Combining multiomics and drug perturbation profiles to identify muscle-specific treatments for spinal muscular atrophy

Abstract

Spinal muscular atrophy (SMA) is a neuromuscular disorder caused by loss of survival motor neuron (SMN) protein. While SMN restoration therapies are beneficial, they are not a cure. We aimed to identify potentially novel treatments to alleviate muscle pathology combining transcriptomics, proteomics, and perturbational data sets. This revealed potential drug candidates for repurposing in SMA. One of the candidates, harmine, was further investigated in cell and animal models, improving multiple disease phenotypes, including lifespan, weight, and key molecular networks in skeletal muscle. Our work highlights the potential of multiple and parallel data-driven approaches for the development of potentially novel treatments for use in combination with SMN restoration therapies.

Keywords: Bioinformatics; Drug therapy; Genetic diseases; Muscle Biology; Neuroscience.

Figure 1. Restoration of protein and transcript expression in skeletal muscle of SMA mice following early SMN restoration treatment.

Smn^–/–;SMN2 mice received a facial i.v. injection at P0 and P2 of Pip6a-scrambled or Pip6a-PMO (10 μg/g). The tibialis anterior was harvested from P2 untreated Smn^–/–;SMN2 and WT mice; from P7 untreated, Pip6a-scrambled-treated, and Pip6a-PMO-treated Smn^–/–;SMN2 mice; and from P7 untreated WT mice. (A) Comparison of the ratio of full-length (FL) SMN2 over total SMN2 quantified by qPCR between P7 untreated, Pip6a-scrambled-treated, and Pip6a-PMO–treated Smn^–/–;SMN2 mice. Data are shown as a scatter plot and are represented as mean ± SEM; n = 4 animals per experimental group, 1-way ANOVA followed by a Dunnett’s multiple comparisons test, F ratio (F) = 34.88, degrees of freedom (df) = 11, ***P < 0.001. (B) Heatmap of the transcriptomic and proteomic expression profiles measured by the Pearson correlation between each pair of samples (after the removal of the first principal component). (C) First 2 principal components based on transcriptomic profiles of P7 untreated WT mice, untreated Smn^–/–;SMN2 mice, Pip6a-PMO–treated Smn^–/–;SMN2 mice, and Pip6a-scrambled Smn^–/–;SMN2 mice. (D) First 2 principal components based on proteomic profiles of P7 untreated WT mice, untreated Smn^–/–;SMN2 mice, Pip6a-PMO–treated Smn^–/–;SMN2 mice, and Pip6a-scrambled Smn^–/–;SMN2 mice.

Figure 2. Identification of disease signal reversed by treatment with Pip6a-PMO by removing the effect of Pip6a scrambled at transcriptomic and proteomic levels.

(A) Venn diagrams show the number of transcripts (top) and proteins (bottom) differentially expressed (DE) between untreated Smn^–/–;SMN2 and untreated WT mice, reversed by treatment with Pip6a-PMO and not DE between Pip6a-scrambled–treated Smn^–/–;SMN2 mice and untreated WT animals. Filtered signatures were named according to the increase (up) or decrease (down) expression in untreated Smn^–/–;SMN2 mice compared with untreated WT animals and are highlighted in the green area of the Venn diagrams. (B) Set of enriched gene ontology (GO) biological processes that show similarity across comparisons. GO enrichment analysis was performed separately for transcripts and proteins that were DE between untreated Smn^–/–;SMN2 and untreated WT mice (blue), DE between Pip6a-PMO-treated Smn^–/–;SMN2 and untreated Smn^–/–;SMN2 mice (purple), and part of the filtered signatures described in A (green).

Figure 3. Harmine target genes, as predicted by CMap analyses, are aberrantly expressed in SMA muscle.

(A) qPCR analysis of genes predicted to be significantly downregulated (Snrnp27, Gls, Aspm, and Mcm2) in the TA of untreated P7 SMA Smn^–/–;SMN2 and WT mice. Data are shown as a scatter plot and are represented as mean ± SEM; n = 4 animals per experimental group, unpaired t test, df = 6 for all, P = 0.041 (Snrnp27), P = 0.0019 (Gls), P = 0.0001 (Aspm), P < 0.0001 (Mcm2). (B) qPCR analysis of genes predicted to be upregulated (Clpx, Ppm1b, Tob2, and Cdkn1a) in the TA of untreated P7 SMA Smn^–/–;SMN2 and WT mice. Data are shown as a scatter plot and are represented as mean ± SEM; n = 4 animals per experimental group, unpaired t test, df = 6 for all except Ppm1b, where df = 5; P < 0.0001 (Clpx), P = 0.0076 (Ppm1b), P = 0.0012 (Tob2), P < 0.0001 (Cdkn1a).

Figure 4. Harmine, as predicted by CMap analyses, is able to reverse the expression of genes significantly downregulated in SMA muscle in several cellular models.

