Interest in RNA dysfunction in amyotrophic lateral sclerosis (ALS) recently aroused upon discovering causative mutations in RNA‐binding protein genes. Here, we show that extensive down‐regulation of miRNA levels is a common molecular denominator for multiple forms of human ALS. We further demonstrate that pathogenic ALS‐causing mutations are sufficient to inhibit miRNA biogenesis at the Dicing step. Abnormalities of the stress response are involved in the pathogenesis of neurodegeneration, including ALS. Accordingly, we describe a novel mechanism for modulating microRNA biogenesis under stress, involving stress granule formation and re‐organization of DICER and AGO2 protein interactions with their partners. In line with this observation, enhancing DICER activity by a small molecule, enoxacin, is beneficial for neuromuscular function in two independent ALS mouse models. Characterizing miRNA biogenesis downstream of the stress response ties seemingly disparate pathways in neurodegeneration and further suggests that DICER and miRNAs affect neuronal integrity and are possible therapeutic targets.
Amyotrophic Lateral Sclerosis‐causing mutations affect DICER complex activity and lead to altered miRNA biogenesis.
Reduction in miRNA levels unifies genetically unrelated forms of human ALS.
miRNA biogenesis is attenuated by cellular stress or by ALS proteins, TDP‐43, FUS or SOD1.
Stress granule protein interactions with DICER complex underlie reduced activity.
Enoxacin, a DICER agonist, mitigates miRNA biogenesis defect and neuromuscular deterioration in models of ALS.
Amyotrophic lateral sclerosis (ALS) is a neurodegenerative disease of the human motor neuron system. Mutations in genes encoding RNA‐binding proteins were recently identified as causative in human ALS. These include TAR DNA‐binding protein 43 (TDP‐43) (Kabashi et al, 2008; Sreedharan et al, 2008; Liu‐Yesucevitz et al, 2010; McDonald et al, 2011), fused in sarcoma (FUS) (Kwiatkowski et al, 2009; Vance et al, 2009; Bosco et al, 2010; Gal et al, 2010) and heterogeneous nuclear ribonucleoprotein A1 (hnRNPA1) (Kim et al, 2013).
Stress granules (SGs) are cytoplasmic sites for modulating mRNA translation and form in response to cellular stress (Anderson & Kedersha, 2008, 2009; Buchan & Parker, 2009). Intriguingly, among the RNA‐binding proteins that are recruited into SGs, several are mutated in ALS or the related disorder fronto‐temporal dementia (FTD), for example, TDP‐43 and FUS.
Stress and SGs are thought to be involved in the pathogenesis of ALS, and SGs are observed in pathologic ALS/FTD specimen (Neumann et al, 2007; Volkening et al, 2009; Bosco et al, 2010; Dormann et al, 2010; Gal et al, 2010; Liu‐Yesucevitz et al, 2010; Dewey et al, 2011; McDonald et al, 2011; Meyerowitz et al, 2011; Aulas et al, 2012; Baron et al, 2013; Daigle et al, 2013; Kim et al, 2013; Vance et al, 2013; reviewed in Dewey et al, 2012; Emde & Hornstein, 2014; Li et al, 2013b; Ling et al, 2013). SG formation is initiated by the phosphorylation of eukaryotic translation initiation factor 2 alpha (EIF2A) on serine‐51 or by the expression of several aggregation‐prone proteins that are capable of initiating SG nucleation (Kedersha et al, 1999). Accordingly, modulation of EIF2A phosphorylation mitigates neurotoxicity in animal models of ALS (Saxena et al, 2009; Kim et al, 2014).
microRNAs (miRNAs) are small endogenous RNAs. Initial processing of primary miRNAs by Drosha/Dgcr8 (Gregory et al, 2004) yields an intermediate pre‐miRNA that is further processed by the 500‐kDa DICER multi‐protein complex to form a mature miRNA. This complex contains DICER 1, ribonuclease type III (DICER) and its partners, Argonaute RISC catalytic component 2 (AGO2), TAR (HIV‐1) RNA‐binding protein 2 (also known as TRBP) and protein kinase, interferon‐inducible double‐stranded RNA‐dependent activator (PACT) (Chendrimada et al, 2005; Haase et al, 2005; Lee et al, 2006). DICER activity is controlled by its core partners (Ma et al, 2008; Shan et al, 2008; Li et al, 2012; Lee et al, 2013) and the mature miRNA is loaded onto AGO2, creating a functional RNA‐induced silencing complex (RISC).
DICER and its co‐factors participate at multiple levels of the stress response (Emde & Hornstein, 2014). Thus, DICER contributes to stress resistance in selected tissues and its deficiency reduces stress tolerance (Mori et al, 2012). In addition, various stress‐signaling cascades control AGO2 post‐translational modifications, its translocation into SGs and its activity (Leung et al, 2006; Qi et al, 2008; Zeng et al, 2008; Shen et al, 2013).
We previously demonstrated that loss of miRNA biogenesis is sufficient to cause spinal motor neuron degeneration in vivo (Haramati et al, 2013). In the current work, we demonstrate that reduction in miRNA levels is a common molecular denominator for multiple forms of familial and sporadic human ALS and that enhancement of DICER activity is beneficial in vivo in two independent ALS mouse models. The observations establish a novel pathway downstream of the stress response that controls DICER complex activity. The regulation of DICER involves stress granule assembly and dynamic interactions with SG proteins. This new mechanism for the control of miRNA biogenesis offers novel therapeutic targets for ALS.
