A small-molecule inhibitor of SOD1-Derlin-1 interaction ameliorates pathology in an ALS mouse model

Amyotrophic lateral sclerosis (ALS) is a devastating neurodegenerative disorder. Despite its severity, there are no effective treatments because of the complexity of its pathogenesis. As one of the underlying mechanisms of Cu, Zn superoxide dismutase (SOD1) gene mutation-induced ALS, SOD1 mutants (SOD1mut) commonly interact with an endoplasmic reticulum-resident membrane protein Derlin-1, triggering motoneuron death. However, the importance of SOD1-Derlin-1 interaction in in vitro human model and in vivo mouse model remains to be elucidated. Here, we identify small-molecular-weight compounds that inhibit the SOD1-Derlin-1 interaction by screening approximately 160,000 compounds. The inhibitor prevents 122 types of SOD1mut from interacting with Derlin-1, and significantly ameliorates the ALS pathology both in motoneurons derived from patient induced pluripotent stem cells and in model mice. Our data suggest that the SOD1-Derlin-1 interaction contributes to the pathogenesis of ALS and is a promising drug target for ALS treatment.

A myotrophic lateral sclerosis (ALS) is a progressive, lateonset neurodegenerative disorder characterized by the selective loss of both upper and lower motoneurons 1 . Riluzole and Edaravone are the only two available treatments for ALS that provide a modest improvement in survival [2][3][4] . Although some clinical trials have been conducted over the past 20 years, most of them have failed, at least in part because of the poor understanding of the pathogenesis of ALS. Multiple motoneurons toxicities have been proposed, such as oxidative stress, excitotoxicity, proteasome dysfunction, endoplasmic reticulum (ER) stress, abnormal mitochondrial function, and altered axonal transport, but no consensus or precise mechanisms have been elucidated 5 . Thus, there remains strong demand for finding new and effective ALS treatments based on the molecular mechanisms of ALS pathogenesis.
Mutations in the Cu, Zn superoxide dismutase (SOD1) gene are frequently found in familial ALS (FALS) patients 6 . Although SOD1 is a well-known antioxidant enzyme, mutant SOD1 (SOD1 mut )-mediated FALS is now considered to result from gain-of-toxic function(s) of SOD1 by mutation rather than from changes in superoxide dismutase activity [7][8][9][10][11][12] . To date, more than 170 different mutations have been reported, but only a few reports have tested the commonality of toxic mechanism(s) by using a substantial number of different ALS-related SOD1 mut . Therefore, further investigation of SOD1 mut -induced toxic mechanism(s) is required for managing this fatal neurodegenerative disease.
We have previously reported that over 100 types of ALS-related SOD1 mut but not wild-type SOD1 (SOD1 WT ) commonly expose the Derlin-1-binding region (termed the DBR) and interact via the DBR with Derlin-1, one of the critical components of the ERassociated degradation (ERAD) machinery 13,14 . Moreover, disruption of the SOD1 mut -Derlin-1 interaction by expressing Derlin-1-derived peptide (termed Derlin-1 (CT4)), which corresponds to the binding site for the DBR and competes for binding to SOD1 mut , has been shown to be protective against SOD1 mutdependent motoneuron loss in the mouse primary spinal cord culture 13 . These findings suggest that the SOD1 mut -Derlin-1 interaction is a common feature of SOD1 mut -caused FALS pathology and that inhibition of this interaction may be a potential target for ALS treatment.
Protein-protein interaction (PPI) is an attractive target for drug development because it plays central roles in large parts of biological processes, and in pathological conditions. Thus, small molecules that inhibit PPIs have potential as therapeutic medicines themselves and also as useful experimental tools to elucidate pathological mechanisms. Although it is considered as challenging to identify small molecules that modulate PPIs, several reports provide encouraging evidence for finding such compounds [15][16][17][18][19][20][21][22][23][24][25] .
