Exploring Motor Neuron Diseases Using iPSC Platforms

Abstract

The degeneration of motor neurons is a pathological hallmark of motor neuron diseases (MNDs), but emerging evidence suggests that neuronal vulnerability extends well beyond this cell subtype. The ability to assess motor function in the clinic is limited to physical examination, electrophysiological measures, and tissue-based or neuroimaging techniques which lack the resolution to accurately assess neuronal dysfunction as the disease progresses. Spinal muscular atrophy (SMA), spinal and bulbar muscular atrophy (SBMA), hereditary spastic paraplegia (HSP), and amyotrophic lateral sclerosis (ALS) are all MNDs with devastating clinical outcomes that contribute significantly to disease burden as patients are no longer able to carry out normal activities of daily living. The critical need to accurately assess the cause and progression of motor neuron dysfunction, especially in the early stages of those diseases, has motivated the use of human iPSC-derived motor neurons (hiPSC-MN) to study the neurobiological mechanisms underlying disease pathogenesis and to generate platforms for therapeutic discovery and testing. As our understanding of MNDs has grown, so too has our need to develop more complex in vitro models which include hiPSC-MN co-cultured with relevant non-neuronal cells in 2D as well as in 3D organoid and spheroid systems. These more complex hiPSC-derived culture systems have led to the implementation of new technologies, including microfluidics, multielectrode array, and machine learning which offer novel insights into the functional correlates of these emerging model systems.

Graphical Abstract

Graphical Abstract.

Significance Statement.

The critical need to understand the cause and progression of motor neuron diseases beyond clinical phenotyping, has motivated the development of innovative in vitro model systems that use human induced pluripotent stem cells (hiPSCs). The ability to differentiate hiPSCs into neuronal and non-neuronal cell types in 2D and 3D cultures, together with functional assays and technologies like multielectrode array, optogenetics and automated machine learning, provides unprecedented insight into the pathophysiological mechanisms that underlie MNDs.

Introduction

Spinal muscular atrophy (SMA), spinal and bulbar muscular atrophy (SBMA), hereditary spastic paraplegia (HSP), and amyotrophic lateral sclerosis (ALS) are genetically, clinically, and epidemiologically diverse in their presentations but all share the pathological feature of motor neuron (MN) loss. While often used as predictive models for clinical outcomes, animal models for these diseases also share significant limitations of not being able to capture the full neurodegenerative cascade by manifesting only a modest subset of phenotypes or because they lack certain phenotypes all together. This is particularly true in mouse models for which corticospinal motor neuron (CSMN) pathology is a prominent feature of human disease since CSMNs comprise such a small percentage of neurons in the motor cortex and thus are difficult targets to manipulate, often leading to minor or no observable phenotypes.¹

Our ability to generate animal models for motor neuron diseases (MNDs) that are more representative of the full spectrum of phenotypic variation is often restricted by the technical limitations of genetic manipulation or our current lack of understanding of the genetic basis of the disorder. While SMA, SBMA, and HSP are diseases that display clear inheritance patterns of specific gene mutations (SMA and SBMA) or can be identified genetically from known disease-causing genes (HSP), ALS has been more difficult to characterize since only about 10% of cases arise from known gene mutations. This means that animal models for ALS are not representative of the much larger proportion of cases that arise sporadically without a known inheritance and contribute to a significant problem in translatability of preclinical modeling.² In diseases like SMA where the genetic basis is known, while the pathology in the central nervous system (CNS) is highly reproducible and generates distinct phenotypes, the neuron-specific contributions to the disease are difficult to dissect as there is multisystem morbidity which complicates the analysis of motor deficits and reduces the life span of the animal such that MN loss can only be observed at early developmental time points.³ In addition to the multisystem involvement seen in SMA, SBMA is uniquely challenging to model because it progresses much more slowly compared to ALS or HSP and results from exonic expansions, which have not been identified in mice.

The constraints of animal modeling of MNDs have incentivized efforts to generate human MNs in vitro that can recapitulate specific phenotypic and genetic characteristics of human disease. These human-induced pluripotent stem cell (hiPSC) platforms benefit not only neuroscientists with unique insight into the neuronal contributions to disease pathogenesis, but to clinicians alike as potential new methods for personalized medicine are developed through the screening of therapeutics. This is particularly important in ALS since hiPSCs can be generated from patients with sporadic forms of the disease and thus, in theory, can provide an understanding of disease mechanisms and possible therapeutic strategies for a larger proportion of ALS cases. Advancements in hiPSC culturing methods have led to an eruption of hiPSC co-culture systems which have allowed us to broaden our understanding of MNDs to one that highlights the role of non-neuronal cells, including astrocytes, oligodendrocytes, and microglia, in the anatomical and temporal progression of the disease. This understanding has been further advanced through the generation of 3D co-culture systems which more closely reflect relevant microenvironments that exist in vivo and the crosstalk that exists between neuronal and non-neuronal cell types therein. Of particular relevance to modeling, MNDs have been the development of co-culture systems that combine hiPSC-derived human skeletal muscle with hiPSC-derived spinal MN in order to study events that take place at the neuromuscular junction (NMJ).

