Transcriptome-based screening in TARDBP/TDP-43 knock-in motor neurons identifies the NEDD8-activating enzyme inhibitor MLN4924

Abstract

A growing body of knowledge implicates perturbed RNA homeostasis in amyotrophic lateral sclerosis (ALS), a neurodegenerative disease that currently has no cure and few available treatments. Dysregulation of the multifunctional RNA-binding protein TDP-43 is increasingly regarded as a convergent feature of this disease, evidenced at the neuropathological level by the detection of TDP-43 pathology in most patient tissues, and at the genetic level by the identification of disease-associated mutations in its coding gene TARDBP. To characterize the transcriptional landscape induced by TARDBP mutations, we performed whole-transcriptome profiling of motor neurons (MNs) differentiated from two knock-in iPSC lines expressing the ALS-linked TDP-43 variants p.A382T or p.G348C. Our results show that the TARDBP mutations significantly altered the expression profiles of mRNAs and microRNAs of the 14q32 cluster in MNs. Using mutation-induced gene signatures and the Connectivity Map database, we identified compounds predicted to restore gene expression toward wild-type levels. Among top-scoring compounds selected for further investigation, the NEDD8-activating enzyme inhibitor MLN4924 effectively improved cell viability and neuronal activity, highlighting a possible role for protein post-translational modification via NEDDylation in the pathobiology of TDP-43 in ALS.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-12147-8.

Introduction

Amyotrophic lateral sclerosis (ALS) is a neurodegenerative disorder caused by the progressive degeneration of motor neurons (MNs) in the brain and the spinal cord resulting in weakness, loss of ambulation, and eventual fatal paralysis of respiratory function¹. From disease onset, the mean survival ranges from two to four years². At present, the few available treatments (i.e., riluzole^3–5, edaravone^6,7) offer modest benefits or are only efficacious in a subset of patients, as is the case for the SOD1-lowering therapy tofersen^8–10. Limitations in treating these patients reflect our incomplete understanding of the molecular basis of ALS and the difficulty in therapeutically addressing the multifactorial nature of the disease.

Several lines of evidence implicate disturbed RNA metabolism in ALS. A pathological hallmark of ALS is the nuclear depletion and cytoplasmic aggregation of TAR DNA-binding protein 43 (TDP-43), an RNA-binding protein encoded by TARDBP involved in nearly all aspects of RNA metabolism^11–13. Mutations in TARDBP and several other genes encoding RNA-binding proteins (e.g., FUS, HNRNPA1, HNRNPB2, MATR3, TAF15) have been implicated in ALS^14,15. Furthermore, transcriptome alterations have been repeatedly reported in the brain and spinal cord of patients who have succumbed to this disease^16–23, establishing a strong link between perturbed RNA homeostasis and ALS.

Initially identified as a transcriptional repressor²⁴, TDP-43 has since been implicated in splicing²⁵, microRNA (miRNA) biogenesis^26,27, RNA stability and transport^28,29, and translation³⁰. In keeping with these functions, TDP-43 localizes to ribonucleoprotein (RNP) condensates both in the nucleus (i.e., nuclear speckles³¹, paraspleckles³², and Cajal bodies³²) and the cytoplasm (i.e., transport granules^28,29, stress granules^33–35, and processing (P)-bodies³⁰), which broadly serve to regulate gene expression in time and space. While the many roles of TDP-43 in RNA metabolism have been extensively documented, the impact of disease-associated mutations in TARDBP on the RNA landscape is still being determined. Several studies have reported RNA abnormalities in TARDBP/Tardbp mutant mouse models^36–41, but whether these observed changes accurately reflect the human condition has been questioned.

As dysregulation of gene expression can have broad downstream consequences, determining the transcriptome alterations that arise in ALS can inform the development of transcriptome-correcting therapies. This drug discovery paradigm, known as “transcriptome reversal”⁴² could give rise to therapies able to normalize several disease pathways simultaneously.

To this end, we performed whole-transcriptome profiling of MNs differentiated from two knock-in induced pluripotent stem cell (iPSC) lines expressing the ALS-linked TDP-43 variants p.A382T or p.G348C, leading to the identification of mutation-induced RNA signatures. Positing that shared alterations may be reflective of underlying disease mechanisms, we queried the Connectivity Map (CMap) database⁴³ for compounds predicted to normalize gene expression changes toward wild-type levels. Selected top-scoring compounds identified in silico were tested experimentally for their ability to ameliorate previously identified disease-relevant readouts⁴⁴ in two phenotypic screens for viability and neuronal activity. This approach led to the identification of the NEDD8-activating enzyme inhibitor MLN4924.

Results

Transcriptomic profiling of MNs differentiated from TARDBP knock-in iPSCs

We have recently reported the generation of two homozygous knock-in iPSC lines with mutations in TARDBP coding for TDP-43^A382T or TDP-43^G348C, two frequent ALS variants of TDP-43⁴⁴. To characterize the transcriptomic profiles of MNs differentiated from those cells, we performed next-generation RNA sequencing (RNA-seq) on total RNA extracted from mutant and isogenic control iPSC-derived MN cultures at 4-weeks post-plating of MN progenitor cells (MNPCs) (Fig. 1a, Supplementary Table S1). Analysis of normalized counts confirmed the expression of the MN markers ISL1, MNX1 (Hb9), CHAT, and SLC18A3 (VAChT) (Supplementary Fig. S1a). Unsurprisingly, MNX1 (Hb9) was detected at relatively low levels as this early-MN marker is gradually downregulated with maturation in limb-innervating MNs⁴⁵. In addition, we noted the expression of the V2 interneuron markers LMO4, VSX2/CHX10, and SOX14, reflecting the anatomical proximity of V2 and pMN domains in the developing spinal cord. Markers for astrocytes, oligodendrocytes, neuronal progenitors, and V1/V3 interneurons were mostly absent or expressed at low levels (Supplementary Fig. S1a). Expression of the stem cell marker NANOG was not detected and POU5F1 (Oct-4) levels were negligible. These results suggest a mixed population of cells enriched in ventral spinal neurons with the presence of some glial and progenitor cells, consistent with previous single-cell RNA-seq experiments⁴⁶. Importantly, the expression levels of cell type markers did not significantly differ between mutant and control samples, indicating a similar cell composition and differentiation propensity, as previously described⁴⁴.

Fig. 1.

Differential gene expression analysis in TDP-43^A382T and TDP-43^G348C MNs. (a) Schematic representation of iPSCs differentiation into MN progenitor cells (MNPCs) and MNs for total RNA extraction. (b–e) Volcano plots (b–d) and Venn diagram (e) comparing differentially expressed genes (DEGs) in TDP-43 MNs differentiated for 28 days (4 weeks) relative to isogenic control. n = 5 independent differentiations. (f) Scatter plot showing a strong correlation between fold changes of overlapping DEGs. (g) Principal component analysis (PCA)-based clustering of RNA-seq samples using Log₁₀(n + 1) transformed normalized read counts of DEGs from the combined contrast. Individual points represent biological replicates from independent differentiations. (h) Validation of selected common DEGs by qPCR. Data shown as mean ± SEM. One-way ANOVA with Dunnett’s post hoc test. P-values shown. Significance was defined as P < 0.05. n = 5 independent differentiations. (i) Representative gene ontology (GO) terms of the “molecular function” category enriched in DEGs of TDP-43^A382T and TDP-43^G348C MNs. (j,k) Representative GO terms of the “cellular component” (j) and “biological process” (k) categories enriched in DEGs of TDP-43^G348C MNs. Of note, genes dysregulated in TDP-43^A382T MNs were not significantly enriched in GO terms of these categories.