Here, we developed a high-throughput screening (HTS) system to find small-molecular-weight compounds that would function as SOD1 mut -Derlin-1 interaction inhibitors. By utilizing timeresolved fluorescence resonance energy transfer (TR-FRET) technology, we successfully established a high-throughput, robust assay system to measure the SOD1-Derlin-1 interaction. We performed an HTS of 160,000 compounds in the public chemical library (at the Drug Discovery Initiative (DDI), The University of Tokyo) and found that an analog of one of the hit compounds prevent 122 types of SOD1 mut from interacting with Derlin-1. The inhibitor significantly ameliorates the ALS pathology in motoneurons derived from patient induced pluripotent stem cells (iPSCs) and delays the onset and prolongs the survival (14.5% and 14.2% improvement, respectively) of ALS model mice expressing SOD1 G93A . Our data emphasize the importance of the SOD1-Derlin-1 interaction as a common mechanism of motoneuron toxicity in SOD1 mut , and we provided a potential mechanism-based ALS treatment.
We screened a total of approximately 160,000 small molecules using the inhibition of FRET signal as a readout. The Z′-factor of all plates in the first screening was set to be over 0.75, suggesting that this assay system was sufficiently robust and reliable (Supplementary Figure 1a). In this stage, 1460 compounds that showed over 15% inhibition were selected for further validation in the second screening ( Fig. 1d and Supplementary Figure 1b). The decrease in FRET signal may occur not only because of the inhibition of SOD1-Derlin-1 interaction but also because of the inhibition of interactions between the tags and the corresponding antibodies. Alternatively, compounds may also target the energy transfer itself by interacting with the emission or excitation processes. To exclude such false-positive compounds, we prepared the monomolecular FRET assay using HA-SOD1 WT -Flag with Flag-Eu and HA-d2 as a counter assay (Supplementary Figure 1c). In the second screening, we reassessed the inhibition of the SOD1-Derlin-1-derived FRET signal of 1460 positive compounds in the first screening and simultaneously checked the effect of each compound on the monomolecular FRET signal of HA-SOD1 WT -Flag. Many of the compounds that were reconfirmed to inhibit the SOD1-Derlin-1-derived FRET signal also inhibited the monomolecular FRET signal (Fig. 1e, lower panel, region A). However, some compounds attenuated only the SOD1-Derlin-1-derived FRET signal with little effect on monomolecular FRET, suggesting that these compounds inhibit the SOD1-Derlin-1 interaction (Fig. 1e, lower panel, region B). Therefore, 131 compounds were selected, each showing over 15% inhibition of the SOD1-Derlin-1-derived FRET signal without inhibiting the monomolecular FRET signal by more than 15% (Fig. 1e, lower panel, region B and Supplementary Figure 1b). Using a titration assay as the third screening, 44 compounds that inhibited the SOD1-Derlin-1-derived FRET signal in a concentration-dependent manner were further selected as positive compounds (Supplementary Figure 1b, d).
In vitro compound validation identified #56 analogs. In the FRET assay, positive compounds may still include compounds that alter only the conformation of SOD1 or Derlin-1 and thereby disrupt FRET signal without dissociation between the two proteins. Thus, we performed an in vitro co-immunoprecipitation assay by adding the compounds to the purified Flag-SOD1 G93A -Derlin-1-HA complex. Twelve compounds out of 44 became positive in this assay (Supplementary Figure 2a-e). Among them, we focused on one of the most prominent inhibitors, named compound #56, because of its dose-dependency in the FRET screening and its drug-like chemical structure defined according to the Lipinski's rule of five, the absence of reactive functional groups, and the exclusion of promiscuous hitters 26,27 . The easiness to speculate the pharmacophore and to synthesize the analogs was also taken into consideration (Fig. 1f, g and Supplementary Figure 2d). Structural analogs of compound #56 were commercially available. More than 200 of those compounds were examined for potential inhibitory effects on SOD1-Derlin-1 interaction. Some of the #56 analogs that showed over 15% inhibition of the SOD1-Derlin-1-derived FRET signal also inhibited the interaction between SOD1 and Derlin-1 in in vitro immunoprecipitation assay (Fig. 2a). Moreover, we found inhibitors that showed lower half-maximal inhibitory concentration (IC 50 ) values than #56 itself (Fig. 2b), including #56-20 and #56-26 (Figs. 2c, d, respectively). These data suggest that the basic core structure of #56 analogs has an inhibitory effect on the SOD1-Derlin-1 interaction.