Alongside the generation of a greater variety of hiPSC models has been the development of novel technologies, including multielectrode array (MEA) electrophysiology, optogenetics, and fully automated imaging that combines robotics and fluorescence microscopy to generate readouts of cell survival and hallmarks of pathology. Together, these offer approaches that have the potential to significantly increase the efficiency at which new therapeutics are screened in hiPSC-derived cultures. Further technological advancements have led to the implementation of machine learning and artificial intelligence-driven prediction models as strategies that could potentially identify phenotypic differences between disease and control cells in response to treatment.

Generating iPSC Across the Spectrum of Motor Neuron Diseases

SMA, SBMA, HSP, and ALS are all considered motor neuronopathies as they share dysfunction of either CSMN, spinal MNs, or both (Fig. 1). Human iPSC-derived spinal alpha motor neurons (hiPSC-MN) have been generated by numerous protocols and validated by way of appropriate morphological characteristics, protein markers, transcription profiles, and electrophysiological measures. While spinal MNs have been cultured to a relatively high purity of over 90% as identified by homeobox gene 9 (Hb9) and choline acetyltransferase (ChAT) immunostaining,⁴ CSMNs have proved more challenging to identify and enrich in culture and are typically seen in vitro at a much lower frequency along with other cortical neuron subtypes. Markers, including BCL11B (CTIP2),⁵ FEZF2,⁶ and L1CAM,⁷ have been used as relevant identifiers of CSMN fate, however, specific CSMN subsets relevant to motor neuronopathies with cortical pathology like ALS and HSP have not been studied to a significant extent in culture.

Figure 1.

Motor neuronopathies and their corresponding levels of MN damage. Spinal muscular Atrophy (SMA) primarily affects spinal MNs and their subsequent innervation of muscle. Human iPSCs have been created to model SMA type 1. Spinal and bulbar muscular atrophy (SBMA) also affects spinal MNs with sparing of CSMNs. CAG trinucleotide expansion repeats in the androgen receptor produce the phenotype. Hereditary spastic paraplegia (HSP) primarily affects CSMN with iPSC developed from a few of the more than 80 genetic subtypes. Amyotrophic lateral sclerosis (ALS) variably affects both CSMN and spinal MNs producing wide ranges in phenotypic presentation. While hiPSC-derived spinal MNs have been made for numerous FALS and SALS subtypes, far fewer models of CSMN biology exist. Green (normal MN), Red (affected MN subtype). Abbreviations: CSMNs, corticospinal motor neurons; FALS, familial amyotrophic lateral sclerosis; hiPSC, human induced pluripotent stem cells; iPSC, induced pluripotent stem cells; MNs, motor neurons; SALS, sporadic amyotrophic lateral sclerosis.

SMA is an autosomal recessive disorder characterized by loss of function mutations in the SMN1 gene⁸ and results in the loss of spinal MNs with subsequent muscle atrophy, weakness in the limb trunk and bulbar muscles, and in its most severe form (SMA1), death during infancy.⁹ In humans, 2 nearly identical copies of the SMN gene are located on chromosome 5q: telomeric SMN1 and centromeric SMN2. SMA is caused by a homozygous mutation of the survival motor neuron 1 (SMN1) gene, which results in reduced levels of the functional SMN protein. A mutation that causes a C to T substitution within the coding region of exon 7 of SMN2 leads to alternative splicing and the subsequent removal of exon 7 from mature SMN2 mRNAs. The truncated SMN2 mRNA product is translated into an unstable protein that is susceptible to degradation.¹⁰ Therefore, the majority of current SMA therapeutic approaches focus on increasing exon 7 retention in SMN2. In 2009, Ebert et al generated the first set of hiPSC-MN from fibroblasts that had been taken from SMA patients. Importantly, the reprogrammed cells showed reduced levels of full-length SMN transcripts, reduced MN survival, and a lack of nuclear gems (sites of SMN localization). These findings were consistent with features that had been previously identified in tissues from SMA patients.¹¹ Since this initial study, numerous SMA iPSC lines have been created and used for studying disease mechanisms and screening potential therapeutics.¹²

SBMA is an X-linked disorder caused by a CAG trinucleotide repeat expansion in the androgen receptor (AR) gene with a corresponding increase in the length of a polyglutamine tract in the AR protein.¹³ This results in slowly progressive spinal MN degeneration with muscle weakness, atrophy, and fasciculations. Additional features related to AR insensitivity result in gynecomastia and reduced fertility.¹⁴ In 2013, Nihei et al generated SBMA-iPSCs from a patient with the disease. Importantly, they were able to demonstrate that the CAG trinucleotide repeat was stable during reprogramming, long-term passage, and differentiation. They were also able to demonstrate the aggregation of the AR in iPSC-derived neurons from this patient.¹⁵ Additional lines of human SBMA iPSC¹⁶ or iPSC-MN were generated which similarly displayed reduced AR levels and reduced histone deacetylase 6 (HDAC6) expression,¹⁷ FAM135B expression,¹⁸ and alterations in synapse-related gene sets.¹⁹ The use of SBMA iPSC-neural progenitor cells (NPCs) led to the observation that a dysregulation of transcription factor EB (TFEB) mediated by an interaction with the poly-Q expanded AR, blocks the fusion of autophagosomes with lysosomes resulting in autophagic dysfunction.²⁰