TDP-43^A382T and TDP-43^G348C affect similar genes and pathways in human MNs

To identify transcriptional changes induced by TARDBP mutations, we performed differential gene expression analysis using DESeq2 (false discovery rate [FDR] < 0.05) by contrasting each mutation to the isogenic control (Fig. 1b-c and Supplementary Data S1). We identified 201 differentially expressed genes (DEGs) in TDP-43^A382T MNs compared with control (137 downregulated genes and 64 upregulated genes), and 324 DEGs in TDP-43^G348C MNs compared with control (161 downregulated genes and 163 upregulated genes) (Supplementary Fig. S1c-d). The expression levels of the ALS-linked genes SOD1, C9ORF72, FUS, and TARDBP did not differ between TDP-43 and isogenic control samples (Supplementary Fig. S1b). When comparing the genes dysregulated in TDP-43^A382T and TDP-43^G348C MNs, there was a significant overlap (P < 1.004e-41, hypergeometric test), revealing a total of 42 DEGs in common (Supplementary Fig. S1d, Supplementary Table S3). Additionally, we combined the datasets from the two mutations and performed differential expression analysis against the isogenic control (TDP-43^A382T + TDP-43^G348C vs. Isog Ctrl) (Fig. 1d, Supplementary Data S1), with the goal of identifying DEGs with a shared direction of effect in TDP-43 samples. This analysis yielded a total of 73 DEGs. Among those, 14 and 18 genes were uniquely dysregulated in TDP-43^A382T and TDP-43^G348C MNs respectively in the prior analyses, while 34 genes were shared between the individual and combined contrasts (Fig. 1e). There was a strong correlation between fold changes in shared differentially expressed genes (r² = 0.9161, slope = 0.9237, P < 0.0001) (Fig. 1f), indicating a similar magnitude of effect on gene expression by the two mutations. The convergence between TDP-43^A382T and TDP-43^G348C MNs was corroborated by findings from a principal-component analysis of DEGs (Fig. 1g), where PC2 distinguished TDP-43^A382T and TDP-43^G348C samples into two distinct but closely related clusters.

We selected a subset of genes for validation by qPCR. We focused on genes showing an overlap between TDP-43^A382T and TDP-43^G348C MNs, particularly those related to cellular processes relevant to ALS and/or associated with ALS or more broadly with neurological disorders (Supplementary Table S3). We also took into consideration the expression levels (i.e., normalized counts) and the log fold changes between conditions, reasoning that a greater magnitude of change and/or alterations in the levels of highly expressed genes are more likely to result in biologically significant effects. These strategies facilitated the selection of a total of 15 genes among those dysregulated in TDP-43 MNs. Differential expression of 10 genes was confirmed in either or both TARDBP mutants in five additional RNA samples from a separate set of extractions (EBF2, PRKCD, CRB1, CHCHD2, GPR50, PCDHA13, PTGDS, RAD51C, ZNF283, ZNF502) (Fig. 1h).

In addition to a shared gene signature, Gene Ontology (GO) enrichment analyses revealed that both mutants showed a dysregulation in genes with molecular functions related to ion binding, DNA binding, and transcription factor activity (Fig. 1i), consistent with altered DNA/RNA-binding protein function. Moreover, genes dysregulated in TDP-43^G348C MNs showed a significant enrichment in the GO cellular component terms “Integral component of plasma membrane”, “intrinsic component of plasma membrane”, and “synaptic membrane” as well as in GO biological processes related to cell adhesion (Fig. 1j, k). Interestingly, the “cell adhesion” category has similarly been reported to be enriched in several transcriptomic studies of ALS patient samples and iPSC-derived MN models^21,47,48. Genes dysregulated in TDP-43^A382T MNs were not significantly enriched in GO terms of the “cellular component” and “biological processes” categories.

To determine whether the gene expression changes identified in our TDP-43 models resemble those dysregulated in other TARDBP mutant iPSC-derived MNs, we turned to a database by Ziff and colleagues⁴⁹ (available from https://oliverziff.shinyapps.io/als_genome_instability/) which catalogs differential expression data generated from RNA-seq datasets of 429 ALS and control iPSC-derived MN lines. Focusing on TARDBP mutations (n = 58 controls vs. n = 10 TARDBP lines [p.Q331K, p.M337V, p.I383T, and p.N390D], we retrieved a total of 3547 DEGs, with 1661 downregulated and 1886 upregulated genes. Comparing this gene set with ours, we found a modest but statistically significant overlap (TDP-43^A382T: P < 2.802e-06, TDP-43^G348C: P < 1.798-11, hypergeometric test) (Fig. 2a, Supplementary Data S2). Approximately 10.4% (21/201) of DEGs in TDP-43^A382T MNs and 9.0% (29/324) of DEGs in TDP-43^G348C MNs were also found to be dysregulated in the retrieved TARDBP mutant dataset. When narrowing our search to genes with a concordant directionality of change, we found 8 co-downregulated genes and 1 co-upregulated between TDP-43^A382T MNs and the compared TARDBP dataset (Fig. 2b, c). Among genes dysregulated in TDP-43^G348C MNs, 9 and 12 genes were co-downregulated and co-upregulated in the TARDBP dataset, respectively (Fig. 2b, c). Furthermore, similar to our findings, the TARDBP MNs showed a significant enrichment in GO terms related to RNA processing and synaptic function⁴⁹. Additional functional terms included “protein binding”, “microtubule cytoskeleton”, and “nervous system development”. Thus, though distinct variants were contrasted in the dataset from Ziff and colleagues, it appears that TARDBP mutations converge, to some extent, on similar genes and pathways.

Fig. 2.

Overlapping RNA alterations across datasets from TARDBP mutant iPSC-derived MNs and post-mortem ALS spinal cords. (a–c) Venn diagram comparing differentially expressed genes (DEGs) in TDP-43^A382T and TDP-43^G348C MNs with other TARDBP mutant iPSC-derived MNs (n = 10 TARDBP MN lines vs. n = 58 controls)⁴⁹. (d–f) Venn diagram comparing DEGs in TDP-43^A382T and TDP-43^G348C MNs with post-mortem spinal cords from patients with genetic and non-genetic forms of ALS (n = 214 post-mortem ALS spinal cords [n = 161 non-genetic ALS, n = 36 C9orf72, n = 5 SOD1, n = 2 FUS, and n = 10 with mutations in 8 other ALS genes] vs. n = 57 control specimens)⁴⁹.

