEGF mRNA, we next determined whether SOD1 could bind other ARE-bearing 3′-UTRs (Fig. 2). We tested IL-8 and TNF-α which, like VEGF, contain class II AREs (Chen and Shyu 1995). These cytokine mRNAs are also highly regulated post-transcriptionally but were not adversely affected by mutant SOD1 in cultured cells (Hoffmann et al. 2002; Newbury 2006; Lu et al. 2007). Probes used in the assay targeted the cluster of AREs in each 3′-UTR (upper panels) (Nabors et al. 2001; Suswam et al. 2008). IL-8 had a higher AU-rich content than VEGF although the AUUUA pentamers were similar. TNF-α, on the other hand, had lower AU-rich content but a larger cluster of pentamers. We used TIAR, another ARE RNA-binding protein (RBP), GST, and HuR as controls. Similar to the VEGF EMSA, G93A SOD1 produced a prominent shift with IL-8 and TNF-α 3′-UTRs in contrast to a very weak shift with WT protein. The lower shift seen with VEGF probe was not observed. HuR and TIAR produced strong shifts for both riboprobes whereas GST showed no binding. To begin characterizing binding preferences of G93A SOD1, we performed EMSA with the VEGF 3′-UTR probe in the presence of various homoribopolymers (Fig. 3). HuR was used for comparison. Poly(A), (U), and (AU) effectively abrogated the upper band shift for G93A SOD1 whereas poly(C) did not. HuR, which has a known affinity for A-, U-, and AU-rich sequences (Ma et al. 1996, 1997), demonstrated a similar pattern of competition with the ribopolymers. Interestingly, the lower shifts for G93A SOD1 and HuR were competed by poly(U), but not poly(A) or (AU), at low levels of competitor. At high levels of competitor, the lower shift was competed by poly(C) for both proteins. Thus, G93A SOD1 and HuR demonstrated very similar binding properties in terms of RNA targets (AREs) and sequence preferences.

Fig. 1.

SOD1 binds to the AU-rich elements of VEGF 3′-UTR. (a) Schematic showing region of the VEGF 3′-UTR analyzed. Probes are shown below with nucleotide limits shown in parentheses. FL, full-length and T, truncation. (b) Western blot of WT and G93A SOD1 proteins probed with an anti-SOD1 antibody. (c) EMSA with SOD1 and HuR proteins using FL probe with and without an excess of unlabeled VEGF 3′-UTR probe or control (Ctl) RNA. (d) EMSA of FL and truncated probes with G93A SOD1. EMSA experiments were repeated at least three times with similar results; -, probe alone; +, probe and protein; *, free probe.

Fig. 2.

G93A SOD1 binds to other class II AREs. Upper panels show schematic of riboprobes derived from IL-8 and TNF-α 3′-UTRs. Lower panels show EMSA gels with these probes using different recombinant proteins. Assays were repeated five times with similar results. P, probe only.

Fig. 3.

Ribopolymer competition for VEGF 3′-UTR binding. Recombinant G93A SOD1 and HuR were incubated with 2.5 fmol of labeled VEGF 3′-UTR in the presence of excess ribopolymers as shown. The reactions were then analyzed by native gel electrophoresis. Results are representative of three independent experiments; P, probe only; *, free probe.