miRNAs are down‐regulated in motor neurons of human ALS patients
We recently demonstrated that conditional inactivation of DICER in motor neurons is sufficient to cause spinal motor neuron degeneration in vivo (Haramati et al, 2010). Additional works from other groups further suggest the involvement of miRNAs in ALS pathogenesis (Williams et al, 2009; Buratti et al, 2010; Kawahara & Mieda‐Sato, 2012; Morlando et al, 2012; Droppelmann et al, 2013; Freischmidt et al, 2013; Koval et al, 2013; Zhang et al, 2013). To pursue this, we examined miRNA expression in lumbar motor neurons of sporadic ALS (sALS) spinal cords isolated by laser capture microdissection (LCM). Global down‐regulation of miRNAs was observed in motor neurons. The observation seemed specific to motor neurons since RNA extracted from surrounding, motor neuron‐depleted ventral‐horn tissue of sALS patients or from neurons of Clarke's column (non‐motor neuron) isolated by LCM within the same nervous systems did not exhibit such global changes in miRNA expression (Fig 1A–C). The observations of miRNA down‐regulation were substantiated by miRNA in situ hybridization, which revealed down‐regulation of miR‐9 and miR‐124 in patient tissue, relative to control tissue and to the hybridization signal of U6 RNA (Fig 1D). We next investigated whether miRNAs display a similar profile in fALS cases. Indeed, analyzing two nervous systems carrying familial SOD1 A4V mutation, revealed similar down‐regulation of miRNAs (Fig 1E and F) that correlated with miRNA changes, observed in sALS (Pearson's correlation coefficient = 0.3). Our data suggest that changes in miRNA expression, which were previously reported in ALS spinal cord extracts (Campos‐Melo et al, 2013), are primarily due to changes in miRNA expression in motor neurons. Attempts to measure pre‐miRNAs in human LCM samples were unsuccessful, probably due to the low abundance of these intermediate precursors. Noteworthy, expression and splicing data for changes in mRNAs encoding miRNA biogenesis factors from the relevant ALS patients did not reveal any change, relative to controls (Rabin et al, 2009). Additionally, 27 long ncRNAs tested were similarly expressed in another cohort of 12 non‐ALS and 12 sporadic ALS nervous systems, with Pearson's correlation of 0.9882. Since miRNAs are globally down‐regulated in sALS cases that are genetically unrelated, reduced miRNA expression in motor neurons appears to be a common denominator of various forms of ALS.
Cellular stress affects pre‐miRNA processing
Stress pathways are thought to control miRNA activity (Leung & Sharp, 2010; Mendell & Olson, 2012; Emde & Hornstein, 2014), but the idea that cellular stress might broadly regulate miRNA biogenesis has not been tested. We sought to do so by the measurement of the relative levels of mature miRNAs and pre‐miRNAs under several stress conditions. We further defined an “inhibition score” that approximates a value of one, when DICER processes pre‐miRNAs effectively, as in wild‐type conditions, and increases >1, when miRNA biogenesis is inhibited, relative to control (Fig 2A).
We chose representative miRNAs that are expressed in human and mouse motor neurons including miR‐10b, miR‐30a and miR‐103 (abundantly expressed in isolated mouse motor neurons); miR‐218, miR‐30c, miR‐138 and let‐7b (highly expressed in human motor neurons); miR‐132, let‐7c and let‐7d (neuronal, mid‐range in human); and miR‐143 and let‐7a (low expression in human neurons).
We triggered cellular stress in the hybrid motor neuron cell line NSC‐34; using thapsigargin induced markers of ER stress (Appendix Fig S1A), down‐regulated mature miRNA levels and up‐regulated pre‐miRNA levels, relative to control, demonstrating the inhibition of miRNA biogenesis (Fig 2B–D). In addition, application of oxidative stressors, paraquat or sodium arsenite, led to the inhibition of miRNA biogenesis (Fig 2E–J, Appendix Fig S1B).
To test whether changes in DICER or its co‐factors are down‐regulated as a result of stress, we measured protein levels by Western blot analysis. The expression levels of the four DICER complex proteins were not down‐regulated by two independent stressors (thapsigargin, paraquat). DICER and AGO2 levels were not reduced by sodium arsenite as well, but PACT and TRBP levels were reduced by 1/3 (Appendix Fig S2 related to Fig 2). We conclude that alteration of DICER activity may occur even without reduction in DICER complex components.
Expression of ALS‐causing genes inhibits pre‐miRNA processing
To assess whether ALS‐causing genes affect miRNA biogenesis similar to chemical stressors, we transfected NSC‐34 cells with vectors of wild‐type FUS, TDP‐43 and SOD1 or mutant forms of the same genes, specifically FUS R495X (Bosco et al, 2010; Waibel et al, 2010), TDP‐43 A315T (Gitcho et al, 2008) and SOD1 G93A (Rosen et al, 1993) (Fig 3A). We observed inhibition in biogenesis of many of the miRNA species tested, relative to the corresponding control sample values, after overexpression of wild‐type or mutant FUS and TDP‐43 (Fig 3B–E, Appendix Fig S3). The inhibitory effect of wild‐type SOD1 or G93A mutant on miRNA biogenesis was consistent with this, but more limited (Fig 3F and G). The deleterious effect of higher‐than‐normal levels of wild‐type FUS, TDP‐43 or SOD1 is in accordance with the previous reports (Jaarsma et al, 2000; Wils et al, 2010; Igaz et al, 2011; Mitchell et al, 2013). Western blot analysis revealed that the levels of DICER, AGO2, PACT and TRBP did not change by the overexpression of wild‐type, or mutant forms of FUS (Appendix Fig S4 related to Fig 3).
In order to test DICER activity directly, we devised a cell‐free Dicing assay, using cell lysates and synthetic double‐stranded RNA, carrying a quencher on a 25‐nt passenger strand and a CY3 fluorophore conjugate on a 27‐nt precursor guide strand. Dicing activity enables the dissociation of a fluorophore conjugate RNA from the quencher and fluorescence (Fig 4A). The assay demonstrated reduction in DICER catalytic activity in lysates of cells treated with sodium arsenite, paraquat or thapsigargin (Fig 4B and C). Importantly, overexpression of ALS‐causing mutants or their respective wild‐type forms also reduced DICER activity (Fig 4D). Noteworthy is that the in vitro assay probably provides an underestimate of the changes in DICER activity, due to the extensive dilution of soluble DICER cofactors in the reaction buffers. Therefore, overexpression of ALS genes, mutation in ALS genes and the application of oxidative or ER stress all inhibit DICER catalytic activity.