(A–D) C2C12, NSC-34, SMA patient fibroblasts, and control fibroblasts were treated with 25, 35, or 50 μM of harmine for 48 hours. Expression of Snrnp27 (A), Gls (B), Aspm (C), and Mcm2 (D) was assessed by qPCR and compared with untreated cells. Data are shown as a scatter plot and are represented as mean ± SEM; n = 3 independent wells, 2-way ANOVA followed by uncorrected Fisher’s least significant difference (LSD), F = 20.20 (Snrnp27), F = 90.95 (Gls), F = 14.16 (Aspm), F = 42.61 (Mcm2), df = 32 for all, *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

Figure 5. Harmine, as predicted by CMap analyses, is able to reverse the expression of genes significantly upregulated in SMA muscle in several cellular models.

(A–D) C2C12s, NSC-34s, SMA patient fibroblasts, and control fibroblasts were treated with 25, 35, or 50 μM of harmine for 48 hours. Expression of Clpx (A), Ppm1b (B), Tob2 (C), and Cdkn1a (D) was assessed by qPCR and compared with untreated cells. Data are shown as a scatter plot and are represented as mean ± SEM; n = 3 independent wells, 2-way ANOVA followed by uncorrected Fisher’s LSD, F = 182 (Clpx), F = 38.49 (Ppm1b), F = 78.17 (Tob2), F = 18.36 (Cdkn1a), df = 32 for all, *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

Figure 6. Administration of harmine to SMA mice partially restores the expression of target genes, as predicted by CMap analyses.

All treated animals received a daily dose of harmine (10 mg/kg, diluted in 0.9% saline) by gavage starting at P0. (A) qPCR analysis of Snrnp27, Gls, Aspm, and Mcm2 in triceps of P7 untreated and harmine-treated Smn^–/–;SMN2 SMA mice and Smn^+/–;SMN2 control littermates. Data are shown as a scatter plot and are represented as mean ± SEM, n = 4 animals per experimental group except for harmine-treated Smn^+/–;SMN2 where n = 3, 2-way ANOVA followed by a Sidak’s multiple comparisons test, F = 25.77 (Snrnp27), F = 1.103 (Gls), F = 0.5143 (Aspm), F = 0.3992 (Mcm2), df = 11 for all, *P < 0.05, **P < 0.01. (B) qPCR analysis of Clpx, Ppm1b, Tob2, and Cdkn1a in triceps of P7 untreated and harmine-treated Smn^–/–;SMN2 SMA mice and Smn^+/–;SMN2 control littermates. Data are shown as a scatter plot and are represented as mean ± SD; n = 4 animals per experimental group except for harmine-treated Smn^+/–;SMN2 where n = 3, 2-way ANOVA followed by a Sidak’s multiple comparisons test, F = 0.4275 (Clpx), F = 0.006960 (Ppm1b), F = 8.167 (Tob2), F = 1.195 (Cdkn1a), df = 11 for all, **P < 0.01.

Figure 7. Administration of harmine to SMA mice improves weight and survival.

All treated animals received a daily dose of harmine (10 mg/kg, diluted in 0.9% saline) by gavage starting at P0. (A) Survival curves of untreated and harmine-treated Smn^–/–;SMN2 mice. Kaplan-Meier survival curve is shown, with n = 10 for untreated Smn^–/–;SMN2 mice, n = 11 for harmine-treated Smn^–/–;SMN2 mice, Log-rank (Mantel-Cox) test, *P = 0.0211. (B) Daily weights of untreated and harmine-treated Smn^–/–;SMN2 mice. Data are represented as mean ± SEM; n = 10 for untreated Smn^–/–;SMN2 mice, n = 11 for harmine-treated Smn^–/–;SMN2 mice, 2-way ANOVA followed by a Sidak’s multiple comparisons test, F = 95.70, df = 202, **P < 0.01, ****P < 0.0001. (C) Daily weights of untreated and harmine-treated Smn^+/–;SMN2 mice. Data are represented as mean ± SEM; n = 13 for untreated Smn^+/–;SMN2 mice, n = 15 for harmine-treated Smn^+/–;SMN2 mice, 2-way ANOVA followed by a Sidak’s multiple comparisons test, F = 2.897, df = 398. (D) Survival curves of untreated and harmine-treated Smn^2B/– mice. Kaplan-Meier survival curves are shown, with n = 9 for untreated Smn^2B/– mice, n = 7 for harmine-treated Smn^2B/– mice, log-rank (Mantel-Cox) test, *P = 0.0221. (E) Daily weights of untreated and harmine-treated Smn^2B/– mice. Data are represented as mean ± SEM, n = 9 for untreated Smn^2B/– mice, n = 7 for harmine-treated Smn^2B/– mice, 2-way ANOVA followed by a Sidak’s multiple comparisons test, F = 96.25, df = 287, *P < 0.05, ****P < 0.0001. (F) Daily weights of untreated and harmine-treated Smn^2B/+ mice. Data are represented as mean ± SEM, n = 13 for untreated Smn^2B/+ mice, n = 8 for harmine-treated Smn^2B/+ mice, 2-way ANOVA followed by a Sidak’s multiple comparisons test, F = 206.3, df = 399, *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

Figure 8. Administration of harmine to SMA mice improves neuromuscular phenotypes.