TDP-43 MNs share gene expression changes with post-mortem ALS spinal MNs

To further assess the relevance of our findings, we compared our data with a previously published dataset obtained from post-mortem spinal cord specimens from 57 neurologically healthy individuals and 214 ALS patients from the NYGC ALS cohort (n = 161 non-genetic ALS, n = 36 C9orf72, n = 5 SOD1, n = 2 FUS, and n = 10 with mutations in 8 other ALS genes), available on the database by Ziff and colleagues⁴⁹ queried above. We retrieved a total of 14,049 DEGs in ALS samples compared with controls (FDR < 0.05); specifically, 6567 upregulated genes and 7482 downregulated genes. When comparing all dysregulated genes, 34.8% (70/201) and 41.4% (134/324) of genes dysregulated in TDP-43^A382T and TDP-43^G348C MNs, respectively, were shared with post-mortem MNs (TDP-43^A382T: P < 6.457e-129, TDP-43^G348C: P < 2.338e-13, hypergeometric test) (Fig. 2d, Supplementary Data S2). A total of 14 genes were commonly dysregulated across all contrasts. When considering directionality of change, 17 downregulated genes and 15 upregulated genes were common between TDP-43^A382T MNs and post-mortem ALS spinal cords (Fig. 2e, f). A total of 25 downregulated genes and 25 upregulated genes were shared between post-mortem and TDP-43^G348C MNs (Fig. 2e, f). Among those, three genes were shared between both mutants and post-mortem samples (PCDHA13 and MEG3, downregulated; PRKCD, upregulated) (Fig. 2e, f). Thus, TARDBP mutant iPSC-derived MNs recapitulate some of the transcriptional dysregulations observed in post-mortem ALS spinal cord samples.

Few dysregulated genes are known TDP-43 targets

To further investigate the mechanisms underlying RNA changes induced by TARDBP mutations, we next considered a relationship between differential gene expression and mRNA binding by TDP-43. We turned to a previous study characterizing TDP-43 target mRNAs by enhanced crosslinking and immunoprecipitation (eCLIP) followed by RNA-seq in motor cortex specimens from sporadic ALS patients and neurologically healthy individuals¹⁹. We compared our DEGs with a list of genes whose mRNAs were previously reported to be bound by TDP-43 (Supplementary Fig. S2). We didn’t find a significant enrichment for TDP-43 targets within our datasets (TDP-43^A382T P < 0.086; TDP-43^G348C P < 0.177, hypergeometric test). Only a small fraction of DEGs were known TDP-43 targets (~ 2.5% in TDP-43^A382T and ~ 1.8% in TDP-43^G348C). Two TDP-43-bound genes were dysregulated in both TDP-43 MN lines (PTGDS, MEG3), while three and four TDP-43-bound genes were only dysregulated in TDP-43^A382T (ZNF532, PAPLN, TNS1) or TDP-43^G348C MNs (GRIN1, ARF3, CNTNAP4, RERE), respectively. Among dysregulated TDP-43-bound genes, the majority (6/9) were downregulated. Given that, to our knowledge, similar cross-linking experiments characterizing TDP-43-bound transcripts have not yet been performed in human spinal cord specimens or spinal MN cultures, additional RNA-protein interaction studies are needed to reveal cell type-specific TDP-43 binding targets. In line with this idea, recent studies have shown that different cellular environments can modulate the RNA processing functions of TDP-43^50,51.

MicroRNA biogenesis is altered in mutant MNs

Given the roles of TDP-43 in miRNA biogenesis^27,52, we next interrogated whether TARDBP mutations could indirectly affect gene expression through changes in miRNA abundance. To this end, we performed miRNA profiling by small RNA sequencing to characterize miRNA abundance in TDP-43^A382T and TDP-43^G348C MN cultures. We conducted differential expression analyses (FDR < 0.05) using the same contrast design used in the RNA-seq experiments (Fig. 3a-c and Supplementary Data S3). Differential expression analyses revealed a total of 40 and 47 miRNAs with an altered abundance in TDP-43^A382T and TDP-43^G348C MNs, respectively. The two mutants had largely overlapping dysregulated miRNAs (P < 0.05, hypergeometric test), with 36 miRNAs in common (Supplementary Fig. S3a, Supplementary Table S4). The combined analysis (TDP-43^A382T + TDP-43^G348C vs. Isog Ctrl) identified a total of 46 dysregulated miRNAs, including the 36 miRNAs shared with the individual analyses (Fig. 3d). Most miRNAs showed decreased abundance with a strong correlation in fold changes between the two mutants (r² = 0.9201, slope = 0.8254, P < 0.0001) (Fig. 3e), which could indicate altered miRNA biogenesis or excess degradation. Downregulation was confirmed in 1 out of 7 miRNAs selected for validation (has-miR-381-3p), although most miRNAs showed a clear trend toward decreased abundance relative to the isogenic control (Fig. 3f).

Fig. 3.

Differential miRNA expression analysis in mutant MNs. (a–d) Volcano plots (a–c) and Venn diagram (d) comparing differentially expressed miRNAs in TDP-43 MNs relative to isogenic control. n = 5 independent differentiations. (e) Scatter plot showing a strong correlation between fold changes of overlapping dysregulated miRNAs. (f) Validation of selected common dysregulated miRNAs by qPCR. Data shown as mean ± SEM. One-way ANOVA with Dunnett’s post hoc test. P-values shown. Significance was defined as *P < 0.05. n = 5 independent differentiations.

Given that TDP-43 can directly associate with pri-miRNA and pre-miRNA precursors during miRNA maturation^27,52, we queried the RBPmap database⁵³ to search for TDP-43 binding motifs in the transcribed sequences of dysregulated miRNAs. We found one or more predicted TDP-43 binding sites in nearly half of the shared dysregulated miRNAs (17/36 miRNAs, Supplementary Table S5). Interestingly, we also noticed that the genomic coordinates of all 35 downregulated miRNAs mapped to the 14q32 miRNA cluster (Supplementary Table S4), which strongly implies that their expression is co-regulated. Accordingly, most of them were downregulated by the same magnitude (~ 3 log fold-changes) in TDP-43 samples relative to the isogenic control in differential expression analyses (Fig. 3e, Supplementary Table S4). Taken together, these results indicate that mutations in TARDBP coding for TDP-43^A382T and TDP-43^G348C lead to a decreased abundance of 14q32-encoded miRNAs.

Functional consequences of dysregulated miRNA in mutant MNs

To explore the potential functional consequences of miRNA perturbances, we used miRgate⁵⁴ to predict targets of shared dysregulated miRNAs (Fig. 4a, Supplementary Data S4), filtering for target genes predicted by at least two out of five computational algorithms used by this database. Given that miRNAs can regulate gene expression post-transcriptionally through mRNA degradation, we considered an inverse correlation between dysregulated miRNAs and the expression levels of their predicted target mRNAs. The list of predicted genes targeted by downregulated miRNAs was compared to genes upregulated in mutant MNs determined by RNA-seq data (and vice versa). These mRNA/miRNA cross-analyses yielded a total of 13 and 30 genes identified as inversely correlated to miRNA abundance in TDP-43^A382T and TDP-43^G348C MNs, respectively, while a total of 6 genes were identified in the combined analysis (TDP-43^A382T + TDP-43^G348C vs. Isog Ctrl) (Fig. 4b, Supplementary Fig. S3b). These results indicate that the detected changes in gene expression could be partially explained by impaired miRNA biogenesis. Interestingly, we noted that several negatively correlated genes encode protocadherins (Supplementary Fig. S3b), which are cell-cell adhesion proteins enriched in the central nervous system (CNS)⁵⁵. Upon re-examination of the RNA-seq data, we also found that DEGs were statistically enriched for the clustered protocadherin locus on chromosome 5 in both mutants (FDR < 0.05) (Supplementary Fig. S3d).