Enoxacin ameliorates ALS‐induced defects in pre‐miRNA processing
Next, we hypothesized that enhancing Dicing complex activity might reverse the negative effect of the ALS‐causing mutant proteins on miRNA processing. We tested this hypothesis by employing enoxacin, a fluoroquinolone antibiotic that is commonly used for the treatment of urinary tract and airway infections. Enoxacin is known to increase miRNA biogenesis via increasing the binding affinity of TRBP and pre‐miRNAs (Shan et al, 2008; Melo et al, 2011). Intriguingly, the impairments in pre‐miRNA processing that were evident in NSC‐34 cells upon transfection with vectors expressing ALS‐causing mutants FUS R495X, TDP‐43 A315T or SOD1 G93A were partially ameliorated by applying enoxacin (100 μM for 72 h) to transfected cultures (Fig 4E–G and Appendix Fig S5). Therefore, enoxacin studies further support the view that DICER activity is impaired in culture models of ALS.
Stress granule formation controls miRNA biogenesis
To gain further mechanistic insight, we performed targeted mass spectrometry analysis of AGO2 and DICER protein interactions under cellular stress (Fig 5A). We examined a set of proteins that were co‐immunoprecipitated with either AGO2 or DICER in a preliminary unbiased screen. Targeted mass spectrometry revealed enhanced interactions of AGO2 with DICER in cells that were treated with sodium arsenite, relative to basal conditions (Fig 5B and C). We report increased AGO2 binding to DICER, which was previously shown to decrease DICER complex activity (Tahbaz et al, 2004).
As DICER complex composition controls miRNA biogenesis (Lee & Doudna, 2012; Lee et al, 2013), we measured additional AGO2 and DICER interactions. HSP90 and co‐chaperone p23, which regulate the interactions of DICER with AGO2 (Tahbaz et al, 2004; Pare et al, 2013), increased their interactions with DICER under stress (Fig 5B). These interactions potentially stabilize an intermediate DICER–AGO2–P23–HSP90 complex (Pare et al, 2013), thereby inhibiting DICER catalytic activity. In addition, cellular stress increased AGO2 and/or DICER protein interactions with SG components EIF2A, EIF3 (Kedersha et al, 2005) and EIF5A (Li et al, 2010) and with poly(rC)‐binding proteins PCBP1 (hnRNP E1) and PCBP2 (hnRNP E2) (Li et al, 2012) (Fig 5B and C). PCBPs are iron chaperones (Shi et al, 2008; Nandal et al, 2011), found in SGs and P‐bodies (Fujimura et al, 2008, 2009). PCBP2 binds to miRNA precursors and presents them to DICER for efficient processing (Li et al, 2012). Therefore, the binding of PCBP1 and 2 to AGO2 and DICER exposes an additional layer of complexity in the regulation of miRNA biogenesis under stress.
Finally, AGO2 is phosphorylated by the p38 mitogen‐activated protein kinase on serine‐378 under stress (Zeng et al, 2008), consistent with a > 10‐fold increase in AGO2 serine‐387 phosphorylation in our system (Fig 5C). Intriguingly, p38 was reported to be persistently activated in ALS (Tortarolo et al, 2003). We further validated increased interactions of AGO2 with PCBP1, by AGO2 immunoprecipitation and Western blot analysis. DICER pull‐down was marginally significant, and we could not validate the increased interaction of PACT with AGO2 (Fig 5D).
Taken together, these observations point toward substantial changes in the interactions of DICER and AGO2 with SG components. Next, we tested the hypothesis that SG formation may be sufficient to attenuate DICER complex activity. We overexpressed TIAR, a known initiator of stress granule formation (Kedersha et al, 1999), which reduced DICER catalytic activity in a cell‐free assay (Fig 5E). EIF5A is a translation elongation factor, which is engaged with SGs under stress (Li et al, 2010) and also with DICER (Fig 5B). We demonstrated that EIF5A overexpression reduced DICER catalytic activity in a cell‐free assay and in cells (Fig 5E and F and Appendix Fig S6A and B).
Several cellular stress‐signaling cascades converge into the phosphorylation of Serine‐51 of the translation initiation factor 2 alpha (EIF2A), thereby driving SG formation (Hinnebusch, 2005; Kedersha et al, 2005). Because EIF2A interactions with AGO2 and/or DICER increased by stress (Fig 5B and C), we tested the implications of overexpressing a phosphomimetic version of EIF2A (S51D). Expression of EIF2A S51D resulted in inhibition of DICER activity (Fig 5G and Appendix Fig S6C and D), further linking stress signaling to reduced miRNA biogenesis. Accordingly, phosphorylation‐resistant serine‐to‐alanine‐mutated EIF2A (S51A) (Costa‐Mattioli et al, 2007) enabled a partial recovery from thapsigargin‐induced attenuation of DICER complex activity in mouse embryonic fibroblasts (Appendix Fig S7 related to Fig 5).
To test whether the experimental manipulations we used are sufficient to induce visible SGs, we utilized immunofluorescence microscopy. This study revealed the onset of TIA1‐positive stress granules with a set of chemical SG inducers, or after transient transfection with vectors expressing wild‐type or mutant forms of FUS, TDP‐43 or SOD1. It may be that high plasmid copy number and the stress of the transient transfection contributed to the induction of SGs by wild‐type FUS, which was not reported in a stably transfected cell line (Bosco et al, 2010; Dormann et al, 2010; Daigle et al, 2013).
In addition, overexpression of SG proteins that were sufficient to inhibit DICER activity, EIF5A, EIF2A S51D or TIAR led to the formation of TIA1‐positive SGs (Fig 6). Intriguingly, under all these conditions, AGO2 was co‐localized with TIA1 in SGs, in accordance with the previous reports that observed AGO2 recruitment into SGs (Leung et al, 2006; Wu et al, 2011). Therefore, interactions of SG proteins with AGO2 and/or DICER directly contribute to the attenuation of DICER activity.