HSPs are disorders with slowly progressive lower extremity spasticity as their most prominent features, although additional symptoms can be present. This spasticity is due to a length-dependent axonopathy of CSMNs.²¹ There are more than 80 distinct genetic loci, most of which gene products play roles in axon development and maintenance.²¹ Human iPSCs were first generated from patients with the most common form of HSP, SPG4. The SPG4 neurons displayed a significant increase in axonal swellings, which stained strongly for mitochondria and tau protein, indicating the accumulation of axonal transport cargoes. In addition, mitochondrial transport was decreased in SPG4 neurons, revealing that these patient iPSC-derived neurons recapitulated disease-specific axonal phenotypes.²² Human iPSCs have also been derived from additional HSP subtypes: SPG47,²³ SPG3A,²⁴ and SPG11.²⁵

ALS has a much more heterogeneous presentation with ~10% of patients having familial ALS (FALS) (mostly autosomal dominant) while the remaining population is classified as having sporadic ALS (SALS) without known inheritance. Presentations can be largely characterized by bulbar onset disease manifested by dysarthria and dysphagia or spinal onset, with limb weakness as the presenting feature. Importantly, a combination of CSMN and spinal MN dysfunction in varying degrees of severity typically evolves following the initial presentation. Human iPSCs were first derived from a patient with FALS, containing a known mutation in the superoxide dismutase 1 (SOD1) gene.²⁶ Since then, hiPSCs have been derived from several other inherited forms of ALS, including C9orf72, TDP-43, FUS, VAPB, and VCP as well as others that likely have ALS relevance (reviewed by Hawrot et al²⁷). Human iPSCs have also been reprogrammed from SALS patients.²⁸ Although SALS accounts for the vast majority of patients with the disease, using a rigorously defined protocol of single-cell RNA sequencing analysis of hiPSC-MN from healthy, FALS, SALS, and even genome-edited iPSC lines across multiple patients, batches, and platforms, investigators have been able to find reproducible ALS signatures which suggest that there are convergent features, even among non-hereditary forms of the disease. Given that ALS is a neurodegenerative disease that manifests in adulthood, an unexpected finding from hiPSC-MN models was that compared to postmortem ALS spinal MNs, hiPSC-MN showed distinct transcriptional patterns suggesting that disruption by ALS conditions occurs as early as embryonic development.²⁹

Emerging iPSC Platforms and Technologies for MND Modeling

Co-Culture of Cell Subtypes

Although MNDs are characterized neurobiologically and clinically by the degeneration of MNs, it is widely accepted that multiple non-neuronal cells, including astrocytes, microglia, and oligodendrocytes, contribute to the progression of each of the MNDs discussed in this review. As our understanding of these non-neuronal contributions to MNDs increases, more complex iPSC models have been used which characterize interactions between MNs and non-neuronal cells so as to more accurately model the disease microenvironment that exists in vivo. These more sophisticated co-culture systems provide the opportunity to not only investigate interactions between different hiPSC-derived disease cell types, but also non-cell-autonomous forms of MN death through combined co-culturing of hiPSC-derived non-neuronal disease cells with healthy hiPSC-derived MN (Fig. 2).

Figure 2.

iPSC-based platforms for motor neuron disease modeling. (A) Healthy control hiPSC-MN (green) can be cultured either alone or with non-neuronal cell types (astrocytes in this case) carrying disease mutations (orange) to establish the significance of non-cell-autonomous effects on cell survival. Neurite outgrown from MNs can also be measured anatomically and temporally using microfluidic chambers and longitudinal imaging. (B) Organoids and spheroids are being used to study either individual anatomical regions (cerebral) of MND pathology or combined to create entire pathways involved in disorders like ALS (corticospinal-muscle). (C) Increasingly sophisticated methods for manipulating (optogenetics) and analyzing electrophysiological of hiPSC-MN (MEA) or force-generating properties of hiPSC-muscle (micropillars) are being used. (D) Automated cell counting and machine learning strategies allow for longitudinal and unbiased analyses of cell survival, morphology, and molecular pathway investigation. Abbreviations: ALS, amyotrophic lateral sclerosis; hiPSC-MN, human iPSC-derived motor neurons; iPSC, induced pluripotent stem cells; MND, motor neuron disease; MNs, motor neurons.

SMA

In SMA, it has been noted that there are pathological changes in non-neuronal cells in patients as well as in animal models of the disease.³⁰ Astrocytes from SMA hiPSCs have abnormal calcium regulation and a reduced production of glial cell line-derived neurotrophic factor (GDNF) suggesting that these astrocytes may lack the capacity to support MN survival.³¹ More recent studies have directly implicated SMA hiPSC-A in MN loss through the upregulation of miR-146.³² Additional studies have also suggested that hiPSC-derived oligodendrocyte lineages in SMA (hiPSC-OL) may also show defects in myelin basic protein (MBP) expression.³³ However, the influences of SMA hiPSC-OL have not been studied in the context of their influence on MN pathology.