Fig. 4.

Functional characterization of dysregulated miRNAs. (a) Bioinformatics analysis workflow for functional characterization of dysregulated miRNAs. (b) Integrated mRNA/miRNA analysis comparing DEGs and predicted mRNA targets of dysregulated miRNAs in TDP-43 MNs. (c) Representative common enriched GO terms of predicted mRNA targets of dysregulated miRNAs.

Aside from targeted mRNA degradation, miRNAs can achieve RNA silencing through translational repression, with little to no influence on mRNA target abundance⁵⁶. To gain further insights into the pathways regulated by altered miRNA, we performed GO enrichment analyses of predicted target genes and investigated shared enriched GO terms per category. Overall, there was a common enrichment in GO terms related to synaptic function, cell adhesion, and neuronal development (Fig. 4c). Furthermore, KEGG pathway analysis^57,58 identified a common enrichment in pathways related to cell signaling, axon guidance, and neurodegeneration (Supplementary Fig. S3c).

In silico screen identifies compounds predicted to restore mutation-induced gene expression signatures

Next, we reasoned that mutation-induced gene expression profiles might represent a potential therapeutic opportunity. We posited that transcriptomic alterations may be reflective of underlying disease mechanisms and that normalizing gene expression may have beneficial effects. Thus, we utilized the CMap database⁴³ (which comprises of transcriptional profiles induced from thousands of small molecules) to screen for compounds predicted to normalize gene expression changes caused by TARDBP mutations toward wild-type levels. We used as inputs our lists of DEGs identified by RNA-seq and obtained as outputs lists of compounds accompanied by their τ scores, indicating the similarity or dissimilarity between the transcriptional changes they induce and the queried DEGs list (Fig. 5a, Supplementary Table S6, Supplementary Data S5). The scores obtained from each queried DEG list were plotted for visualization of hits common between individual and combined datasets (Fig. 5b). Among 16 top-compounds identified (score<-85), we selected 6 candidate compounds from the lower left corner (i.e., predicted to reverse gene expression changes) with high scores in the combined dataset and/or at least one of the individual datasets for further investigation (Fig. 5c). For this prioritization, we took into consideration compound characteristics that could facilitate clinical translation, including currently approved therapeutic uses or investigational uses, known mechanisms of action, and blood-brain barrier penetrance.

Fig. 5.

Transcriptome-based in silico and in vitro phenotypic screens identify one compound that ameliorates MN survival and firing activity. (a) Screening funnel used to identify candidate compounds. (b) Scatter plot of the τ scores from the individual (y-axis) versus combined (x-axis) datasets gene signatures. Each data point represents one compound of the CMap library. Arrow heads show selected compounds. (c) Description and τ scores of selected compounds. (d) Heatmap of mean MN survival relative to DMSO control after treatment with compounds (0.1 µM and 1.0 µM) in culture conditions without neurotrophic factors (NFs) supplementation. n = 4 independent differentiations. One-way ANOVA with Sidak’s post hoc test. (e) Overview of in vitro phenotypic screens. (f) Bar graph of MN viability after treatment with MLN4924. Data points represent individual wells from n = 4 independent differentiations. One-way ANOVA with Sidak’s post hoc test. (g) Mean firing rate of MNs treated with candidate compounds (1.0 µM) calculated from per-well mean values from n = 4 independent differentiations. Two-way ANOVA with Dunnett’s post hoc test. Data shown as mean ± SEM. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

Treatment with hit compound MLN4924 ameliorates MN viability and firing activity

In our next steps, we aimed to determine whether the six selected hit compounds could improve ALS-relevant phenotypic readouts, including MN viability (Fig. 5a). In previous work, we showed that removal of neurotrophic factors (NFs) supplementation from the differentiation medium significantly decreased the viability of MNs irrespective of the cell line or genotype⁴⁴. Thus, we treated MNs cultured without NFs with two concentrations of the candidate compounds (0.1 µM and 1.0 µM) for 6 days to screen for compounds with neuroprotective properties (Fig. 5e). We also included in our compound panel two FDA-approved drugs for the treatment of ALS, namely riluzole and edaravone, as well as the combination of sodium phenylbutyrate (PB) and tauroursodeoxycholic acid (TUDCA), which initially showed promising outcomes in ALS randomized controlled trials^59–61. Following treatment, one compound (MLN4924) effectively improved viability of isogenic control and TDP-43^A382T MNs compared with the vehicle-treated condition at the highest tested concentration, with a similar trend observed in TDP-43^G348C MNs (Isog Ctrl: +29.3%, P = 0.0352; TDP-43^A382T: +28.4%, P = 0.0472; TDP-43^G348C: +14.5%, P = 0.8681 by one-way ANOVA with Sidak’s multiple comparisons test) (Fig. 5d-f). Perhaps surprisingly, treatments with the ALS drugs did not show similar neuroprotective effects at the concentrations used.

We previously showed using multi-electrode array (MEA) that TDP-43 MNs displayed hypoactivity compared with isogenic control MNs after several weeks in culture⁴⁴. Thus, in addition to viability, we also explored if one or more of the compounds might ameliorate this mutation-induced phenotype. We examined the effects of the six selected hit compounds on neuronal activity in a single-dose screen at the highest concentration (1.0 µM) (Fig. 5a). For this assay, MNs were cultured in complete final differentiation medium with NFs to maintain optimal MN viability for electrophysiological recordings (Fig. 5e). Interestingly, we found that MLN4924 treatment significantly increased the mean firing rate of TDP-43^A382T and TDP-43^G348C MN cultures compared to untreated cells (TDP-43^A382T: +1.7 Hz, P = 0.0091; TDP-43^G348C: +1.5 Hz, P = 0.0485 by two-way ANOVA with Dunnett’s multiple comparison test) (Fig. 5g). Overall, we demonstrate that combining in silico and multi-phenotypic screens approaches enabled the identification of one compound able to improve neuronal viability and activity.