We next tested the effect of SGs on DICER activity via chemical induction or blockade of SG formation (Fig 7A). NSC‐34 cells were treated with puromycin (1 μg/ml, 24 h), which drives dismantling of translating polysomes, thereby inducing SGs (Blobel & Sabatini, 1971). Puromycin decreased the efficacy of miRNA biogenesis in a cell‐free Dicing assay (Fig 7B) and in cells, as measured by qPCR (Fig 7C and Appendix Fig S8A and B). Notably, puromycin (1 μg/ml, 24 h) reduced DICER and TRBP levels by 44 and 31%, respectively (Appendix Fig S9A and B related to Fig 7). Because the interpretation of the puromycin experiment can take into consideration reduction in DICER/TRBP levels, we sought a complementary approach to test whether inhibition of SG formation recovers miRNA biogenesis from ALS mutant‐induced Dicing defects. Cycloheximide (CHX) is known to dissolve pre‐formed SGs (Kedersha et al, 2000). The application of CHX to cultures transfected with FUS R495X‐, TDP‐43 A315T‐ or SOD1 G93A‐expressing vectors was able to mitigate the reduction in DICER complex activity (Fig 7D–F and Appendix Fig S8C–H) and reduced the abundance of visible SG that were quantified by confocal microscopy (Fig 7G and H). Noteworthy, it is unlikely that the effect of CHX was due to a general inhibition of protein synthesis since puromycin, which equally inhibits translation, had an opposite effect (Fig 7B). Furthermore, with the CHX concentrations we used, we did not see a reduction in the ectopic expression of FLAG‐TDP‐43 A315T or GFP‐FUS R495X and CHX was unable to affect miRNA levels when applied to untransfected NSC‐34 cells (Appendix Fig S10A–C, related to Fig 7).
miRNAs down‐regulation in the SOD1 G93A mouse model of ALS is reversed by enoxacin
We next extended our observations to the SOD1 G93A mouse model of ALS (B6SJL‐Tg (SOD1*G93A)1Gur/J mouse strain) (Gurney et al, 1994). We detected down‐regulation of several miRNAs in SOD1 G93A spinal cords at day 137, relative to the expression of the same miRNAs at day 77. These changes were not observed in age‐matched and litter‐matched wild‐type controls (Appendix Fig S11A related to Fig 8). The mRNA and protein levels of AGO2, DICER, PACT and TRBP remained unchanged (Appendix Fig S11B and C related to Fig 8). The relatively mild reduction may be due to extracting RNA from the whole spinal cord tissue.
Recently, enoxacin was shown to increase the expression of miRNAs in the frontal cortex of rats with beneficial impact on depression (Smalheiser et al, 2014). Therefore, we tested whether miRNA levels are changed in vivo after oral application of enoxacin (n = 5) or carrier (water, n = 5) to SOD1 G93A male sibling‐matched cohorts. Treatment starting on day 42 of the mouse life did not affect miRNA levels in wild‐type motor cortices that were harvested at day 90, relative to controls. However, the levels of several miRNAs were up‐regulated by enoxacin in SOD1 G93A motor cortices (Appendix Fig S11D and E related to Fig 8).
Enoxacin therapy is beneficial for neuromuscular function in the SOD1 G93A mouse model of ALS
We next tested whether enoxacin has a beneficial effect on neuromuscular function of SOD1 G93A mice (Gurney et al, 1994). We observed a ~7‐day delay at the onset of neurological symptoms of the enoxacin‐treated group (n = 40) relative to sibling‐matched male cohort treated with carrier (water, n = 37) (Fig 8A). Consistently, weight peak and onset of weight decline, which is defined by the loss of 1 g bodyweight after the weight peak, were delayed in enoxacin‐treated mice (Fig 8B and C). Kaplan–Meier survival analysis did not yield significant differences between the two groups (Fig 8D). However, evaluation of neurological status by employing a common neurological scoring system, with numerical values increasing from 0 to 4 as the disease progresses (Gill et al, 2009). Neurological scoring revealed that the enoxacin‐treated cohort was superior to untreated controls (Fig 8E).
To further evaluate motor function, we performed automated quantitative gait analysis, using the CatWalk system (Neumann et al, 2009). Swing speed was significantly higher and stride length was significantly increased in the enoxacin‐treated group, relative to untreated controls (ten matched siblings per group, Fig 8F and G). Additionally, the performance of the enoxacin‐treated group was superior to untreated siblings in a rotarod test (n = 25 control, n = 29 enoxacin, Fig 8H). Therefore, enoxacin has a beneficial effect on multiple clinical parameters of the SOD1 G93A mouse model. Furthermore, even a lower enoxacin dose significantly improved gross strength on a hang‐wire assay and the clinical neurological score, relative to controls (16 animals treated with enoxacin, 15 untreated controls; Fig 8I and J). Finally, enoxacin‐treated animals (n = 6) moved significantly more than untreated controls (n = 5) in a fully automated infrared‐based home cage locomotion assay (Fig 8K). Although enoxacin did not extend the lifespan of the SOD1 G93A model, an array of different assays revealed improved neuromuscular function, suggesting a beneficial profile. Noteworthy, riluzole, the only FDA‐approved drug for ALS, lacks survival benefit in SOD1G93A mice (Scott et al, 2008; Li et al, 2013a). Therefore, as previously suggested (Scott et al, 2008), the predictive value of the aggressive SOD1 G93A model may be limited due to its 23 transgene copies in testing novel pharmacological interventions.
We also tested enoxacin in the TDP‐43 A315T transgene (Wegorzewska et al, 2009). Enoxacin‐treated TDP‐43 A315T females displayed an improved neurological score, relative to a control cohort (nine treated animals, eight untreated controls; treatment started on day 42; Fig 8L), performed better on the Hang‐wire test (Fig 8M) and displayed improved stride length (Fig 8N). Of note, characterization of the TDP‐43 A315T mice was performed until the onset of clinical gastro‐intestinal pathology, which leads to early death of this TDP‐43 A315T mouse line and precludes the full development of ALS (Esmaeili et al, 2013).
miRNAs function at an intriguing interface of cellular stress and disease (Leung & Sharp, 2007; Mendell & Olson, 2012; Emde & Hornstein, 2014). Accordingly, facets of miRNA malfunction in ALS or frontotemporal dementia were reported in the past: We demonstrated that loss of DICER is sufficient to cause progressive degeneration of spinal motor neurons (Haramati et al, 2010). In addition, specific miRNA genes are involved in feedback loops upstream or downstream of FUS (Morlando et al, 2012; Dini Modigliani et al, 2014) or TDP‐43 (Buratti et al, 2010; Kawahara & Mieda‐Sato, 2012). miRNAs were suggested to regulate neuromuscular junction repair (Williams et al, 2009; Valdez et al, 2014), to plausibly regulate neuro‐inflammation in ALS (Koval et al, 2013) and may serve as circulating biomarkers (De Felice et al, 2014; reviewed in Droppelmann et al, 2014; Gascon & Gao, 2014).