Other mutant SOD1 proteins bind to VEGF 3′-UTR

We next sought to determine whether other mutant forms of SOD1 have binding characteristics similar to G93A. We cloned three additional ALS-associated mutants, A4V, G37R, and L38V, and expressed them as GST fusion proteins. For consistency of comparison, we expressed WT and G93A SOD1 in the same manner. A Coomassie-stained gel and Western blot, probed with an anti-SOD1 antibody, show that the proteins are comparable (Fig. 4a). Using the same EMSA assay with VEGF 3′-UTR probe, we observed a similar shift pattern among mutant proteins in the metallated state (Fig. 4b, M+ lanes). A4V and G37R SOD1 showed more intense shifts compared with L38V and G93A. Lower shifts were observed with the former two mutants whereas fainter shifts were seen in the latter two. WT SOD1 again showed only a faint upper shift. Comparison of densitometry for the lower shift region showed that WT had values similar to GST control (data not shown). The patterns of binding were not only similar to untagged SOD1 proteins produced elsewhere (see Materials and methods and Figs 1 and 2), but also with different preparations of tagged proteins (data not shown). Interestingly, when the proteins were not metallated, there was an inversion of shift intensities, with the lower one becoming more pronounced than the higher one (Fig. 4b, M- lanes). We also tested TNF-α and IL-8 3′-UTR probes and observed a similar gradient of gel shift intensities with mutant and WT SOD1 (Fig. 4c). In contrast to G93A in Fig. 2, which lacked a GST tag, we observed very faint lower bands in the metallated mutant proteins. These lower bands may reflect subtle effects of the tag on RNA-protein complex binding or migration through the gel. They intensified somewhat in the unmetallated state, but remained substantially fainter than the lower shift for VEGF indicating a consistent difference among probes as with Figs 1 and 2. No lower shift was seen with WT SOD1, either in the metallated or unmetallated state. We then tested each protein with the ribopolymers to determine whether binding preferences among the mutant SOD1 forms. We found a pattern of competition for all proteins that was similar to G93A and HuR (G37R is shown in Fig. 4d; see Fig. S1 for other mutants), with poly(U) and (AU) competing effectively for RNA binding at low concentrations. Again, the lower shift remained with high concentrations of poly(AU). Poly(C) had no effect on the upper shift pattern or intensity, but competed away the lower band at high levels. We also evaluated poly(GC), and this ribopolymer had no effect on either shift. Thus, three different ALS-associated SOD1 mutants bound to VEGF 3′-UTR, similar to G93A, but distinct from WT SOD1.

Fig. 4.

Other ALS-associated SOD1 mutants bind to VEGF, IL-8, and TNF-α 3′-UTRs. (a) The upper panel is a Coomassie-stained gel showing WT and mutant SOD1 proteins tagged to GST. The lower panel is a Western blot of the same proteins using an anti-SOD1 antibody. (b) EMSA analysis of mutant and WT SOD1 proteins with the VEGF 3′-UTR riboprobe; P, probe only; M+, metallated; M-, unmetallated. (c) Same as (b) except TNF-α and IL-8 riboprobes were used. (d) Ribopolymer competition of G37R SOD1 for VEGF 3′-UTR binding as described in Fig. 3. Panels (b-d) are representative of three to seven independent experiments.

Mutant SOD1 proteins display different affinities for VEGF 3′-UTR

The variation in band shift intensities observed with different SOD1 forms suggested differences in affinity for the VEGF 3′-UTR. To address this possibility, we performed EMSA over a range of protein concentrations with limiting amounts of riboprobe (Fig. 5). We observed a marked separation of curves between the group of mutant SOD1 proteins and WT SOD1. The affinity curve of WT SOD1 did not plateau even at higher concentrations, and thus the apparent K_d is likely higher than shown. Even with an estimated K_d of 5 μM, mutant proteins had an 11- to 72-fold higher binding affinity. The affinity of A4V (70 nM), approached that of HuR (30 nM). As the affinity of HuR was somewhat lower than reported elsewhere for other mRNA targets (< 10 nM), it is possible that small amounts of Rnase in the binding assay led to an underestimation of affinity (Ma et al. 1996; Fialcowitz-White et al. 2007). Among the group of mutant SOD1 proteins, there was a hierarchy of affinities with A4V showing the lowest K_d and G93A the highest (476 nM). At substantially higher amounts of protein (760 nM and greater), a lower shift faintly appeared for WT SOD1, in contrast to the mutants where the lower shift was prominent at very low concentrations. The appearance and intensity of the lower band seemed to parallel the hierarchy of affinities. A4V and G37R showed a second shift at concentrations as low as 47.5 nM versus G93A which had a fainter second shift appearing at 285 nM. L38V resided in between G93A and the other two mutants.

Fig. 5.

Comparison of binding affinities among SOD1 and HuR proteins. With limiting quantities of VEGF 3′-UTR probe (0.25 fmol), each protein was analyzed by EMSA over a range of concentrations as shown. Bound and free probe bands were quantitated by densitometry and expressed as a ratio of bound/(bound + free). Curves were calculated as described in Materials and methods. Apparent K_d are shown under each curve. Data points are the mean ± SEM of three independent experiments.