#56-59 disrupts all SOD1 mut -Derlin-1 interactions analyzed. Unfortunately, none of the #56 analogs tested above showed inhibitory activities in the cell-based co-immunoprecipitation assay. To evaluate the possibility that the undesired interaction with serum-derived substances was the cause of the inactivation, we assessed the activity in serum-starved condition. Within the serum-depleted medium, some of the compounds showed the inhibition activity, suggesting that the undesired interaction in the culture medium was one of the causes of the inactivity (Supplementary Figure 3a). In addition, #56-26 was relatively stable and showed low permeability, suggesting that the inability might also be due to the low permeability of the plasma membrane (Supplementary Figure 3b).
Although there were some compounds that inhibited the SOD1-Derlin-1 interaction in serum-depleted condition, we have previously reported that the serum depletion causes the conformational change in SOD1 14,28 . Thus, we synthesized new #56 derivatives to obtain novel inhibitors that are active in the  culture medium containing serum. In an in vitro immunoprecipitation assay, two (#56-40, #56-41) out of six new compounds worked as well as #56 itself, correlating with the result of FRETbased assay (Fig. 3a, b and Supplementary Figure 3c). Moreover, #56-40 showed an improved permeability and inhibitory effect on the SOD1-Derlin-1 interaction even in a cell-based co-immunoprecipitation assay ( Fig. 3c and Supplementary Figure 3b). However, this compound has the inconsistent effect, suggesting that the effective concentration depends on cellular conditions. Therefore, we synthesized additional compounds and found a  more stable inhibitor in cell-based assay, #56-59 ( Fig. 3d-f).
Through the further evaluation of newly synthesized #56 derivatives, we found that the character of potent inhibitors in cell-based assay is provided by the replacement of hydrogen with alkyl groups at the amide bond, turning that amide into a tertiary amide (Fig. 3g, h). Testing several derivatives with a tertiary amide, we found even more potent inhibitor, #56-111 (Supplementary Figure 4a, b). SOD1 G93A , which we used for the above screening steps, is a representative SOD1 mut among 122 types of ALS-related SOD1 mut that interact with Derlin-1. Thus, we next examined whether #56-59 could also inhibit the interactions of Derlin-1 with 122 types of SOD1 mut in the cell-based co-immunoprecipitation assay. #56-59 clearly attenuated the interactions between Derlin-1 and all tested SOD1 mut (Supplementary Figure 5a-h). These results suggest that ALS-related SOD1 mut commonly possess a similar mechanism of interaction with Derlin-1, consistent with our previous findings of the importance of DBR exposure in SOD1 for the interaction with Derlin-1 14,28,29 . SOD1 WT is known to form homodimers, and Derlin-1 forms homomeric and heteromeric complexes with Derlin family proteins and with other ERAD components such as Hrd1 30 . Under the same conditions as those in which #56-59 inhibited the SOD1-Derlin-1 interaction, these interactions were not affected, suggesting that #56-59 inhibits the SOD1-Derlin-1 interaction with some specificity (Supplementary Figure 6).
#56 analogs target SOD1 DBR. There are two putative modes of action of PPI inhibitors: intervention by direct binding to the interface or allosteric interference. To reveal the mechanism of #56-mediated PPI inhibition, we evaluated the inhibition activity of #56 analogs against the SOD1-Derlin-1 interaction using fulllength proteins and/or fragments containing binding sites in the in vitro assay. Due to the low activity of #56-59 in vitro, two prominent in vitro inhibitors, #56-20 and #56-26, were assessed. Both of them inhibited all combinations of the interaction (Fig. 4a). The inhibition of the interaction between SOD1 (1-20) and Derlin-1 (CT4) suggested that #56 analogs exert their activities by interacting with the binding region of either SOD1 or Derlin-1. To identify the direct binding target of #56 analogs, we tried the surface plasmon resonance (SPR) method with biotinylated peptide as a ligand and #56 analogs as an analyte. However, the interaction could not be detected probably due to the relatively weak interaction and low solubility of the compounds (Supplementary Figure 7a). Thus, we next performed fluorescence polarization (FP) assays in the presence of SOD1 (5-20) peptide or Derlin-1 (CT4) peptide, utilizing the fluorescence properties of #56-20 per se. The degree of FP increased in the presence of SOD1 (5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16)(17)(18)(19)(20) peptide rather than the Derlin-1 (CT4) peptide, suggesting that #56-20 preferentially and directly interacts with the SOD1 DBR (Fig. 4b). We also compared the binding ability of #56-20 to the full-length recombinant SOD1 WT and SOD1 G93A . The preferential interaction of #56-20 with SOD1 G93A was observed (K d = 81.1 μM and Supplementary  Figure 7b). This is consistent with our model in which SOD1 mut possess a DBR-exposed conformation and that #56-20 interacts with the DBR 14,29 .