Components of the NEDDylation pathway are expressed in iPSC-derived MNs and are modulated by MLN4924 treatment

To explore the mechanisms underlying MLN4924’s effects in our models, we considered its documented mode of action. MLN4924 is an inhibitor of NEDDylation, a post-translational modification (PTM) similar to ubiquitylation, where an activated NEDD8 is conjugated to a protein substrate⁶². This pathway involves pre-NEDD8 processing enzymes, NEDD8-activating enzyme (NAE) (a heterodimer consisting of NAE1 and UBA3), NEDD8-conjugating enzyme E2, NEDD8-E3 ligases, and de-NEDDylation enzymes (Fig. 6a)⁶². Specifically, MLN4924 inhibits NAE activity, thereby blocking the initial step of this process. Returning to our RNA-seq data, we found that all cell lines expressed numerous genes of the NEDDylation pathway at levels nearly comparable to those of the neuronal markers tubulin β-3 (TUBB3) and neurofilament heavy chain (NEFH) (Fig. 6b). Western blot experiments revealed similar levels of free NEDD8, NEDD8-conjugated proteins, and NAE1 across cell lines at baseline (DMSO-treated samples) (Fig. 6c-g, Supplementary Fig. S4). We observed a strong trend towards decreased levels of NEDD8-conjugated proteins upon treatment with MLN4924, though only reaching statistical significance in TDP-43^G348C MNs (Isog Ctrl P = 0.0799; TDP-43^A382T P = 0.0705; TDP-43^G348C P = 0.0266 by one-way ANOVA with Sidak’s multiple comparisons test) (Fig. 6c, f, Supplementary Fig. S4). There was also a mild upward trend of free NEDD8 and NAE1 levels post-treatment, mostly in TARDBP mutant MNs (Fig. 6c, d,e, g, Supplementary Fig. S4). This effect was more evident at the RNA level, with MLN4924 treatment inducing upregulation of NEDD8, UBE3, and NAE1 transcripts to varying degrees across lines (Fig. 6i-k), perhaps reflecting a compensatory response to NEDDylation inhibition. We also analysed the effect of MLN4924 treatment on TDP-43 expression and did not detect any significant changes both at the RNA and protein levels (Fig. 6d, h,l). Overall, our results show that NEDDylation pathway components are expressed in iPSC-derived MNs and appear to be modulated by MLN4924 treatment.

Fig. 6.

MLN4924 treatment inhibits protein NEDDylation and induces a compensatory upregulation of pathway components. (a) Schematic representation of the NEDDylation pathway and target of MLN4924. (b) Heatmap of normalized counts of genes involved in the NEDDylation pathway determined by RNA-seq. Color legend refers to NEDDylation steps illustrated in (a). Normalized counts of neuronal (MAP2, TUBB3, NEFH), astrocyte (S100B) and stem cell (POU5F1) markers are displayed for comparison. (c–h) Representative immunoblots (c,d) and quantification of free NEDD8 (e), NEDD8-conjugated proteins (f), NAE1 (g), and TDP-43 (h) levels in 2-week MNs treated with DMSO or MLN4924 (1.0 µM). βIII-tubulin was used as loading control. n = 3 independent experiments. Original uncropped immunoblots are presented in Supplementary Fig. S4. (i–l) Quantification of relative transcript levels of NEDD8 (i), UBA3 (j), NAE1 (k), and TARDBP (l) in 4-week MNs treated with DMSO or MLN4924 (1.0 µM) by qPCR. n = 4 independent experiments. Data shown as mean ± SEM. Two-way ANOVA with Tukey’s and Sidak’s post hoc tests. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

Discussion

Accumulating evidence implicates RNA dyshomeostasis in the pathobiology of ALS. In particular, the egress of TDP-43 from the nucleus implies a loss of nuclear RNA processing functions at least in advanced stages of ALS, prompting many recent transcriptomic studies in TARDBP knockdown cells^16,17,63,64. However, less attention has so far been given to the impact of ALS-linked TARDBP mutations on global RNA metabolism in human iPSC-derived neurons^51,65,66. Here, we performed whole-transcriptome profiling of MNs differentiated from iPSCs with TARBDP mutations coding for TDP-43^A382T or TDP-43^G348C. Our study is the first to provide both mRNA and miRNA expression profiles with integrated mRNA/miRNA analyses, with the aim of increasing our understanding of TDP-43 pathobiology in ALS and identifying molecules able to restore disease-relevant phenotypes.

Our findings highlight significantly overlapping RNA alterations in TDP-43^A382T and TDP-43^G348C MNs, indicating common pathogenic effects of these mutations on RNA processing. Further comparison of our results with published datasets from other TARDBP mutant iPSC-derived MN lines and post-mortem spinal cord specimens revealed shared transcriptomic alterations. Importantly, these results point to several genes previously linked to ALS and other neurodegenerative disorders including frontotemporal dementia (FTD), Parkinson’s disease (PD), and Lewy body dementia (LBD). Among shared DEGs, the long non-coding RNA (lncRNA) MEG3, implicated in MN development⁶⁷, was downregulated in both TDP-43^A382T and TDP-43^G348C datasets and in post-mortem ALS spinal cord specimens. This lncRNA is a known TDP-43 target^19,68 and loss of TDP-43 binding to MEG3 was previously observed in brains of patients with FTD⁶⁹. Levels of the PD- and LBD-associated gene CHCHD2^70–72 were also significantly reduced in TDP-43^A382T and TDP-43^G348C MNs, but were increased in the post-mortem dataset. Coiled-coil-helix-coiled-coil-helix domain containing 2 (encoded by CHCHD2) forms a complex with CHCHD10, another nucleus-encoded mitochondrial protein of the CHCHD-containing protein family, known to be genetically linked to ALS and FTD^73–75. Interestingly, CHCHD10 was significantly downregulated in TDP-43^A382T MNs and post-mortem ALS spinal cords. Previous studies have similarly reported loss of CHCHD2 expression in TDP-43^M337V and TDP-43^Q331K iPSC-derived MNs⁶⁵ and decreased CHCHD10 expression in sporadic ALS patient-derived MNs⁷⁶. Additionally, CHCHD2 and two other shared DEGs identified here (PRKCD, MYH14) were predicted to be candidate ALS genes in a machine learning study⁷⁷.

Nonetheless, a considerable number of changes observed in post-mortem samples are not recapitulated in iPSC-derived MNs, as previously observed^49,78. These differences may reflect the limitations of iPSC-based models such as cell immaturity⁷⁹, the loss of age-related epigenetic marks during reprogramming⁸⁰, and the absence of other cell lineages needed to capture non-cell-autonomous effects. It is also worth noting that TDP-43^A382T and TDP-43^G348C MNs did not mirror the transcriptional signature of TDP-43 loss-of-function (implicating STMN2, UNC13A, ELAVL3, PFKP, RCAN1, SELPLG) established by previous knockdown studies and other ALS iPSC models^{16,17,51,63,64,81,82}, perhaps consistent with the preserved nuclear localization of TDP-43 in our cells⁴⁴. Thus, it appears that the TDP-43^A382T and TDP-43^G348C variants impact RNA processing by mechanisms distinct from nuclear clearance, as noted in previous TARDBP/Tardbp mutant mouse models^36–39. Like these studies, we posit that the observed mutation-induced transcriptome profiles may reflect early RNA perturbations in the pathogenesis of TDP-43-ALS, preceding nuclear TDP-43 loss.

We explored the possibility that the TARDBP mutations studied could impact gene expression through altered miRNA biogenesis. miRNAs are highly expressed in the CNS and have been shown to be dysregulated in ALS patient specimens^83–89. We found a marked overlap in dysregulated miRNAs in both mutant MNs with concordant fold changes and directionality, some of which were previously linked to ALS. Specifically, dysregulation of miR-370-3p, miR-409-3p, and miR-495-3p was documented in iPSC-derived MNs with homozygous FUS^P517L mutations⁹⁰. Other dysregulated miRNAs, namely miR-432-5p, miR-323a-3p and miR-453, were predicted here to target ALS2, associated with juvenile ALS^91,92. Additionally, the ALS-associated gene SARM1^93,94 was predicted to be regulated by miR-432-5p, mir-370-3p, miR-453, miR-495-3p, and miR-654-5p, all miRNAs that were downregulated in our study. In agreement with the studies cited above, our work may support the view that miRNAs contribute to ALS pathophysiology.