Mutant SOD1 can compete with HuR for VEGF 3′-UTR binding

Based on our finding that mutant SOD1 binds to AU-rich sequences and that the apparent affinity approached that of HuR, we wanted to determine whether the two proteins could compete for VEGF 3′-UTR binding. We used an ELISA-based RNA-binding assay as previously described (King 2000; Nabors et al. 2003). In this assay, HuR is affixed to the ELISA well and the RNA-binding reaction is initiated with addition of biotinylated VEGF 3′-UTR riboprobe and competitor protein. Probe bound by competitor protein is unavailable for HuR binding and subsequently removed during the wash phase. The well is developed colorimetrically using streptavidin conjugated with alkaline phosphatase. All comparisons are represented as a percentage of the signal detected when no competitor is added. As shown in Fig. 6a, there was a concentration-dependent inhibition of HuR binding to VEGF 3′-UTR in the presence of mutant SOD1. A4V showed significant inhibition at 20-fold excess (p < 0.001) whereas G37R, L38V, and G93A required 40-fold excess or greater for inhibition (p < 0.001). WT SOD1, on the other hand, did not show any inhibition. GST, a second control, also did not inhibit binding (not shown). As VEGF mRNA is expressed in motor neurons, we next examined HuC to determine whether binding competition occurred with neuronal-specific ELAV members. This family member has been linked to motor neuron development (Akamatsu et al. 1999). We observed a similar pattern of competition although there was more potent inhibition at 20× for all mutant proteins. At that concentration, A4V and L38V showed the most significant inhibition compared with WT SOD1 (p < 0.001) followed by G37R (p < 0.01) and G93A (p < 0.05). At higher concentrations, the inhibition was similar (p < 0.001). Thus, the pattern of competition, in general, paralleled the relative affinities of these proteins for VEGF 3′-UTR.

Fig. 6.

Mutant SOD1 competes with HuR for binding to VEGF 3′-UTR probe. An ELISA-based RNA-binding assay was used to assess competition between SOD1 and HuR (top panel) or HuC (bottom panel) (see Materials and methods and King 2000). With HuR or HuC affixed to the ELISA well, the binding reaction was initiated by addition of a biotinylated VEGF 3′-UTR riboprobe and SOD1 protein. VEGF 3′-UTR binding was assessed by colorimetry. Results were expressed as a percentage of the value when no competitor was added. All data represent the mean ± SD of four independent measurements. HuR: ***p < 0.001; HuC: ***p < 0.001; **p < 0.01 and *p < 0.05. At 20× for HuC, ***A4V and L38V, **G37R, and *G93A.

Localization of RNA-binding sites of SOD1

As SOD1 has no identifiable RNA-binding domain, we tested truncated peptides by EMSA to localize further the RNA-binding activity to VEGF, TNF-α, and IL-8 AREs. A schematic of SOD1 shows the location of dimer sites, metal-binding sites, and peptides analyzed for RNA binding (Fig. 7a) (Deng et al. 1993). We used G37R, G93A, and WT SOD1 as templates and all peptides were expressed as fusions to GST (Fig. 7b). The EMSA pattern for all three probes seen with full-length G37R was replicated in peptide A* which incorporates the initial 85 amino acids plus the mutant arginine residue (Fig. 7c). Interestingly, when that residue was replaced with WT glycine, two similar but less intense shifts were observed (peptide A). Peptide B, which spans amino acids 1-31 showed little or no shift for VEGF and faint upper shifts for IL-8 and TNF-α (lane 4). An additional 18 amino acids (peptide C, 1-49), however, produced dual shifts similar to peptide A. The next peptide which spanned amino acids 32-98, including mutant arginine at position 37, also produced lower shifts but fainter upper shifts (peptide D*). Replacement of the mutant arginine with WT glycine (peptide D) led to some diminishment of the shifts for VEGF and loss of the lower shift for IL-8 and TNF-α. Peptides representative of the C-terminal portion of the protein showed gradual loss of binding. Peptide E (amino acids 49-153, including the G93A mutation) showed only a weak upper shift but no lower shift, and peptide F (amino acids 102-153) showed no binding. In summary, the N-terminal portion of the protein that included amino acids 32-49 reproduced shift patterns similar to full-length protein.