Next, to confirm the target of the compounds within the cells, we embedded benzophenone as a photoactivatable crosslinking moiety, and a biotin as a tag to #56-59, a potent inhibitor in cellbased assay. We named this compound Photo-Biotin-PEG 3 -#56-59, PB56 for short (Fig. 4c, right). Although it was less potent than the original compound (#56-59), PB56 also inhibited the SOD1-Derlin-1 interaction in a dose-dependent manner (Supplementary Figure 7c). In the crosslinking and pull-down assays using SOD1 G93A -transfected or Derlin-1-transfected cells treated with PB56, we clearly observed the biotin-labeled band at the same position of monomer Flag-SOD1 G93A and high-molecular smear bands probably representing sodium dodecyl sulfate (SDS)-resistant SOD1 G93A aggregates, suggesting that PB56 selectively binds to SOD1 and not Derlin-1 (Fig. 4c, left). Pre-treatment with #56-59 competitively displaced the binding of PB56 to SOD1 G93A , indicating that the common basic structure of #56-59 in PB56 mediated the interaction with SOD1 (Fig. 4d). Moreover, we confirmed that PB56 interacted with SOD1 G93A more strongly than SOD1 WT within the cells (Fig. 4e).
Additionally, we evaluated the effects of #56 analogs to SOD activity in vitro. Both #56-20 and #56-26, which possess strong inhibitory activity on the SOD1-Derlin-1 interaction in vitro, showed no effects on SOD activity (Supplementary Figure 7d). Taken together, these data suggest that #56 analogs might specifically inhibit the interaction with Derlin-1 by directly binding to the SOD1 DBR without affecting the enzymatic activity of SOD1.
#56-59 ameliorates the pathology of ALS. Finally, we evaluated the therapeutic effects of #56 analogs in in vitro human model utilizing ALS patient iPSCs as well as in in vivo ALS mouse model. Recently, we have characterized and established the in vitro ALS model system by utilizing patient iPSCs 31 . Therefore, we treated #56-40 and #56-59 to motoneurons derived from ALS patient iPSCs with L144FVX mutation in SOD1 (ALS1). Although #56-40 was less effective, probably because of the narrow effective dose, treatment with #56-40 and #56-59 increased the survival of ALS motoneurons (Fig. 5a). Then, we further investigated the effect of #56-59 to other iPSC-derived motoneurons. The number of motoneurons generated from healthy control-derived iPSC was not increased in the presence of #56-59, indicating that the increase of ALS motoneurons with #56-59 was not due to the improvement of differentiation efficiency or survival independent of ALS pathology (Supplementary Figure 8a). Although #56-59 did not improve the viability of motoneurons generated from ALS3 (SOD1 G93S ), it restored the motoneuron survival of ALS2 (SOD1 L144FVX ) (Supplementary Figure 8a). These data suggested that #56-59 could ameliorate SOD1 mut toxicity in in vitro human ALS model. The conformational changes in SOD1 WT have been reported in FALS with other ALS-causative gene mutations, including FUS and TDP-43 32 . Thus, we examined the effect of #56-59 on motoneuron toxicity with TDP-43 mutant; however, #56-59 failed to improve the survival of ALS4 motoneurons (TDP-43 M337V ) at least in this condition (Supplementary Figure 8a).