The current work provides novel evidence that global down‐regulation of miRNAs is a molecular commonality for genetically unrelated forms of human ALS and suggests that similar previous observations (Campos‐Melo et al, 2013) are primarily due to changes in miRNA expression in motor neurons.
The observations in human samples led us to investigate the mechanism. We demonstrated that decreased DICER catalytic activity is reducing miRNA levels. Several lines of evidence suggest that DICER complex activity is inhibited by the emergence of SGs: (i) The interaction of several RNA‐binding proteins that are normally found in stress granules with DICER or AGO2 increases upon stress. (ii) Initiating SGs by overexpressing EIF5A or TIAR, the phosphomimetic form EIF2A, or by puromycin, was sufficient to reduce DICER activity in cells and in cell lysates. (iii) AGO2, a cofactor of DICER, co‐localizes with TIA1 in cytoplasmic SGs after transfection of ALS‐causing genes. (iv) Blocking SGs with cycloheximide or with non‐phosphorylatable serine‐to‐alanine 51 mutant of EIF2A reversed the negative effect of ALS‐causing mutants on miRNA biogenesis. Puromycin and cycloheximide are two inhibitors of translation, yet they display opposite effects both on pre‐miRNA processing and, respectively, on SG formation. Thus, the protective activity of cycloheximide in motor neurons (Yang et al, 2013) is consistent with its ability to inhibit SG formation and to improve miRNA biogenesis. (v) These observations are in line with the previous identification of TDP‐43 and FUS in SGs and with the fact that TDP‐43 binds DICER and AGO2 directly (Kawahara & Mieda‐Sato, 2012), providing a conceivable explanation also for the stronger molecular phenotypes observed with TDP‐43 and FUS, relative to SOD1.
Therefore, we suggest that stress and SG formation initiate dynamic changes in DICER interactions with its co‐factors, in a way that diminishes DICER complex activity. This may be because of acquired or lost interactions under stress and may be related to AGO2 post‐translational modifications (Leung et al, 2006; Qi et al, 2008; Zeng et al, 2008; Shen et al, 2013) or to increased AGO2 binding to DICER, which counter‐intuitively decreases DICER activity (Tahbaz et al, 2004).
Enoxacin increases the binding affinity of TRBP to pre‐miRNAs, thereby augmenting pre‐miRNA processing (Shan et al, 2008; Melo et al, 2011). We demonstrated that impaired miRNA biogenesis is mitigated by enoxacin with beneficial clinical outcome on the neuromuscular performance of two different ALS mouse models. We did not observe any lifespan extension, but also Riluzole, the only FDA‐approved drug for ALS, lacks survival benefit in SOD1 G93A mice (Scott et al, 2008; Li et al, 2013a). Therefore, as previously suggested (Scott et al, 2008), the predictive value of the aggressive SOD1 G93A model may be limited in testing novel pharmacological interventions. Furthermore, SOD1 effect on DICER activity may be more limited than the effects mediated by FUS or TDP‐43, which are RNA‐binding proteins that are directly engaged in SGs.
Nonetheless, if DICER and miRNAs are indeed instrumental in the pathogenesis of ALS in humans, as we suggest, these observations hold promise for future intervention by validating DICER as a new therapeutic target. Furthermore, enoxacin is a particularly intriguing candidate for clinical assessment, based on its established safety and pharmacokinetic profile and because several forms of ALS converge in the down‐regulation of miRNAs. However, more effective DICER agonists might exhibit more dramatic beneficial effects in ALS in the future.
miRNAs impart robustness to cellular programs in development and in the adult life (Hornstein & Shomron, 2006; Pelaez & Carthew, 2012; Cassidy et al, 2013; Emde & Hornstein, 2014). Accordingly, even moderate DICER and miRNA insufficiency contributes to failed homeostatic mechanisms, allowing the activation of myriad aberrant pathways that were otherwise repressed under normal miRNA activity. The consequences may contribute particularly to diminished stress resistance, creating a vicious cycle, which further increases neuronal stress up to the threshold of malfunction and death. Accordingly, in one example, even mild but long‐term downregulation of brain‐enriched miR‐124 affected AMPA expression and frontotemporal dementia phenotype (Gascon et al, 2014).
The sensitivity of DICER activity to cellular stress is not limited to motor neurons and may be probably relevant to other tissues under chronic stress, where decrease in miRNA or DICER abundance was reported (Kaneko et al, 2011; Mori et al, 2012; Inukai & Slack, 2013; Nidadavolu et al, 2013). Thus, evaluating whether stress impinges on DICER activity in other disease states will be valuable. The variability in miRNA biogenesis with different stressors or by using different cell types suggests additional factors that are currently unknown.
In summary, we provide evidence for the control of miRNA biogenesis at the level of DICER complex activity, by several stress‐related cellular changes. The consecutive reduction in miRNA levels unifies various forms of ALS, suggesting that the development of novel therapies might focus on DICER or miRNAs as common critical effectors of neuronal integrity.