Fig. 7.

Localization of RNA-binding activity in SOD1. (a) Schematic of human SOD1 and truncated peptides used in the analysis of RNA binding (Deng et al. 1993). Numbers in parentheses refer to the amino acids included in each truncation. ▲, dimer contact sites; *, metal binding sites; S, residue involved in disulfide bridge formation. Peptides included mutant or WT sequence as indicated in parentheses. (b) Coomassie stain of peptides fused to GST. (c) EMSA analysis of peptides using VEGF, IL-8, and TNF-α 3′-UTR riboprobes as indicated. This assay was repeated three times with similar results.

Discussion

In this study, we show that ALS-associated mutations of SOD1 confer a high binding affinity for VEGF (and other) AREs in a pattern similar to HuR. The RNA-binding activity is localized to the N-terminal portion of the protein and is qualitatively affected by the state of metallation. This gain of function enables mutant SOD1 to compete with HuR and neuronal HuC in vitro. These findings provide a plausible explanation for the prior observation that mutant, but not WT, SOD1 was present in the ribonucleoprotein complex of VEGF 3′-UTR (Lu et al. 2007). This acquired RNA-binding activity may represent a toxic gain of function for mutant SOD1 that leads to disruption of post-transcriptional VEGF mRNA processing.

HuR is a ubiquitously expressed RBP that functions as a major stabilizer of mRNAs containing AREs in the 3′-UTR (Brennan and Steitz 2001). HuR belongs to the ELAV family where three of the four members (Hel-N1, HuC, and HuD) are limited in expression to neurons (Szabo et al. 1991; King et al. 1994). The role of HuR in VEGF mRNA stabilization has been characterized in a number of cell systems (Levy et al. 1998; Goldberg-Cohen et al. 2002; Nabors et al. 2003; Datta et al. 2005; Cherradi et al. 2006; Suswam et al. 2008). Additionally, HuR plays a direct role in translation of ARE-containing mRNAs through an association with polysomes (Mazan-Mamczarz et al. 2003; Lal et al. 2004; Kawai et al. 2006; Galban et al. 2008). Neuronal Hu members have also been linked to ARE-mediated stabilization with other mRNAs such as Nova, glucose transporter, and growth-associated protein-43 (Jain et al. 1997; Mobarak et al. 2000; Pascale et al. 2005; Ratti et al. 2008). In contrast to the stabilizing function of Hu proteins, other RBPs such as tristetraprolin, A + U-rich binding factor 1, or K homology-type splicing regulatory protein vie for binding to the same AREs and destabilize the mRNA (Barreau et al. 2006). Our ELISA assay (Fig. 6) raises the possibility that mutant SOD1 antagonizes the stabilizing function of HuR by competing for ARE binding. The potent pattern of binding inhibition with HuC suggests that neuronal ELAV members may also be adversely affected by mutant SOD1. While HuC has not been directly tested for stabilizing effects on VEGF mRNA, it is heavily expressed in ventral spinal cord and is important for motor neuron development (Okano and Darnell 1997; Akamatsu et al. 1999). The impact of altering HuR expression levels on RNA stabilization and translation has been well established (Barreau et al. 2006). In glioma cells, for example, ectopic expression of HuR led to stabilization and up-regulation of VEGF mRNA (Nabors et al. 2003). Stabilization of VEGF mRNA was impaired, on the other hand, when a truncated version of HuR was expressed in vivo (Datta et al. 2005). The truncated form lacked the first RNA recognition motif but retained RNA-binding capability. Our data suggest that mutant SOD1 may function as a dominant negative inhibitor of HuR or HuC by competing for the same binding sites. The abundance of mutant SOD1 in the cytoplasm compared with HuR and other neuronal ELAV proteins which are predominantly nuclear would favor this competition. The excess amount of protein required, however, may still exceed physiological concentrations as the affinity of HuR was 2- to 16-fold higher than the mutant SOD1 proteins. Interestingly, A4V which is associated with a more virulent phenotype (mean survival of 1 year after onset) (Juneja et al. 1997) had the highest affinity for VEGF ARE and displayed the most potent competition with HuR and HuC. As our ELISA assay does not reconstitute the intracellular milieu, it is unclear what other in vivo conditions may exist that affect binding affinity. The presence of cofactors or concentration gradients of ELAV proteins within the cytosol, for example, may favor mutant SOD1 binding. Toxicity may also be related to harmful protein-protein interactions that are created by the proximity of mutant SOD1, vis-à-vis the ARE, with ELAV proteins or other essential components of the RNA-stabilizing/translational machinery. This mechanism, for example, has been implicated for Bcl-2 where toxic interactions of mutant SOD1 may disrupt function of this anti-apoptotic factor (Pasinelli et al. 2004). In the spinal cord, VEGF is produced by a number of sources including astrocytes, motor neurons, and microglia (Bartholdi et al. 1997; Oosthuyse et al. 2001; Brockington et al. 2006), and thus its neuroprotective effect on motor neurons is from paracrine and autocrine pathways. Impairment of HuR and neuronal ELAV proteins thus could negatively impact all sources of VEGF and produce motor neuron degeneration. Indeed, we previously observed a global decline of VEGF mRNA in the spinals cords of mutant SOD1 mice starting before onset of clinical symptoms (Lu et al. 2007). While VEGF mRNA has been our focus, impairment of HuR function could potentially impact a number of different targets. Silencing of HuR, for example, significantly impaired translation of hypoxia-inducible factor alpha-1α, a protein which positively regulates VEGF transcription (Brockington et al. 2004; Mukhopadhyay and Datta 2004; Galban et al. 2008). Likwise, p53 and cytochrome c were significantly down-regulated with loss of HuR (Galban et al. 2003; Kawai et al. 2006). Numerous other targets for HuR have also been identified (Barreau et al. 2006). Our finding that mutant SOD1 bound to two other class II AREs, IL-8, and TNF-α in a similar pattern however, suggests additional determinants of post-transcriptional dysregulation as the half-lives of these mRNAs were not adversely affected by G93A SOD1 in the glioma cell (Lu et al. 2007). Furthermore, Strong and colleagues observed SOD1 binding to low molecular weight neurofilament 3′-UTR in cultured cells (Ge et al. 2005). This 3′-UTR has only short and scattered U-rich sequences. They found that mutant forms of SOD1 bound to the 3′-UTR and produced a destabilizing effect.