Next, we performed continuous delivery of the compounds to SOD1 G93A transgenic male mice by using osmotic pumps. Because we were concerned that the efficacy of the compounds might be compromised by a limited ability to cross the blood-brain barrier and to access the target motoneurons, intracerebroventricular (i.c.v.) cannulation was chosen as the method for delivery. The start point of administration was set at 22 weeks of age, approximately 6 weeks before the usual onset timing as defined in our previous study 13 . We infused the mice with dimethyl sulfoxide (DMSO) as control, 1 mM #56-40, and   Table 2). Consistent with these results, the number of motoneurons detected by Nissl staining of lumbar spinal cord sections at 31 weeks of age was significantly increased in #56-59-treated ALS model mice (Fig. 5c, d). These data clearly show that the SOD1-Derlin-1 interaction inhibitor can ameliorate ALS pathology both in in vitro human model and in vivo mouse model, demonstrating the importance of the SOD1-Derlin-1 interaction in the pathogenesis of SOD1 mut -induced FALS and the potential of the SOD1-Derlin-1 interaction as a therapeutic target in ALS.

Discussion
In the present study, we designed and developed a highthroughput, robust screening assay system for measuring the interaction between two proteins, SOD1 and Derlin-1 (Fig. 1a-c).
We screened approximately 160,000 compounds and selected one potential scaffold, #56 (Fig. 1d-g). We found that some analogs of #56 also possessed inhibitory activities in vitro (Fig. 2a-c). Moreover, newly synthesized #56 analogs inhibited the SOD1-Derlin-1 interaction in cell-based assays (Fig. 3c, d, g). One of these inhibitors, #56-59, exerted its activity on all types of SOD1 mut -Derlin-1 interaction that we previously reported 14 ( Supplementary Figure 5a-h). Furthermore, we show that the SOD1-Derlin-1 interaction inhibitor can ameliorate ALS pathology both in in vitro human model and in vivo mouse model (Fig. 5). We used two inhibitors, #56-40 and #56-59, to assess the effect to the ALS pathology. However, unlike #56-59, #56-40 showed only modest effects to the ALS pathology (Fig. 5a, b). Our concern was that the effective concentration of #56-40 was in a very narrow range. Thus, we assume that the dose of #56-40 might be insufficient to show a therapeutic effect on ALS model mice under these conditions. Moreover, ALS4 motoneurons showed even a vulnerability to #56-59. The ALS motoneurons would be feeble, and the effective dose of #56-59 might be different among iPSC lines. The failure of improvement in ALS3 and the significant reduction in ALS4 could be caused by the toxicity of #56-59.
The detected concentration of #56-59 in the brain and spinal cord of the mice were very low (Supplementary Figure 8b and  Supplementary Table 2). In addition, we could not evaluate the inhibition activity in vivo, because the level of the SOD1-Derlin-1 interaction varies even in the non-treated ALS model mice (Supplementary Figure 9a). Taking account of the unbound fractions within the serum, the effective concentration of #56-59 could be estimated between 0.25 to 1 μM in cell-based immunoprecipitation assay ( Supplementary Figure 4a and Supplementary Figure 8c). One of the possible reasons of this discrepancy is the modification of #56-59. #56-40 showed a relatively unstable character (Supplementary Figure 3b). Although HEK293A cells and the cells in central nervous system do not express typical metabolizing enzymes compared to hepatic cells, #56-40 and probably #56-59 could be metabolized in cells and in vivo. In the present method that we measured the concentration of #56-59, the modified compounds could not be estimated. Therefore, it is possible that the modified #56-59, which we could not identify in this study, contributed to the inhibition of SOD1-Derlin-1 interaction in the present assays.
Another possibility is the difference of the treatment time course. In the long-term treatment in vivo, #56-59 could occupy the DBR of newly synthesized SOD1 mut before its interaction with Derlin-1. Thus, #56-59 would be able to efficiently inhibit the SOD1-Derlin-1 interaction compared to the cell-based immunoprecipitation assay, where #56-59 might dissociate the pre-formed complexes. It is also possible that as we could find the increase of #56-59 in spinal cord at 5 weeks of infusion (Supplementary Figure 8b and Supplementary Table 2), the longer continuous infusion would reach higher concentration.