SBMA

Although poly-Q nuclear inclusions have been seen in astrocytes of an SBMA mouse model,³⁴ this was not accompanied by astrocytosis, and to date, we are not aware of the use of hiPSC-A, hiPSC-OL, or hiPSC-microglia in the study of SBMA.

HSP

Mutations in the ATL1 gene in HSP (SPG3A) were found to impair cholesterol homeostasis in SPG3A cortical projection neurons. Because glial cells are a major source of cholesterol, conditioned media from control astroglial cells added to hiPSC-neuronal cultures were able to reduce axonal swellings in SPG3A cortical neurons. Furthermore, SPG3A hiPSC-astrocytes were noted to have lipid droplet defects, emphasizing that there may be neuronal/glial crosstalk for lipid metabolism in the pathobiology of SPG3A.³⁵ A family with HSP containing a mutation of connexin 47 (Cx47) in oligodendrocytes has been described³⁶ suggesting that oligodendroglial influences could result in HSP-relevant pathologies, but this has not been studied using hiPSC co-culture platforms.

ALS

The importance of non-MN cell types, including astrocytes and microglia, has been described in the context of ALS. This was originally described as an in vivo phenomenon implicating astrocytes³⁷ and microglia³⁸ as drivers of disease progression in mSOD1 mice. Astrocyte-mediated toxicity to MNs in ALS has been recapitulated with hiPSCs from both familial as well as SALS,^39-42 the latter being of particular importance since mouse models for SALS do not exist. Non-cell-autonomous forms of MN death in ALS have also been investigated using a co-culture system consisting of iPSC-derived oligodendrocytes⁴³ with hiPSC-MN and has allowed for the dissection of the specific contributions of these non-neuronal cell types to disease progression. Beyond toxicity, the electrophysiological study of iPSC spinal cord-specific astrocytes with iPSC-MNs has also allowed for opportunities to leverage MEA studies in non-cell-autonomous modeling.⁴⁴

Modeling the Motor Unit

While much of the focus has been on MN loss as a process occurring primarily in the soma, there is now abundant evidence suggesting that among the earliest features of these motor neuronopathies is the “dying back” of axons from the NMJ.⁴⁵ Since loss of strength and muscle atrophy are products of denervated muscle, the capacity to recapitulate the entire motor unit using iPSC technologies has gained traction. Several models of the motor unit in 2D platforms have been generated,^45,46 including a recent elegant study which constructed a motor unit consisting of MNs coupled to skeletal muscles interacting via the NMJ within a microfluidic device. Microfluidic devices allow for the isolation of somatic and axonal microenvironments by way of a minute difference in volume between the 2 compartments.⁴⁵ The successful recapitulation of the anatomical properties of the motor unit across 2 fluidically isolated compartments has allowed for more sophisticated studies of properties particularly relevant to MNDs like axon growth and regeneration, especially as it relates to the effect of novel therapies that specifically target events at the NMJ.

The replication of the motor unit within microfluidic devices has been used to investigate neurite outgrowth and NMJ pathology in ALS. Specifically, ALS hiPSCs harboring fused in sarcoma (FUS) mutations were studied following treatment with the HDAC6 inhibitor Tubastatin A which was found to improve FUS iPSC-MN neurite outgrowth and NMJ formation.⁴⁷ In a separate study, investigators used a microfluidic device to create what they coined “ALS-on-a-chip-technology” consisting of 3D bundles of skeletal muscle in combination with iPSC-MN derived from a SALS patient, and demonstrated that stimulation of ALS hiPSC-MN generated fewer muscle contractions. This deficit, as well as the MN degeneration which was observed to accompany it, was rescued with the combination treatment of rapamycin and bosutinib.⁴⁸ Despite the impressive capacity for recapitulating the motor unit, one shortcoming of these strategies remains myelinating axons to more fully recapitulate an in vivo paradigm.

Organoids and Spheroids

Alongside these advancements in 2D co-culture systems has been the development of 3D cerebral and spinal organoids which, in combination with muscle spheroids, can be used to model cell-to-cell interactions within a cortical spinal circuit. ALS and HSP both have CSMN dysfunction as part of their pathobiology while most in vivo modeling of MNDs has focused on spinal MN pathology, as noted above. The most common mouse model used to study the potential translatability of ALS therapeutics has abundant spinal MN pathology with little CSMN pathology, a factor that potentially has contributed to paucity of therapeutic successes in ALS patients.² It is possible that the dearth of effective therapeutics for diseases like HSP or ALS may be the result of inadequate study and development of compounds that target CSMN dysfunction.