Strikingly, most altered miRNAs mapped to the 14q32 miRNA cluster, implying that their expression is regulated by a common transcriptional or epigenetic process. Members of miRNAs clusters typically share similar biological functions by targeting the same sets of genes or functionally related genes of the same pathways⁹⁵. The miRNAs of the 14q32 cluster (also known as the miR379–410 or C14MC cluster) are highly conserved, brain-enriched miRNAs which are involved in many aspects of neuronal development and function including neuronal migration, neurite formation, and synaptic plasticity^96,97. Accordingly, we found several differentially regulated miRNA/mRNA pairs implicating protocadherins, which play key roles in neural circuit formation⁵⁵. The miRNAs of this cluster were also previously associated with brain disorders including epilepsy^98,99, schizophrenia^100,101, autism spectrum disorder^102,103, and brain cancers^104,105. Interestingly, miRNAs of this cluster were shown to be selectively exported via exosomes¹⁰⁶, and several have been detected in the serum¹⁰⁷, suggesting they could constitute promising biomarkers. In fact, levels of the 14q32-encoded miRNA miR-127-3p were altered in MNs in this study as well as in the serum of sporadic and familial ALS patients in two previous studies^108,109.

In addition to biomarkers, miRNAs are also increasingly regarded as potential therapeutic targets, given that miRNAs can modulate hundreds of targets simultaneously¹¹⁰. In future work, individual miRNA mimics or synthetic miRNA clusters¹¹¹ may be employed to modulate levels of dysregulated miRNAs and their targets to assess phenotypic rescue. Such therapeutic strategies capable of acting more broadly on multiple cellular pathways, rather than on a few targets, may be more likely to have a clinically significant impact on the disease course. However, an important consideration for miRNA replacement therapy is the potential for undesirable off-target effects, which can be hard to predict or avoid¹¹² as well as the invasive nature of the treatments (i.e., intrathecal injections) that are not without risk for patients.

Small orally bioavailable molecules able to improve the function and viability of MNs may be promising ALS therapies, given their ease of administration. Thus, we leveraged the transcriptomic profiles of TDP-43 MNs to search for compounds predicted to correct gene expression using the CMap database. This drug discovery strategy was previously applied to several human conditions including aortic valve disease^113,114, non-alcoholic steatohepatitis¹¹⁵, epilepsy^116–118, schizophrenia¹¹⁹, neurodevelopmental disorders⁴², and FTD¹²⁰, yet remains largely unexplored in the ALS field. Among top-scoring compounds, several had pharmacological effects that could potentially be relevant for the treatment of ALS. Some were known to act in the CNS for the treatment of neurological symptoms (e.g., neuroleptic/antipsychotic, anticonvulsant) and/or were ligands of neurotransmitter receptors (e.g., dopamine receptor, serotonin receptor), namely sulpiride, QS-11, GR-46,611, and CS-110266. Interestingly, sulpiride is a neuroleptic and in a previous large-scale chemical screen conducted in TDP-43^A315T worms and TDP-43^G348C zebrafish, all the compounds identified as hits were also neuroleptics¹²¹. Additionally, while there is evidence of increased MAP kinase signaling in ALS patients as well as in animal and iPSC models^{49,66,122–124}, two compounds identified here were MAP kinase inhibitors (XMD-892, XMD-885). Finally, XMD-892¹²⁵ and other compounds (tacedinaline^126–128, piperlongumine^129,130, prunetin¹³¹) have also been investigated for their antioxidant, anti-inflammatory, anti-ischemic, neuroprotective, and/or antitumour properties. In silico compound prediction was followed by two in vitro screens based on viability and hypoactivity phenotypes described in previous work⁴⁴. We tested a panel of 6 top-scoring compounds along with three ALS drugs: riluzole, edaravone, and PB/TUDCA. These efforts led to the identification of the NEDDylation inhibitor MLN4924, which effectively improved neuronal viability and firing activity. Finally, we showed that components of the NEDDylation pathway are enriched in MNs and that MLN4924 treatment reduces levels of NEDD8-conjugated proteins, thereby demonstrating target engagement.

Protein NEDDylation has been positively implicated in neuronal development, synapse formation, and neurotransmission^132–136. However, overactivation of NEDDylation has been linked to neuronal apoptosis¹³⁷ and brain malignancies^138,139. NEDD8 has been detected within ubiquitin-positive inclusions in several neurodegenerative diseases including ALS, ALS/FTD, PD, and Alzheimer’s disease¹⁴⁰. Furthermore, a single nucleotide polymorphism (SNP) in DCUN1D1, a component of the NEDDylation pathway¹⁴¹, was found to increase the risk of developing FTD by about 4-fold¹⁴². MLN4924 is a first-in-class drug to have entered phase II and III clinical trials for the treatment of various cancers¹⁴³. As it is brain-penetrant and generally well tolerated by patients¹⁴³, its therapeutic effects are also being considered for the treatment of neurological disorders. MLN4924 was shown to be neuroprotective against ischemic injury in a mouse model of ischemic stroke¹⁴⁴. This compound was also reported to attenuate reactive oxygen species production and cell damage and death induced by H₂O₂ stress in rat primary neurons¹⁴⁵. Recently, genetic modulation of the NEDDylation pathway ameliorated motor phenotypes in nematode models of ALS expressing mutant forms of SOD1 and C9ORF72¹⁴⁶. Here, we showed that MLN4924 exerted desirable effects in TARDBP mutant iPSC-derived MNs, although its mechanisms of action in our models remains to be defined. Given that PTMs can regulate protein-protein interactions and liquid-liquid phase separation behavior¹⁴⁷, we speculate that NEDDylation inhibition by MLN4924 may modulate the formation of RNP condensates, a process central to several RNA processing functions of TDP-43. In fact, modulating the NEDDylation pathway has been linked to the assembly and disassembly of stress granules^146,148,149. NEDDylation has also been shown to promote nuclear protein aggregation under proteotoxic stress conditions¹⁵⁰. These findings and the apparent neuroprotective effects of MLN4924 in our iPSC models highlight the therapeutic potential of this compound and calls for further investigation into the roles of protein NEDDylation in TDP-43 biology and ALS.

In summary, this study highlights the value of whole-transcriptome profiling for shedding light on disease-relevant perturbed pathways, and for identifying potential therapeutic candidates through a transcriptome reversal paradigm. In silico prediction of MLN4924 translated into improvement of disease-relevant phenotypes in vitro in our models, demonstrating the potential of this strategy for drug discovery in ALS and other neurodegenerative diseases.