Notably, one ALS model mouse treated with #56-59 did not show any symptoms until 58 weeks of age. Although it is unclear whether #56-59 could really suppress the pathology completely in some population or whether additional factor(s) might exist, further improvement of the administration method, including doses, timing of initial administration, and pharmacokinetics, should provide us with more effective treatment for ALS pathologies.
In the present study, we administered the compounds before the appearance of defects in motor function. Our compound delayed the onset of the disease, suggesting that the SOD1-Derlin-1 interaction is involved in the onset of ALS, and earlier treatment might improve the phenotype more effectively. The early diagnosis of conformation-disordered SOD1 using our ELISA system that can easily detect the DBR-exposed SOD1 may enable us to perform a differential diagnosis of the applicable patients before onset and to provide pre-onset treatment 14,26 .
We have previously reported that ER stress evoked by the SOD1-Derlin-1 interaction activates ASK1, and ASK1 deficiency prolonged the survival of ALS model mice by improving the disease progression but not onset 13 . Moreover, the administration of ASK1 inhibitors showed the reduction of the glial cell activity 33 , which is known to contribute to the progression of the pathology 34,35 . On the other hand, the #56-59 treatment had little effect on the progression and did not show significant suppression of gliosis (Supplementary Figure 9b). Thus, we estimate that the SOD1-Derlin-1 interaction may primarily target the signaling pathways that contribute to the disease onset independent of ASK1 pathway. It would be an important issue to reveal the detailed signaling pathways affected in the present treatment protocol and the role of ASK1 in non-cell autonomous motoneuron death, as well as the effects of #56-59 on the ASK1 pathway in in vitro and in vivo.
#56 analogs preferentially interact with SOD1 mut rather than SOD1 WT . Although we cannot exclude the possibility that #56 analogs inhibit the interaction of SOD1 mut through the DBR not only with Derlin-1 but also with other proteins, the SOD1-Derlin-1 interaction is the most likely and promising target for ALS treatment. It is also important to carefully interpret the specificity of #56 analogs in cells or in vivo because of the very weak K d value of them (Supplementary Figure 7a). Further improvements of #56 analogs' affinity to SOD1 mut would be required. The elucidation of compound-SOD1 complex atomic structure should resolve these questions and facilitate further optimization of the compounds.
We also found that zinc deficiency caused a conformational change of SOD1 WT to a mutant-like form (exposure of the DBR), resulting in SOD1-Derlin-1 interaction 28 . Recently, it was reported that conformationally disordered SOD1 WT was observed in SOD1 mutation-negative sporadic ALS (SALS) patients 36 . SOD1 WT has also been shown to be required for non-cell autonomous motoneuron toxicity in astrocytes or oligodendrocytes derived from SALS patients 37,38 . These data suggest that under certain conditions, including genetic factors and environmental factors, SOD1 WT may also serve as the cause of ALS pathogenesis and that inhibition of the SOD1-Derlin-1 interaction may be a potential target for ALS therapy development not only for SOD1 mut -caused FALS but also for a subset of SOD1 mutation-negative FALS and SALS, in which conformationally disordered SOD1 WT is proposed to be involved. To approach this possibility, we investigated the effect of #56-59 on the iPSCderived motoneurons with TDP-43 mutant. However, consistent with our previous observation that the exogenously expressed TDP-43 mutants could not alter the SOD1 WT conformation 29 , administration of #56-59 did not improve motoneuron viability (Supplementary Figure 8a). Since we cannot exclude the possibility that the vulnerability masked the positive effect of #56-59 in ALS4, it is important to evaluate the involvement of the SOD1-Derlin-1 interaction by using multiple inhibitors, different iPSC-derived motoneurons, and/or other methods.
In summary, we succeeded in providing a proof-of-concept regarding our SOD1-Derlin-1 interaction hypothesis by ameliorating the disease both in in vitro human model and in vivo mouse model by using a developed PPI inhibitor. These data reemphasize the importance of the SOD1-Derlin-1 interaction as a common mechanism of motoneuron toxicity in SOD1 mut , and we provided a potential approach for a molecular mechanism-based ALS treatment.
according to the manufacturer's instructions (Dojindo). The experiment was repeated three times on three separate days.