So how do the ALS-associated mutations confer high-affinity RNA-binding activity to SOD1? It has become increasingly clear that these mutations lead to a gain of function through improper protein folding (Cleveland and Rothstein 2001; Valentine and Hart 2003; Pasinelli and Brown 2006). The conversion to a high affinity RBP may relate to these conformational changes. SOD1 is typically a stable dimer that maintains enzymatic activity even when exposed to denaturing agents such as urea or ionic detergents (Forman and Fridovich 1973). Mutations associated with ALS, however, disrupt the thermodynamic stability of SOD1 by weakening the dimer interface, destabilizing the monomers, or both (Lindberg et al. 2005; Khare and Dokholyan 2006). Misfolding results in exposure of normally buried hydrophobic residues in the dimer (Tiwari et al. 2005). Unique epitopes produced by these conformational changes have led to the development of antibodies that can distinguish misfolded mutant SOD1 from the native dimer (Rishi et al. 2007). Some of the exposed hydrophobic residues, such as phenylalanine, could potentially interact with RNA ligands through ring-stacking similar to HuR and other RBPs (Kenan et al. 1991; Burd and Dreyfuss 1994; Wang and Tanaka Hall 2001). All of the phenylalanine residues, for example, reside in the N-terminal region and were contained within the G37R truncation (amino acids 1-85) that reproduced the EMSA pattern of full-length protein (Fig. 7). Moreover, amino acids 32-49, which contain six hydrophobic residues found in ribonucleoprotein consensus motifs of RBPs, were required to produce shifts similar to full-length protein (Kenan et al. 1991). The pattern was observed for all 3′-UTR probes suggesting a common mode of binding. On the other hand, the C-terminal portion (amino acids 102-153) of SOD1, which did not bind to the any of the 3′-UTR probes, contains no phenylalanines or tyrosines for ring stacking. The very weak affinity of WT SOD1 (and peptide A in Fig. 7) for AREs indicates that the enzyme has some intrinsic RNA-binding activity. It is unclear whether this is related to the dynamic structure of the dimer (Khare and Dokholyan 2006) or the monomer itself. ARE-binding has been observed in other enzymes that lack canonical RNA-binding motifs such as glyceraldehydes-3-phosphate dehydrogenase, 3-oxoacyl-CoA thiolase, AUH (AU enoyl-CoA-hydratase binding protein) and hydroxyacyl-CoA dehydrogenase/3-ketoacyl-CoA thiolase/enoyl-CoA-hydratase β subunit (Nanbu et al. 1993; Nagy and Rigby 1995; Nakagawa et al. 1995; Adams et al. 2003).