Methods

All methods were performed in accordance with the relevant guidelines and regulations.

iPSC lines and culture

The use of human cells in this study was approved by McGill University Health Center Research Ethics Board (DURCAN_iPSC / 2019–5374). The TARDBP knock-in cell lines used in this study were generated from the control cell line AIW002-02, as detailed in our earlier work⁴⁴. This control cell line was reprogrammed from peripheral blood mononuclear cells (PBMCs) of a 37-year-old Caucasian male, as previously described¹⁵¹. Informed consent was obtained from this subject. A summary of the iPSC lines used can be found in Supplementary Table S1. iPSCs were maintained on dishes coated with Matrigel (Corning Millipore; Cat#354277) in mTeSR1 (StemCell Technologies; Cat#85850) with daily media change and passaged at 80% confluence using Gentle Cell Dissociation Reagent (StemCell Technologies; Cat#07174). Cultures were routinely tested for mycoplasma using the MycoAlert Mycoplasma Detection kit (Lonza; Cat#LT07-318).

Differentiation of MNs from iPSCs

IPSCs were first induced into MN progenitor cells (MNPCs) by dual-SMAD signaling inhibition and ventral neural patterning as previously described¹⁵². Banked MNPCs were thawed onto dishes coated with 10 µg/ml Poly-L-ornithine (PLO) (Sigma-Aldrich; Cat#P3655) and 5 µg/ml laminin (Sigma-Aldrich; Cat#L2020) in “expansion medium” composed of basic neural medium (1:1 mixture of DMEM/F12 medium (Gibco; Cat#10565–018) and Neurobasal medium (Life Technologies; Cat#21103–049), 0.5X N2 (Life Technologies; Cat#17502–048), 0.5X B27 (Life Technologies; Cat#17504–044), 0.5X GlutaMAX (Gibco, Cat#35050-061), 1X antibiotic-antimycotic (Gibco, Cat#15240–062), and 100 µM ascorbic acid (Sigma-Aldrich; Cat#A5960)) supplemented with 3 µM CHIR99021 (Selleckchem; Cat#S2924), 2 µM SB431542 (Selleckchem; Cat#S1067), 2 µM DMH1 (Selleckchem; Cat#S7146), 0.1 µM retinoic acid (Sigma-Aldrich; Cat#R2625), 0.5 µM purmorphamine (Sigma-Aldrich; Cat#SML-0868), 0.5 mM valproic acid (Sigma-Aldrich; Cat#P4543)) and 10 µM ROCK inhibitor Y-27632 (Selleckchem; Cat#S1049) (for the first 24 h), and were allowed to recover up to 6 days with medium fully changed every other day.

For final MN differentiation, MNPCs were dissociated as single cells with Accutase (StemCell Technologies; Cat#07922) and were seeded on dishes coated with 10 µg/ml PLO and 5 µg/ml laminin (Life Technologies; Cat#23017-015) in “final differentiation medium” composed of basic neural medium supplemented with 0.5 µM retinoic acid, 0.1 µM purmorphamine, 0.1 µM Compound E (StemCell Technologies; Cat#73954), and 10 ng/ml insulin-like growth factor-1 (IGF-1, Peprotech; Cat#100–11), brain-derived neurotrophic factor (BDNF, Peprotech; Cat#450–02) and ciliary neurotrophic factor (CNTF, Peprotech; Cat#450–13)), with half-changes at least every other week. Alternatively, for viability and MEA experiments, MNPCs were first passaged at a 1:3 ratio in “priming medium” composed of basic neural medium supplemented with 0.5 µM retinoic acid and 0.1 µM purmorphamine with medium changed every other day for 6 days before they were plated for final differentiation as described above.

Library preparation and sequencing

Total RNA was extracted from MN cultures using the miRNeasy micro kit (Qiagen; Cat#217004) with DNase treatment (Qiagen; Cat#79256) following the manufacturer’s instructions. Samples with RNA integrity numbers above 6.5 determined by a Bioanalyzer (Agilent) were used for library preparation using the NEB mRNA stranded library preparation kit (for RNA-seq) and the NEB small RNA library preparation kit (for small RNA-seq). Libraries were prepared and sequenced at Genome Québec (Montreal, Canada) on a Illumina NovaSeq 6000 sequencing system with paired-end 100 bp (PE100) strategy.

RNA-seq analysis

Sequencing files were analyzed using the GenePipes RNA-seq pipeline 3.6.0¹⁵³. Briefly, raw reads were clipped for adapter sequence, trimmed for minimum quality (Q30) in 3’ and filtered for minimum length of 32 bp using Trimmomatic¹⁵⁴. Surviving read pairs were aligned to the human genome assembly hg38 by the ultrafast universal RNA-seq aligner STAR¹⁵⁵ using the recommended two passes approach. Aligned RNA-seq reads were assembled into transcripts and their relative abundance was estimated using Cufflinks¹⁵⁶. Differential expression analyses were conducted using DESeq2¹⁵⁷ with a false discovery rate of 5% with no cut-off on log fold-change. Volcano plots were created using the Molecular and Genomics Informatics Core (MaGIC) Volcano Plot Tool, available at https://volcano.bioinformagic.tools/.

miRNA profiling

Sequencing datasets were analyzed with a modified version of the workflow proposed by Potla et al.¹⁵⁸. We adapted this workflow to analyze reads without UMI’s, single end and paired end readsets, and multiple readsets per sample (this workflow is available at https://github.com/neurobioinfo/miRNA_workflow). After quality control with FastQC (http://www.bioinformatics.babraham.ac.uk/projects/fastqc/), we used cutadapt to remove reads shorter than 15 nt and larger than 32 nt. We used Bowtie v1.3¹⁵⁹ to align readsets to the miRBase database v22^160–164 and to the human reference genome v104. In order to maximize the number of alignments and to detect rare miRNAs, we used a seed length 20 (Bowtie option -l 20) and allowed 3 mismatches within the seed (Bowtie option -n 3). MiRNAs aligned to miRBase were counted with Samtools idxstats¹⁶⁵. MiRNAs aligned to the reference genome were filtered with TAGBAM from BEDTools¹⁶⁶ and counted with a locally developed script that optimizes counting time (https://github.com/neurobioinfo/miRNA_workflow). Finally, the local workflow adds miRNAs aligned to miRBase to those aligned to the reference genome for each sample, and generates a miRNA counts matrix with all samples counts. Before differential expression analyses with the R package DESeq2¹⁵⁷, the count matrix was pre-filtered to keep miRNAs having at least 10 counts. DESeq2 estimates dispersions using the Cox-Reid method¹⁶⁷. It uses negative binomial GLM fitting for log2 fold changes estimation and the Wald statistics for hypothesis testing. We used a false discovery rate of 5% with no cut-off on log2 fold change.

Gene ontology and KEGG enrichment analysis

Terms from the Gene Ontology were tested for enrichment with the ShinyGO online tool, available at http://bioinformatics.sdstate.edu/go/.

miRNA target prediction

Prediction of miRNA targets was performed using miRGate⁵⁴, which uses five different public prediction algorithms. We considered target genes predicted by at least two algorithms.