Animals. All experiments were performed in accordance with protocols approved by the Animal Research Committee of the Graduate School of Pharmaceutical Sciences, The University of Tokyo (Tokyo, Japan). The mouse genotypes were verified by PCR at weaning and before the administration of compounds. At 22 weeks of age, approximately 6 weeks before the usual onset timing as defined in our previous study 13 , male mice with the human SOD1 G93A transgene (G1L/+ line, backcrossed to C57BL/6) were randomly separated into three groups and continuously infused with DMSO as control, with 1 mM #56-40, or with 3 mM #56-59, using osmotic pumps (flow rate 0.15 μl h −1 , Alzet Mode 2006) by i.c.v. cannulation with the Brain Infusion Kit 3 (Alzet). Pumps were placed on the back of the mouse and were replaced every 6 weeks until the mouse showed paralysis. Onset was determined at the time point when the average of week's rotarod performance (three times per week) began to decline at an accelerated speed to 40 rpm for 5 min. The genotype of the mouse that did not show any symptoms was again confirmed by PCR, and the treatment trial of this mouse was stopped at 58 weeks of age. The investigators were not blinded to allocation during experiments and outcome assessment.
Measurement of #56-59 concentration. Male mice (C57BL/6) were continuously infused with 3 mM #56-59 using osmotic pumps by i.c.v. cannulation. The brain and spinal cord were extirpated at 1, 3, and 5 weeks after the start of the infusion and homogenized with phosphate buffer. The tissue homogenates were deproteinized with acetonitrile containing the internal standard (methyltestosterone) and centrifuged at 10,000 × g for 5 min at 4°C. The supernatants were subjected to liquid chromatography-tandem triple quadrupole mass spectrometry analysis (LCMS-8060, Shimadzu, Japan) to quantify the concentration of #56-59 in the brain and spinal cord. The samples that were below the detection limit (Supplementary Table 2) were eliminated in the Supplementary Figure 8b.
Nissl staining and immunohistochemistry. Mice were perfused with PBS and then with 4% paraformaldehyde (PFA) in PBS and the spinal cords (L2-L5) were fixed with 4% PFA in PBS for 1 day. The lumbar spinal cords were then incubated in 30% sucrose for 24 h, embedded in CryoMount I (Muto PureChemicals), and frozen sections (40 μm) were produced. Nissl stained were performed using 0.1% cresyl violet acetate (WALDECK). Motoneurons showing a clear nucleolus and distinctly labeled cytoplasm with neuronal morphology, with cell bodies >20 μm in diameter and located in the anterior horn, were included in cell counts. All motoneuron counts were performed in a blinded fashion. For the immunohistochemistry, frozen sections were blocked with 1% BSA at room temperature, and each section was incubated with the primary antibodies. After three washes with PBS, sections were incubated with the secondary antibody solution at room temperature. All images were obtained using a Leica TCS SP5 confocal laser scanning microscope.
Statistical analysis. No statistical methods were used to predetermine sample size. Motoneurons survival assay were analyzed using one-way analysis of variance followed by Tukey's post hoc test (with similar variance) or Kruskal-Wallis followed by Dunn's test (#56-59-treated ALS1 with non-similar variance) to determine statistical significances of the data (n = 6; p < 0.05, post hoc test, *p < 0.05). Onset and survival were statistically analyzed using the Kaplan-Meier method followed by the log-rank test and the Gehan-Breslow-Wilcoxon test (*p < 0.05). To ensure randomization of spinal cord section for motoneurons count, the five frozen sections of L2 to L5 were randomly chosen (WT mice (34 weeks old): n = 4; DMSO-treated and #56-59-treated SOD1 G93A transgenic mice (31 weeks old): n = 8 per group, *p < 0.05: the significance with similar variance was evaluated using unpaired, two-tailed Student's t test).
Data availability. The data that support the findings of this study are available from the corresponding author on reasonable request.