Materials and Methods
Human tissue miRNA analysis
Human nervous tissues were acquired by way of an Investigational Review Board and Health Insurance Portability and Accountability Act compliant process. Sporadic ALS nervous systems were from patients who had met El Escorial criteria for definite ALS. Informed consent was obtained from all subjects. Tissue measurements were carried out post‐mortem and conformed to the WMA Declaration of Helsinki and the Department of Health and Human Services Belmont Report. Sporadic ALS nervous systems were from patients who had bulbar or arm‐onset disease and caudal progression, and thus, the lumbar regions had relatively abundant residual motor neurons. Tissue collections were as previously described: Completed within 4–6 h of death, the motor system was dissected, embedded in OCT and stored at −80°C. RNA quality was assessed using microelectrophoresis on an Agilent 2100 Bioanalyzer as previously described. Microdissected cell punches were captured on CapSure™ Macro LCM Caps (Arcturus Bioscience). 35 to 50 cryosections were cut at a thickness of 9 μm in a −18°C cryotome and placed onto uncharged glass slides. The sections were returned to −80°C for a minimum of 3 h. They were stained with cresyl violet acetate in a 10‐step, timed, nuclease‐free immersion process. Motor neurons were microdissected using a Pixcell IIe Laser Capture Microdissection (LCM) System (Arcturus Bioscience) and CapSureTM Macro LCM Caps (Applied Biosystems). Each LCM session collected 500–1,000 motor neuron punches and lasted < 2.5 h. Total RNA was isolated using the RNAqueous Micro kit (Life Technologies) as per the manufacturer's procedure. RNA quality was assessed by spectrophotometric analysis using a Nanodrop ND‐1000 spectrophotometer and an Agilent 2100 Bioanalyzer. Expression of 667 microRNAs was assessed using the microRNA TaqMan® qPCR Megaplex pools array on an ABI 7900HT Fast Real‐Time PCR System (Life Technologies). Technical replicates were run after pre‐amplification using the manufacturer's protocol. Relative quantification was calculated by comparing arithmetic mean of three small RNA control genes nucleolar RNU48/SNORD48, RNU44/SNORD44 and splicosome‐related U6. Two‐way ANOVA test was performed with DataAssist Software (Life technologies Inc.). In situ hybridization was performed on 7‐μm sections of frozen spinal tissue from lumbar regions, briefly fixed in 4% paraformaldehyde and incubated in 2 μg/ml proteinase K. Slides were then incubated for 10 min in 0.13 M 1‐methylimidazole (Sigma), 300 mM NaCl, pH 8.0 and further fixed in EDC (Sigma) following Pena et al (2009), acetylated for 30 min in a solution of freshly prepared 0.1 M triethanolamine and 0.5% (v/v) acetic anhydride. Hybridization of sections with 4 pmol of 5′ and 3′ DIG‐labeled miR‐9 and miR‐124 LNA probes followed manufacturer's instructions (Exiqon).
Tissue culture vectors and small molecules
HEK293 cells (American Type Culture Collection) and NSC‐34 cells (Cellutions Biosystems Inc.) were cultured in high‐glucose DMEM (Gibco), 1% l‐glutamine, 10% fetal calf serum and 1% penicillin/streptavidin. Mouse embryonic fibroblasts (MEFs) were cultured in DMEM (Gibco), 15% fetal calf serum, l‐glutamine, 1% penicillin/streptavidin and sodium pyruvate. AGO2‐FLAG‐expressing HEK293 cells were a gift of Markus Landthaler. EIF2A S51A mouse embryonic fibroblasts were provided by Prof. Chaim Kahana, Weizmann Institute of Science, Israel. Mouse embryonic motor neurons were harvested as described in Milligan and Gifondorwa (2011). Expression vectors are as follows: EIF2A 3 (S51D) from David Ron (Addgene plasmid # 21809); pBabe_puro_DEST_Flag_EIF5A2 from William Hahn (Addgene plasmid # 45266); and pLJM60‐Tiar from David Sabatini (Addgene plasmid # 38244). Wild‐type and R495X FUS vector were provided by Daryl Bosco (University of Massachusetts, Boston), wild‐type and A315T TDP‐43 vectors by Markus Landthaler (MDC Berlin, Germany), and wild‐type and G93A SOD1 vectors by Yoram Groner (WIS, Israel). DICER‐FLAG and AGO2‐FLAG plasmids were kindly provided by Narry Kim (Seoul National University, South Korea). Enoxacin was supplied by Sigma‐Aldrich (E3764‐5G) or Buckton‐Scott (Germany). Thapsigargin, paraquat, cycloheximide and puromycin are purchased from Sigma‐Aldrich, Israel.
RNA was extracted using Tri reagent (Molecular Research Center Inc.) or Qiazol with consecutive use of the miRNeasy Mini Kit (Qiagen). Pre‐miRNA and miRNA reverse transcription was performed using miScript Kits No. I & II (Qiagen). Quantitative analysis of miRNA and pre‐miRNA expression was performed in > 3 independent biological repeats and additionally in technical duplicates with StepOnePlus quantitative Real‐Time PCR System, and two‐way ANOVA test was performed with DataAssist Software (Life technologies Inc.). Primers are described in Table EV1 and additional pre‐miRNA sequences are from Jiang et al (2005).
Cell‐free dicing assay
Dicing assay is described in Melo and Melo (2014) with the following modifications: dsRNA substrate for measuring DICER activity was annealed from 5′Phos/mCmU rCmArU mUrUrC rCrUrG mGrUmA rUmGrA mCrArA rCrGrA mAmU/3′IABkFQ and 5′Cy3/mAmU mUrCrG rUrUrG mUrCmA rUmArC rCrArG rGrArA rArUmG rAmGmC rU (Integrated DNA Technologies). Cells were lysed via sonication in hypotonic solution, with consecutive DICER assay reaction assembly by mixing 3 μl of 10× Buffer K (200 mM Hepes–KOH, pH 7.0, 20 mM dithiothreitol (DTT), 20 mM MgCl2); 3 μl of 10 mM ATP; 1.5 μl of 0.5 M, freshly dissolved, creatine phosphate; 0.9 μl of 1 mg/ml creatine phosphate kinase; 0.6 μl of ribonuclease inhibitor (40 units/μl); 3 μl of labeled dsRNA (50 nM); and 18 μl of cell extract. DICER assay mix was incubated at 30°C for 2 h and diluted with 100 μl of ultra‐pure water, and then, 10 μl was transferred to 384‐multiwell plate (10 technical repeats for each reaction) and measured on PHERAstar FS microplate reader (BMG Labtech). Pre‐annealed ssRNA Cy3 fluorescent oligo served as positive control for calibration curve. Annealed and quenched dsRNA without cell extract served as negative control for quantification of background fluorescence.