Quantitative PCR

Expression levels of genes of interest were quantified using quantitative PCR (qPCR). cDNA synthesis was performed with 500 ng of total RNA using the M-MLV Reverse Transcriptase kit (Thermo Fischer Scientific; Cat#28025013) in a total volume of 40 µL. To quantify expression levels of selected miRNAs, cDNA synthesis was performed using 10 ng of total RNA with the TaqMan^® Advanced miRNA cDNA Synthesis Kit following the manufacturer’s instructions. Real-time qPCR reactions were set up in triplicates with the Applied Biosystems Applied Biosystems SYBR green (Applied Biosystems; Cat#A25778) or TaqMan^® Fast Advanced Master Mix (Applied Biosystems; Cat#A44360) and run on a QuantStudio 5 Real-Time PCR system (Applied Biosystems Cat#A28140). Primers and TaqMan^® probes references are provided in Supplementary Table S2. Levels of the endogenous control miRNA hsa-miR-191-5p¹⁶⁸ and the geometric mean of the endogenous control genes 18 S (RN18S1), HPRT1, PPIA and POLR2A were used for normalization. Normalized expression was displayed relative to the relevant control condition.

In silico screen with the CMap database

The lists of down/upregulated genes of each contrast were used as inputs in the CMap database⁴³ to assess the similarity (or dissimilarity) between the query gene set and gene expression profiles induced by a library of small molecules. The obtained output is a list of the small molecules rank-ordered by their CMap connectivity score (τ), a normalized metric ranging from − 100 to 100. A negative score indicates opposing profiles between the small molecule and the query gene set, thereby predicting the small molecule to reverse gene expression changes. In contrast, a positive score indicates similar profiles.

Two-dose viability screen

MNPCs were plated as 15,000 cells per well in opaque white optical 96-well plates coated with PLO/laminin. Cells were cultured in final differentiation medium for one week, after which they were treated with 1 µM cytosine arabinoside (AraC, Sigma-Aldrich; Cat#C6645) overnight (~ 17 h) to eliminate any residual proliferating cells. The next day, medium was fully changed to final differentiation medium with or without supplementation of neurotrophic factors (NFs) (i.e., BDNF, CNTF, and IGF-1), with half-changes every other week until initiation of the assay. After 5 weeks of final differentiation, MNs were treated with the candidate compounds MLN4924, piperlongumine, prunetin, QS-11, XMD-885, XMD-892 at the final concentrations of 0.1 µM and 1.0 µM, or with 3 µM riluzole, 0.025 µM edaravone or the combination of 250 µM sodium phenylbutyrate (PB)/10 µM tauroursodeoxycholic acid (TUDCA) for 6 days with medium renewed every second day. After treatment, viability was assessed with an ATP-based chemiluminescent assay (Cell Titer-Glo^®, Promega; Cat#G7570) following the manufacturer’s instructions. The luminescence readings were acquired using a GloMax Microplate Reader (Promega). Raw luminescence values were normalized to the vehicle condition (mean of DMSO-treated wells, medium without NFs) for each cell line. The percentage increase in viability post-treatment reported in text represents the difference in normalized luminescence values between DMSO- and MLN4924-treated cells.

Single-dose functional screen

To assess the effect of compounds on neuronal activity, treatments were performed in MNs differentiated for 5 weeks in 24-well MEA plates (Axion Biosystems) in complete final differentiation medium. MNs were treated with candidate compounds for 6 days at a final concentration of 1.0 µM or with the ALS drugs as described above. After treatment, MEA recordings and analyses were carried out as previously described⁴⁴. Briefly, MNs were incubated in freshly prepared pre-warmed carbonated artificial cerebrospinal fluid (aCSF)¹⁶⁹ for at least 1 h. Spontaneous activity was recorded for 5 min in a Maestro Edge MEA system (Axion Biosystems) and the AxiS v.66,465 software (Axion Biosystems).

Western blot

Cell pellets were collected from MN cultures treated with DMSO or 1.0 µM MLN4924 for 6 days as described above at 2 weeks post-plating of progenitors. Proteins were extracted using ice-cold RIPA buffer (Millipore; Cat#20–188) supplemented with protease inhibitors (Roche; Cat#11697498001) and phosphatase inhibitors (Roche; Cat#04906837001), as previously described⁴⁴. Protein concentrations were determined using the DC Protein Assay (Bio-Rad; Cat#5000111). A total of 20 µg of protein was resolved by 10% or 13% SDS/PAGE and transferred to nitrocellulose membranes using a Trans-Blot Turbo Transfer System (Bio-Rad). After transfer, membranes were blocked in 5% BSA in TBS-T 0.1% for 1 h at room temperature and incubated with primary antibodies in blocking solution overnight at 4 °C. After three washes with TBS-T 0.1%, membranes were incubated with horseradish peroxidase (HRP)-conjugated secondary antibodies (1:10000) in blocking solution for 1 h at room temperature. Blots were developed with Clarity Western ECL Substrate (Bio-Rad; Cat#170–5061) and pictures were acquired with a ChemiDoc MP Imaging System (Bio-Rad). Semiquantitative analysis of western blots was performed with the Image Lab 6.0.1 software (Bio-Rad), using βIII-tubulin as a loading control. The following primary antibodies were used: mouse anti-βIII-tubulin (Millipore; Cat#MAB5564; 1:40000), rabbit anti-TDP-43 (C-terminal) (Proteintech; Cat#12892-1-AP; 1:4000), rabbit anti-NEDD8 (Cell Signaling Technology, Cat#2754; 1:1000), and rabbit anti-NAE1 (Cell Signaling Technology, Cat#14321; 1:2000). The following secondary antibodies were used: HRP-conjugated anti-Mouse (Jackson Immunoresearch, Cat#115-035-003) and HRP-conjugated anti-Rabbit (Jackson Immunoresearch, Cat#111-035-144).

Statistics

Biological replicates were defined as independent differentiations. Hypergeometric testing was performed to compare the lists of genes and miRNAs using the following web app: http://nemates.org/MA/progs/overlap_stats.html. For qPCR experiments, Grubbs’ test was used to determine significant outliers. Statistical analyses were performed with the GraphPad Prism software. Data distribution was assumed to be normal although this was not formally tested. Differences between mutants and control were analyzed using one-way or two-way analysis of variance (ANOVA) tests. Means and standard errors of the mean were used for data presentation. Significance was defined as P < 0.05.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

Conceptualization, S.L., G.M., M.C., and T.M.D.; Methodology, S.L., G.M., L.G., M.C., and T.M.D.; Software, G.J.A. and D.S.; Validation, S.L. and A.N.J.; Formal analysis, S.L., G.M., G.J.A., and D.S.; Investigation, S.L., G.M., A.S., A.N.J., M.J.C.-M., N.A., A.K.F.-F., G.H., M.R.F., A.A.D., and M.C.; Resources, M.C., G.M., S.M.K.F., and T.M.D.; Data curation, S.L.; Writing—original draft, S.L.; Writing—review and editing, S.L., G.M., M.C., and T.M.D.; Visualization, S.L., G.M., M.C., and T.M.D.; Supervision, G.M., M.C., and T.M.D.; Project Administration, S.L., M.C., and T.M.D.; Funding acquisition, T.M.D. All authors have read and agreed to the published version of the manuscript.

Data availability

The data supporting the findings of this study are available from the corresponding authors on reasonable request. The RNA-seq and small RNA-seq datasets generated during the current study are available in the Gene Expression Omnibus (GEO) repository (http://www.ncbi.nlm.nih.gov/geo) under the accession numbers GSE277084 and GSE277085.

Declarations

Competing interests

The authors declare no competing interests.