In addition to the weak affinity of WT SOD1 for VEGF ARE, there was also a qualitative difference, with only a single shift observed (Figs 1 and 4). The dual shift observed with mutant forms suggests additional conformational changes that alter electrophoretic mobility of the RNA-protein complex. This observation is supported by accentuation of this altered shift pattern in the unmetallated (apo) state (Fig. 4b and c). It has long been recognized that native SOD1 is inherently less stable in the apo state (Forman and Fridovich 1973; Roe et al. 1988) and this instability may be increased in the presence of ALS-associated mutations (Lindberg et al. 2002; Rodriguez et al. 2002; Tiwari and Hayward 2003; Doucette et al. 2004; Banci et al. 2007). These mutations are thought to shift the equilibrium to abnormally structured apo-monomers which oligomerize and ultimately form pathogenic intracellular aggregates (Lindberg et al. 2002, 2005; Khare et al. 2004, 2006; Banci et al. 2007). In another study, a faster migrating electrophoretic band was observed with mutant SOD1 proteins in a partially denaturing gel, especially under reducing conditions or when the mutant had reduced metal binding capacity (Tiwari and Hayward 2003). The lower band suggested monomerization, and raises the possibility that the lower band in our gel-shift assay represented the monomer form. Although each of the mutant SOD1 proteins analyzed in this study had relatively preserved metal-binding capacity, the lack of metallation after biosynthesis may have sufficiently destabilized the dimer (Borchelt et al. 1994; Hayward et al. 2002; Rodriguez et al. 2002). Furthermore, the RNA-binding reaction was carried out in the presence of mild reducing conditions (1 mM DTT). The lack of a lower shift for WT SOD1 would be supportive of this possibility. The lower shift was consistently more pronounced with VEGF than IL-8 or TNF-α regardless of metallation status. The explanation for this difference is not yet clear, but may relate to the distribution of AREs, the composition of the ARE, or unique flanking sequences (Figs 1 and 2). All of these factors influence secondary or tertiary structure of the 3′-UTR, and thus may impact binding by mutant SOD1. The relevance of these in vitro binding differences to in vivo biological effects is unclear. Although there is a possibility that the patterns could tie in to the differential effects of mutant SOD1 on these mRNAs in cultured cells (Lu et al. 2007), further studies will be required.

In summary, we have shown that ALS-associated mutations of SOD1 confer high affinity RNA-binding activity to elements that govern the stability and translation of VEGF mRNA. This gain of function may disrupt ELAV protein function or other post-transcriptional regulatory components, leading to loss of expression of a growth factor that is essential for motor neuron survival.

Abbreviations used

ALS

amyotrophic lateral sclerosis

ARE

adenine and uridine-rich element

DTT

dithiothreitol

ELAV

embryonic lethal abnormal visual

EMSA

electrophoretic mobility shift assay

GST

glutathione S-transferase

IL

interleukin

RBP

RNA-binding protein

SOD

copper/zinc superoxide dismutase

TIAR

TIA-1 related protein

TNF-α

tumor necrosis factor-α

UTR

untranslated region

VEGF

vascular endothelial growth factor

WT

wild-type.