Cerebral organoids have been developed to primarily study diseases of development but could potentially have been used in modeling CSMN degeneration in MNDs as well.⁴⁹ An air-liquid interface-cerebral organoid (ALI-CO) developed by Giandomenico et al included CSMN subtypes, identified by single-cell RNA sequencing, to express subcortical projection neurons identified by CTIP2 and FEZF2. Using a unique co-culture method, investigators cultured the ALI-CO with the embryonic mouse spinal cord (with associated peripheral nerves and paraspinous muscles intact). Output from the ALI-CO through subcerebral projections onto the mouse spinal cord was able to elicit muscle contractions from mouse paraspinous muscle.⁵⁰ This technique offers a unique perspective on combining human and animal modeling to generate more complex systems that, at least at this time, would be too difficult to developmentally recapitulate from single cells. Using a combination of 3D in vitro culture and in vivo transplantation, investigators first cultured human embryonic stem cell (hESC)-derived cerebral organoids and then transplanted them into the cerebral cortices of mice to show that they were able to extend axons into the corticospinal tract of adult animals. While these transplanted grafts became vascularized over time, they were also prone to overgrowth. Transplantation into the prefrontal cortex of monkeys also showed subcerebral projections. This approach, despite the potential ethical challenges, provides evidence that this combinatorial approach could be useful for reconstructing neural circuits in vivo.⁵¹

Spinal cord organoids from hiPSC representing both ventral, intermediate, and dorsal patterns have been developed using a serum-free, embryoid body-like aggregation approach (SFEBq).⁵² Using a protocol for neuralization and caudalization of hiPSC from patients with SMA, Hor et al were able to develop a spinal organoid that showed a typical apical to basal patterning and diverse cell populations, including interneurons and MNs that extended neurites and developed NMJ. However, dorsal cell types which would include sensory neurons were absent. Some features of SMA1 were loosely recapitulated in these organoids, including the normal developmental profile of the appropriate number of MN which reached a plateau followed by degeneration over time.⁵³ Spinal cord organoids showing morphological and electrophysiological properties of MNs and interneurons among other neuronal subtypes have also been generated by other groups using a variety of differentiation protocols and culture platforms.^52,54-57

In an elegant and complex study by Andersen et al, hiPSC-derived spheroids resembling the hindbrain/cervical spinal cord were combined with cortical spheroids and then assembled with human skeletal muscle spheroids in order to generate an “assembloid” that recapitulates a corticospinal-muscle pathway. This corticospinal-muscle assembloid would be useful in the investigation of not only spinal MN predominant disorders like SMA and SBMA, but also holds significant promise as a model for ALS where both CSMN and spinal MN pathologies exist. Importantly, not only were the cell subtypes connected morphologically, but stimulation of cortical neurons using optogenetic strategies resulted in skeletal muscle contraction⁵⁸ suggesting the formation of a functional circuit.

Among the challenges of organoid differentiation remain the scarcity or absence of non-neuronal cell types, including microglia, pericytes, and vascular components.^59-61 This is particularly notable given the known contributions of non-neuronal subtypes to the progression of ALS. Other limitations of organoid development include central necrosis of the organoid as its size expands. Nevertheless, investigators interested in studying more complex network interactions between neuronal and non-neuronal cell types as they pertain to MND will be buoyed by the impressive pace of discovery and manipulation of these iPSC model systems.

Physiological Correlates: Optogenetics, Micropillars, and Multielectrode Array

The introduction of channelrhodopsins (ChRs) into hiPSC-MN is an optogenetic strategy for the light-induced activation of these neuronal populations grown on micropillar scaffolds, followed by the electrophysiological and biophysical measurement of muscle contraction, thus demonstrating the formation of not only an anatomically relevant model of the neuromuscular circuit but a functionally relevant one as well.^62-64 These models have been applied to SMA and ALS,^48,65 notably by Osaki et al, who injected skeletal myoblasts into ChR-containing MN spheroids to generate micropillar structures from skeletal muscle that could be stimulated to contract with 488 nm light.⁶⁴

The variety of 2D and 3D iPSC culture systems that have been developed to model MNDs has grown in proportion with technological advancements designed to measure the physiological correlates of these more complex cultures, often in a high-throughput manner. This has been highlighted through the use of MEA technology, which enables electrophysiological recordings to be taken from multiple populations of neurons in parallel, in addition to providing electrophysiological readouts of network activity.^44,66,67

Automated Imaging and Machine Learning

Automated imaging using a robotic microscope platform to measure the presence of fluorescently labeled neurons as well as proteins⁶⁸ has been used to evaluate iPSC-MN survival⁶⁹ and allows for high-content screening and analysis. In a proof-of-concept study using hiPSC-MN morphology as defined by soma size and neurite outgrowth, investigators demonstrated that an AI-based prediction model could potentially be applied to differentiate between ALS and healthy controls. While significant differences to allow for prediction between ALS and controls were not observed, challenges, including rigorous comparisons of passage numbers, strategies for reproducible differentiation, and an increase in sample sizes could help to improve the utility of these strategies.⁷⁰ Fujimori et al used automated imaging to examine hiPSC-MN neurites, protein aggregates, stress granules, and cytotoxicity. Using a larger sampling of hiPSC from ALS and healthy controls, they were able to identify ropinirole as a potential therapeutic agent among 1232 screened compounds.⁷¹ Similar technologies have been applied in the pathological examination of HSP-hiPSC neurites.⁷²