All treated animals received a daily dose of harmine (10 mg/kg, diluted in 0.9% saline) by gavage starting at P0. (A) Relative frequency of myofiber sizes in P7 untreated and harmine-treated Smn^–/–;SMN2 and Smn^+/–;SMN2 mice. Data are shown as percentages, with n = 3 animals per experimental group and > 400 myofibers per experimental group. (B) Western blot and quantification of GLT-1/vinculin expression in the spinal cord of P7 untreated and harmine-treated Smn^–/–;SMN2 and Smn^+/–;SMN2 mice. Data are shown as a scatter plot and are represented as mean ± SEM; n = 3 for untreated and harmine-treated Smn^+/–;SMN2 mice, n = 4 for untreated and harmine-treated Smn^–/–;SMN2 mice, 2-way ANOVA followed by a Sidak’s multiple comparisons test, F = 35.01, df = 10, ****P < 0.0001. (C) Number of motor neuron cell bodies per ventral horn area in the spinal cord of P7 untreated and harmine-treated Smn^–/–;SMN2 and Smn^+/–;SMN2 mice. Data are represented as mean ± SEM n = 3 for untreated Smn^+/–;SMN2 mice, n = 4 for harmine-treated Smn^–/–;SMN2 and Smn^+/–;SMN2 mice, n = 5 for untreated Smn^–/–;SMN2 mice, 2-way ANOVA followed by a Tukey’s multiple comparisons test, F = 4.617, df = 12, *P < 0.05, **P < 0.01. Images are representative spinal cord ventral horn areas of untreated and harmine-treated Smn^–/–;SMN2 mice. Total original magnification, ×20.

Figure 9. RNA sequencing and pathway analysis reveals full rescue of 20% of dysregulated genes in SMA muscle following harmine administration.

All treated animals received a daily dose of harmine (10 mg/kg, diluted in 0.9% saline) by gavage starting at P0. TAs were harvested at P7 from untreated and harmine-treated Smn^–/–;SMN2 mice and WT animals and were processed for RNA sequencing. (A) Venn diagram representation of the differentially expressed (DE) genes based on the negative binomial distribution (DESeq2) in untreated Smn^–/–;SMN2 mice versus untreated WT mice (blue), harmine-treated Smn^–/–;SMN2 mice versus untreated Smn^–/–;SMN2 mice (purple), and untreated WT mice versus harmine-treated WT mice (orange). (B) Venn diagram representation of the DE genes based on the negative binomial distribution (DESeq2) in untreated Smn^–/–;SMN2 mice versus untreated WT mice (blue), harmine-treated Smn^–/–;SMN2 mice versus untreated Smn^–/–;SMN2 mice (purple), and harmine-treated Smn^–/–;SMN2 mice versus untreated WT mice (green). (C) Gene ontology (GO) biological processes enriched in genes DE in untreated Smn^–/–;SMN2 mice versus untreated WT mice (blue), in harmine-treated Smn^–/–;SMN2 mice versus untreated Smn^–/–;SMN2 mice (purple), in untreated WT mice versus harmine-treated WT mice (orange), and in harmine-treated Smn^–/–;SMN2 mice versus untreated WT (green). –LogP values for the enrichment are reported.

Figure 10. Identification of molecular effectors involved in harmine activity in SMA muscle.

(A) A gene functional network was built extracting gene interactions from a Phenotypic Linkage Network (45) for the top 500 most differentially expressed (DE) genes (ordered by adjusted P value) in untreated Smn^–/–;SMN2 mice versus untreated WT mice. Genes are represented as nodes and are colored by direction expression change in untreated Smn^–/–;SMN2 mice versus untreated WT mice (left) and by direction of expression change in harmine-treated Smn^–/–;SMN2 mice versus untreated Smn^–/–;SMN2 mice (right). Gray nodes correspond to genes that are DE in the disease model (untreated Smn^–/–;SMN2 mice versus untreated WT) mice but have not been restored by harmine treatment. (B) Top MGI enriched phenotypes for the 4 identified modules in the network (shown in A) that show reversed expression profile after harmine treatment. –LogP values for the enrichment are reported. (C) Ingenuity Pathway Analysis (IPA) tool was used to identify upstream regulators of the top 500 most differentially expressed genes in untreated Smn^–/–;SMN2 mice versus untreated WT mice (shown in A). For each of the top 50 most significant upstream regulators shown (ordered on enrichment P values from left [most significant] to right [less significant]), we calculated the proportions of target genes within each of the 6 modules that are predicted to be regulated by the corresponding upstream regulator. Represented is a selected reduced list of regulators based on high proportion of target genes from Module 1 (muscle phenotypes) and Module 2 (glucose and lipid metabolism).