Western blot analysis
Cell extracts or immune‐purified proteins were eluted by boiling in X5 sample buffer (60 mM Tris–HCl pH 6.8, 25% glycerol, 2% SDS, 14.4 mM β‐mercaptoethanol, 0.1% bromophenol blue) for 5 min. The samples were separated via 8% SDS–polyacrylamide gel electrophoresis at 120 V for 70 min, electro‐transferred onto nitrocellulose membrane (Whatmann, 10401383) at 100 V for 1 h, stained with Ponceau (Sigma, P7170) to asses transfer quality and then blocked for 1 h at room temperature with 5% milk protein in PBS + 0.05% Tween‐20 (0.05% PBST). Blocked membranes were then incubated with shaking at 4°C overnight with primary antibodies in 5% bovine serum albumin, 0.02% sodium azide and five drops of phenol red in 0.05% PBST. Membranes were next washed 3 times for 5 min at room temperature with 0.05% PBST and then incubated with horseradish peroxidase‐conjugated species‐specific secondary antibodies in 5% milk for 1 h at room temperature. Membranes were washed three times for 5 min in 0.05% PBST and visualized by ImageQuant™ LAS 4000 (GE Healthcare Life Sciences) using EZ‐ECL Chemiluminescence detection kit (Biological Industries, 20‐500‐120). Primary antibodies used are as follows: anti‐DICER (ab13502) 1:500; anti‐AGO2 (ab32381) 1:500; anti‐PACT (ab75749) 1:1,000; anti‐TRBP (ab42018) 1:1,000; anti‐PCBP1 (ab74793) 1:1,000 from Abcam; anti‐GAPDH (Ambion, AM4300) 1:4,000; anti‐phospho‐eIF2α (Ser51) (Cat# 04‐342EMD, Millipore) 1:500; anti‐eIF2α (sc‐11386, Santa Cruz Biotechnology) 1:200; anti‐TLS/FUS (cat. 611385, BD Transduction Laboratories) 1:500; and anti‐TDP‐43 (10782‐2‐AP, Proteintech) 1:1,000.
For mass spectrometry, HEK293 cells that stably express AGO2‐FLAG (n = 3 for each group) or DICER‐FLAG‐transfected cells (48 h post‐transfection, n = 4 for each group) were untreated or stressed with 0.5 mM sodium arsenite for 60 min before harvest. IP was performed with the FLAG‐IPT1 Kit (Sigma‐Aldrich). Negative controls were HEK293 cells not containing a FLAG construct. Eluted proteins and the FLAG peptide were filtered using 3‐kDa molecular weight cutoff spin columns (Amicon, Millipore). Buffer was exchanged to 50 mM ammonium bicarbonate (Sigma‐Aldrich) in the same step. Proteins were reduced by addition of dithiothreitol (Sigma‐Aldrich) to a final concentration of 5 mM and incubation for 30 min at 60°C and alkylated with 10 mM iodoacetamide (Sigma‐Aldrich) in the dark for 30 min at 21°C. The proteins were then digested using trypsin (Promega; Madison, WI, USA) at a ratio of 1:50 (w/w trypsin/protein) for 16 h at 37°C. Digestion was stopped by addition of 1% trifluoroacetic acid (TFA). Samples were stored in −80°C until analysis. Each sample was spiked with the mixture of 51 heavy isotopically labeled peptides (JPT Technologies). Heavy labels included U‐13C6; U‐15N4 for peptides terminating with Arg and U‐13C6; U‐15N2 for Lys. ULC/MS grade solvents were used for all chromatographic steps. Each sample was loaded using split‐less nano‐ultra‐performance liquid chromatography (10k psi nanoAcquity; Waters, Milford, MA, USA). Mobile phase was as follows: A) H2O + 0.1% formic acid and B) acetonitrile + 0.1% formic acid. Desalting of samples was performed online using a reverse‐phase C18 trapping column (180 μm i.d., 20 mm length, 5 μm particle size; Waters). The peptides in samples were separated using a C18 T3 HSS nano‐column (75 μm i.d., 250 mm length, 1.8 μm particle size; Waters) at 0.3 μl/min. Peptides were eluted from the column and into the mass spectrometer using the following gradient: 3–30% B in 90 min, 30–35% in 10 min, 35–90% B in 5 min, maintained at 90% for 5 min and then back to initial conditions. The analytical column was coupled with a quadruple orbitrap mass spectrometer (Q Exactive, Thermo Scientific), via a nano‐ESI interface. Parallel reaction monitoring of mass spectrometry (MS) was according to Peterson et al (2012) at maximum injection time of 300 ms, resolution of 35,000 and target automatic gain control (AGC) of 2e5. Samples were analyzed in a random order, measuring the light and heavy forms of each target peptide. Data were processed with Skyline software (MacLean et al, 2010). Extracted ion chromatograms were integrated, and the ratio of the light to heavy forms was calculated. Student's t‐test was used, after logarithmic transformation, for statistical calculations of the differential peptide expression. Peptide sequences for mass spectrometry are described in Table EV1.
Wild‐type or stably transfected AGO2‐FLAG HEK293 cells were cultured on poly‐A‐lysin‐coated coverslips and further subjected to transfection or treated with stress‐inducing chemicals. For microscopy, cells on coverslips were rinsed with PBS, fixed with 4% formaldehyde, permeabilized with 0.1% Triton X‐100 (Sigma‐Aldrich, St. Louis, MO), blocked with PBS, 2% BSA and incubated with anti‐TIA1 antibody (sc‐1757, Santa Cruz Biotechnology), anti‐Flag antibody (F1804, Sigma‐Aldrich) and anti‐AGO2 (ab32381, Abcam) overnight. Coverslips were washed three times with PBS and incubated with secondary antibodies (Cy2 donkey anti‐goat 705‐495‐147 and Cy5 donkey anti‐mouse 715‐175‐150, Jackson Immunoresearch) for 1 h. Coverslips were washed 3 times with PBS, mounted with Fluoroshield™ with DAPI (F6057, Sigma) and captured with a LSM 710 confocal microscope (Carl Zeiss AG).