While automated imaging has been successfully applied in 2D hiPSC cultures of MND, automation in the generation and analysis of cerebral organoids has been challenged by batch variability,⁷³ the scalability of performing studies in plates that are limited in the number of wells, and reproducibility of data between research groups. Imaging 3D organoids is challenging because of the complexity of these cultures related to deposits of proteins, carbohydrates, and lipids.⁷⁴ Machine learning using cerebral organoids is still in its early stages and to our knowledge has not been applied in MND investigation.⁷⁵

To understand the potential power of using hiPSC technologies, correlations between their properties and other human CNS tissues are warranted. In comparing ALS brain and spinal cord tissues with ALS hiPSC-MNs, investigators used machine learning to identify novel RNA-binding proteins—both validating some that were already known as well as several others that had not yet been associated with ALS.⁷⁶ Machine learning, as applied to neurodegenerative diseases and previously reviewed by Myszczynska et al, may encompass clinical, neuroimaging, pathological, and other biofluid/biomarker datasets, among which could include iPSCs.⁷⁷ In ALS, the Answer ALS program plans to combine many of these features along with analyses of hiPSC-MNs using a multi-omics approach in an attempt to identify specific subgroups of ALS (www.answerals.org)

Drug Screening in iPSC and Translational Opportunities for MND

The original promise following the development of iPSC reprogramming technologies was the development and testing of new compounds that would be more effective in the treating MNDs. This platform has the potential to help satisfy the critical need for novel treatments for MNDs by improving the translatability of therapeutics from in vitro models (Table 1).

Table 1.

iPSC-based platforms for translational medicine.

Disease	Gene/mutation	Targeted pathway	iPSC subtype	Potential therapeutic agent	Reference
SMA	SMN	HDAC inhibition	MN	HDAC inhibitors	Lai et al, 201778
		HDAC inhibition	MN	HDAC inhibition—valproic acid	Ebert et al, 200911 Garbes et al, 201379 Yoshida et al, 201580
		HDAC inhibition	MN	HDAC inhibition—tobramycin	Ebert et al, 200911
		CDK4/6 inhibition	Spinal organoid		Hor et al, 201853
		SMN2 splicing modifier	MN	Rigosertib	Son et al, 201981
				TEC-1	Ando et al, 202082
				ASO	d’Ydewalle et al, 201783 Yoshida et al, 201580
				Morpholino oligomer	Ramirez et al, 201884 Osman et al, 201685
		Gene conversion SMN2- to SMN1-like	MN	CRISPR/Cpf1 and oligodeoxynucleotides	Zhou et al, 201886
		lncRNA AS1 PRC2 binding	MN	ASO	Woo et al, 201787
		Notch inhibition	MN, astrocytes, oligodendrocytes	LY-411575	Ohuchi et al, 201933
SBMA	ARpolyQ	Autophagy	MN	Trehalose	Rusmini et al, 201988
HSP	SPAST (SPG4)	Lipid metabolism liver X receptor	CSMN	GW3965	Rehbach et al, 201972
	SPG11 (SPG11)	GSK3β/βCat signaling pathway	CSMN	Tideglusib	Pozner et al, 201889
ALS	SOD1	K+ channels	MN	Retigabine	Wainger et al, 201490
		ER stress	MN	Salubrinal	Kiskinis et al, 201491
		Pathway inhibitors: inhibiting the ERK1/2, p38/MAPK, JNK, WNT, and TP53 pathways, respectively	MN	FR180204, SB203580, SP600125, XAV939, and pifithrin-α hydrobromide	Bhinge et al, 201792
		Hyperexcitability	MN	AMPA receptor antagonists, Kv7 potassium channel agonists, and DRD2 agonists	Huang et al, 202193
	TDP-43	HDAC inhibition and RNA metabolism	MN	Anacardic acid	Egawa et al, 201294
		Autophagy	MN and astrocytes	Fluphenazine and methotrimeprazine	Barmada et al, 201495
		Stress granule formation	MN	Nucleic acid intercalating molecules and cardiac glycosides	Fang et al, 201996
		HDAC6 inhibition	MN	Tubastatin a	Fazal et al, 202197
	FUS	HDAC6 inhibition		ACY-738 and tubastatin a	Guo et al, 201798
		Poly(ADP-ribose) glycohydrolase inhibition	MN	Gallotannin	Naumann et al, 201899
		Autophagy	MN	Torkinib and PQR309	Marrone et al, 2019100
	FUS and SOD1	K+ channel blockade	MN	4-Aminopyridine	Naujock et al, 2016101
	C9orf72	Anti-oxidant	MN	Trolox	Lopez-Gonzalez et al, 2016102
		Autophagy	MN	YM201636 and apilimod	Shi et al, 2018103
		Proteasome activation	MN	Rolipram	Khosravi et al, 2020104
		RNA transcripts	MN	Oligonucleotides (antisense and sense)	Donnelly et al, 2013105 Lagier-Tourenne et al, 2013106 Sareen et al, 2013107
	SALS, SOD1, TDP-43, C9orf72	Src/Abl inhibition	MN	Bosutinib	Imamura et al, 2017108
	SALS, FUS, TDP-43	Dopamine D2 receptor agonism	MN	Ropinirole	Fujimori et al, 2018109
	SALS, FUS, TDP-43, SOD1	Na+/K+ ATPase pump inhibition	MN	Digoxin, lanotoside c, proscillardin	Burkhardt et al, 201328

Abbreviations: SMA, spinal muscular atrophy; SBMA, spinal and bulbar muscular atrophy; HSP, hereditary spastic paraplegia; ALS, amyotrophic lateral sclerosis; FUS, fused in sarcoma; SOD1, superoxide dismutase 1; TDP-43, TAR DNA-binding protein 43; C9orf72, chromosome 9 open reading frame 72; SALS, sporadic amyotrophic lateral sclerosis; MN, iPSC-derived motor neurons; ASO, antisense oligonucleotide; DRD2, dopamine receptor D2.

SMA

Splicing Modifications of SMN

Because of the demonstration that antisense oligonucleotides (ASOs) targeting the splicing regulatory sequences resulting in the production of full-length SMN protein have shown clinical efficacy in treating SMA,¹¹⁰ significant attention has been drawn to the potential of using hiPSCs from SMA patients to examine mechanisms for SMN protein production and screen for ASO and morpholinos of therapeutic interest.^80,83-85,87 Other small molecules that have been studied using SMA iPSC include the small-molecule compound rigosertib, identified as an SMN2 splicing modulator, that led to enhanced SMN protein expression⁸¹ and TEC-1, identified as part of a 300 compound screen for selectivity to SMN1 that increased SMN2 transcripts and protein levels.⁸²

Histone Deacetylase (HDAC) Inhibition

HDAC inhibitors have been studied in several in vitro and in vivo models as well as in clinical trials of patients with SMA. Investigators have hypothesized, however, that those HDAC inhibitors previously studied may lack the isoform specificity to provide adequate neuroprotection. With this in mind, SMA hiPSC-MNs were used to screen over a dozen HDAC inhibitors with the ability to differentiate between the isoforms. By measuring SMN2 mRNA levels, investigators were able to predict which compounds could potentially be therapeutically relevant. These compounds were also shown to have biological activity as they increased the number of gems.⁷⁸ Other compounds with HDAC inhibition, including valproic acid^11,79,80 and tobramycin¹¹ have also been studied in this context.

Other Pathways

Hor et al used the SMA spinal cord organoid model first to show that spinal MNs from SMA patients were enriched in the expression of cell cycle genes CDK1, CDK2, CCNA2, CCNB1, and CCNB2, possibly due to the loss of the SMN protein. As a proof of principle, they tested the efficacy of the compound PD 0332991 (a CDK4/6 inhibitor) in reversing SMA MN degeneration by treating SMA spinal cord organoids for 7 days. As predicted, this compound also provided neuroprotection to MNs.⁵³ Other investigators have also used SMA iPSC-derived MNs, astrocytes, and oligodendrocytes in an effort to identify other pathways that could increase SMN transcripts or provide neuroprotection.¹²

SBMA

Protein misfolding in SBMA as well as other neurodegenerative diseases may play an important role in MN pathology because of abnormalities in autophagic pathways. With this in mind, investigators used hiPSC-MNs expressing the mutant AR, which contains an elongated polyglutamine tract (ARpolyQ), to show that the nutraceutical compound trehalose (an inducer of autophagy) could reduce ARpolyQ protein expression suggesting that this nutraceutical could be useful as a potential therapeutic.⁸⁸ However, hiPSCs have not, to date, been used as a platform to more extensively examine potential therapeutics for this disease.

HSP

Using SPG4 hiPSC-neurons, investigators established a platform of phenotypic assays to examine neurite outgrowth and pathology. As predicted, SPG4 hiPSC-CSMN showed increases in axonal swellings and growth cone area, accompanied by reductions in neurite length. Using a semi-automated image acquisition pipeline, a selection of compounds thought to have potential in the rescue of neurodegeneration phenotypes were tested. Analysis of the phenotypic readouts found that an agonist for liver X receptor (LXR), GW3965, decreased axonal swelling and growth cone area as well as increased neurite length in SPG4 hiPSC-CSMN, thereby helping uncover a potential new drug target for HSP treatment.⁷²

SPG11 encodes for spatacsin which is a transmembrane protein believed to be involved in axonal maintenance and linked to autophagic-lysosomal machinery. Based upon a previous study in iPSC-NPCs which demonstrated that a reduction in NPC proliferation was mediated by increased GSK3β activity, and that blocking this activity with the SDK3β inhibitor rescued this defect,¹¹¹ Pozner et al examined neurite length, membranous inclusions, and neuronal death in SPG11 hiPSC-neurons to examine the therapeutic potential of tideglusib, a GSK3β inhibitor. These investigators demonstrated additional restoration of neurite length accompanied by reduced membranous inclusions and improvement in neuronal survival.⁸⁹

Given the plethora of genes associated with HSP, it will be interesting to see if therapeutics targeting specific mutations will have broader applicability among the HSP phenotypes. Another challenge in translating discoveries in this disorder is that individual HSPs are relatively uncommon in their incidence and indolent in their presentation—making the timing of therapeutic initiation more challenging in these patient populations.