In vivo enoxacin study
All experiments were performed according to Weizmann Institute of Science guidelines and IACUC approval (protocol No. 03040512‐1). Mice, B6SJL‐Tg(SOD1*G93A)1Gur/J, which were first reported in Gurney et al (1994), carry a hemizygous high transgene copy number of the SOD1 G93A mutant on a B6SJL hybrid background (Jackson Laboratories, 002726). The TDP‐43 A315T mouse line (Wegorzewska et al, 2009) was on a C57BL/6J background (Jackson Laboratories, 010700). All experiments were performed sibling‐matched, except the mice in the 200 mg/kg bodyweight SOD1 G93A enoxacin group, in which mice were randomly assigned to treatment groups. Animals were group‐housed on a 12‐h light–dark cycle. Food and water were provided ad libitum. Food pellets at cage floor, long sipper tubes on bottles and nutrient gel were used to ease accessibility to nutrition when mice deteriorated according to Gill et al (2009). RotaRod™ performance (San Diego Instruments) was evaluated at ramp speed 4, 40 rpm. Data are presented as the average latency to fall off the rotarod of three independent rounds per time point and are normalized to average initial performance at pre‐symptomatic stage for each individual for each time point. A fully automated gait analysis with the CatWalk XT™ (Noldus) system was performed. Data were obtained of > 4 independent rounds on the apparatus. For hang‐wire muscle strength assessment, mice were allowed forelimb grip onto a 2‐mm‐thick horizontal metal wire, suspended 80 cm above surface. Ability to successfully raise hindlimbs to grip the wire and crawling to the end of the wire was scored: 1 = successful four‐paw grip and crawling to the end of the wire within < 60 s; 2 = four‐paw grip for > 15 s without crawling; 3 = failure to successfully establish hindlimb grip in 60 s; and 4 = inability to sustain forelimb grip for 60 s. Each mouse was evaluated in two sessions on two consecutive days, consisting of three trials each, with a 15‐min inter‐trial interval. The second day was used to obtain the data as displayed. Neurological score was assessed according to Gill et al (2009): 0 = > 2 s extension of hindlimbs away from lateral midline when mouse is suspended by its tail 2–3 times; 1 = collapse or partial leg collapse toward midline or hind leg trembling during tail suspension; 2 = toes curl under at least twice during a 12‐inch walk, or any part of foot is dragging along cage/table bottom; 3 = rigid paralysis or minimal joint movement, foot not being used for forward motion; and 4 = humane endpoint, mouse cannot right itself within 30 s on either side. The locomotion of animals was quantified over a period of 46 h in the home cage, by automated sensing of body‐heat image using an InfraMot (TSE‐Systems). Individual animal movements were summed up every 30 min. Animal statistics: rotarod performance was evaluated using the two‐way ANOVA with repeated measures for each individual and post hoc Sidak–Holm tests. Weight peak, weight decline, onset of symptoms and survival were calculated with log‐rank (Mantel–Cox) test. All other tests employed two‐way ANOVA with post hoc Sidak–Holm tests. All animal statistics were calculated with GraphPad Prism 6 statistics program.
AE conceived and led the research, performed all non‐human experimental activities and data analysis, including molecular biology and animal studies. RTL and JMR provided human autopsy material. L‐LL, ASi and BA performed laser capture microdissection and human tissue RNA analysis. CE performed protein quantification studies. NR performed microscopic studies, developed and performed in vitro DICER assays. HO and RE performed miRNA in situ hybridization study. ASa and YL performed mass spectrometry. IR, JA, IM, RS, IZB‐D and SMH helped conducting research or provided critical input for scientific interpretations. JMR, TM and EH conceived research and supervised the study. AE and EH developed the interpretations presented and wrote the manuscript with comments from JMR and TM. JMR is the corresponding author for the human analysis, and EH is the corresponding author for all other facets of the study. TM and EH share senior authorship.
Conflict of interest
The authors declare that they have no conflict of interest.
Source Data for Appendix
We thank Ranjan Batra (University of California, SD) for sharing bioinformatics analysis from ongoing studies; Markus Landthaler (Max Delbrück Center for Molecular Medicine), Daryl Bosco (University of Massachusetts, Boston), Zissimos Mourelatos (University of Pennsylvania), Yoram Groner, Menachem Rubinstein, Ari Elson (Weizmann Institute of Science), David Ron (CIMR, University of Cambridge), William Hahn (Dana Farber Cancer Institute, HMS) and David Sabatini (Whitehead Institute, MIT) for vectors and reagents; Pat Janssen and Fred Farin (UW Functional Genomics and Bioinformatics Core Laboratory) for technical assistance and troubleshooting; and Tal Yardeni (Weizmann Institute of Science) for whole spinal cord RNA. We further thank Daniel Dar (Weizmann Institute of Science) for RNA analysis, Chaim Kahana (Weizmann Institute of Science), Orna Elroy‐Stein (Tel Aviv University), Markus Weber (Kantonsspital St. Gallen), Witold Filipowicz (FMI, Basel), Bernard Schneider (EPFL, Lausanne), Mark A. Behlke (Integrated DNA Technologies, Inc.) and Peter Andersen (Umea University) for discussions and advice and Danielle Aronowitz for English editing. This work was supported by grants to E.H. from the Israel Science Foundation, the ISF Legacy‐heritage Program, The Bruno and Ilse Frick Foundation for Research on ALS, Yeda‐Sela, Yeda‐CEO fund, Minna‐James‐Heineman Stiftung through Minerva, ERC Consolidator Program, Israel Ministry of Industry, Trade and Labor “Kamin Program”, Nella and Leon Benoziyo Center for Neurological Diseases, Y. Leon Benoziyo Institute for Molecular Medicine and the ALS‐Therapy Alliance. Additional funding was from Kekst Family Institute for Medical Genetics, David and Fela Shapell Family Center for Genetic Disorders Research, Crown Human Genome Center, Nathan, Shirley, Philip and Charlene Vener New Scientist Fund, Julius and Ray Charlestein Foundation, Fraida Foundation, Wolfson Family Charitable Trust, Adelis Foundation, MERCK UK, Maria Halphen, Estates of Fannie Sherr, Lola Asseof, Lilly Fulop, The “National Network of Excellence in Neuroscience” (NNE), initiated by Teva and miCure Therapeutics. Grants to T.M. from the ALS Association (grant 1765); to L‐L.L. from NIH T32 fellowship NS051171; and to J.R. by Microsoft Research, the family of T. Evans Wyckoff, the friends of Greg Brooks, and Lois Caprile. A.E. was supported by Deutsche Forschungsgemeinschaft. E.H. was the incumbent of the Milton A. Kimmelman Career Development Chair, and the Hornstein Laboratory is further supported by Dr. Sidney Brenner and friends.
FundingIsrael Science Foundation
- © 2015 The Authors