ALS

ALS hiPSCs have been used much more extensively than those from the other motor neuronopathies for the screening of therapeutics. Some screens have been based on electrophysiological assays, ability to provide neuroprotection, pathway-specific therapeutics, and gene therapy targets (C9orf72 ASO) among others.²⁷ Most recently, high-content live cell imaging of ALS hiPSC-MN has been used to screen a large chemogenomic library in order to identify a subset of compounds that reduce hyperexcitability.⁹³ Some platforms have advanced from in vitro discovery to early-phase clinical trials in ALS patients. A large repository of ALS iPSCs have been generated to include clinical, pathological, fluid biomarker, molecular, and genetic correlates of the disease (www.answerals.org). Table 1 shows therapeutics that have been studied in ALS hiPSCs and/or subsequently translated to clinical trials for the disease. Identifying novel drug targets for the treatment of ALS is uniquely challenging because unlike other MNDs discussed in this review, the majority of ALS cases do not arise from known heritable mutations which greatly hinders the identification of relevant druggable pathways. This is why screening drugs using hiPSCs generated from SALS patients is relevant to the identification of new therapeutics for the disease and has the potential to significantly increase the translatability of clinical trials in a way that benefits a larger proportion of ALS patients.

Limitations and Future Directions

iPSC technologies have incredible potential in the field of MND research and will continue to provide platforms for novel discoveries. However, they are not without their limitations, many of which result from practical demands, including significant time, labor, and cost required to generate and store human cells of interest. Additionally, as iPSCs for MNDs become more ubiquitously used, greater uniformity in MN induction and differentiation protocols will be necessary to have confidence in the reproducibility of results.

One particular future direction that may help to refine our understanding of MND pathology lies in the efficient and reproducible generation of homogenous populations of MN subtypes that display the same molecular markers as are observed in vivo. Terminal differentiation of hiPSC-MNs is typically assessed through immunocytochemical analysis for pan-MN markers Isl1 and Hb9¹¹² in combination with more specific markers for certain classes of MN, including L1CAM, LHX3, and FOXP1.^7,113,114 A study by Thiry et al, demonstrated, however, that the heterogeneity of the neural population following hiPSC differentiation extends beyond that which can be captured by immunocytochemistry. Using single-cell RNA sequencing, they found that of the 58% of cells that were identified through immunocytochemistry as FOXP1⁺/LHX3⁻ suggesting lateral motor column (LMC) MN identity, 4 LMC MN subtypes could be identified based on differential expression of HOX genes, further diversifying the predominant neural population into cervical medial LMC, cervical lateral LMC, lumbar LMC, and immature cervical LMC.¹¹⁵ This result is consistent with other work by Amoroso et al, which found that LMC MN identities are enriched following hiPSC induction protocols that use retinoic acid and purmorphamine gradients.¹¹² Since our knowledge of the transcriptional events that lead to the diversification of MN subtypes is more limited, an important future direction will involve the refinement of induction and differentiation protocols so that specific neural subtypes can be enriched in order to better study the molecular mechanisms that lead to the selective loss of MN subtypes in MND.

Summary/Conclusion

Although the neurobiological underpinnings of the MNDs discussed in this review are quite different, human MNs and other non-neuronal cells generated from hiPSCs have been used to uncover unique properties of their pathobiologies, independent of the shared constraints of animal modeling. The field of hiPSC modeling is one that is evolving rapidly and is underscored by the need to develop more complex models that can more accurately recapitulate disease microenvironments in vitro. ALS may represent the most challenging of these investigations since the majority of ALS patients from whom iPSCs are derived do not have known gene mutations and have significant heterogeneity in their presentation and progression. This is compounded by the variable involvement of both CSMN and MN in these ALS populations. Conversely, SMA has provided an excellent opportunity to validate iPSC platforms because of its known mutations as well as the development of therapeutics which are known to be effective both in vitro and in patients—thus providing insights into how iPSC platforms may inform about relevant disease-specific pathways and clinical responses to translational therapeutics. The study of hiPSC-MN derived from patients with MNDs has increased our understanding of their genetic determinants and has furthermore provided greater knowledge of the convergent pathways/phenotypes between the MNDs discussed in this review. In the future, iPSC-based drug discovery platforms used for a specific MND may very well have applications for others.

Funding

This work was supported by grants from the NIH/NINDS 1R01NS117604-01 and the Dept. of Defense ALS Research Program W81XWH2010161. Illustrations were created with BioRender (https://biorender.com/).

Conflict of Interest

N.J.M. declared Advisory role with BrainStorm Cell Therapeutics, Amylyx, Apellis, Janssen, Orion, Cytokinetics; research funding from Biogen Idec, MediciNova, MT Pharma, Anelixis, Apellis Pharma, Orion Pharma, Helixmith; stock ownership with Johnson & Johnson. A.E.J. declared no potential conflicts of interest.

Author Contributions

Conception and design, manuscript writing: A.E.J. and N.J.M.

Data Availability

No new data were generated or analyzed in support of this research.