The Path to Progress Preclinical Studies of Age-Related Neurodegenerative Diseases: A Perspective on Rodent and hiPSC-Derived Models

Abstract

Alzheimer’s disease (AD) and Parkinson’s disease (PD) are the two most prevalent age-related neurodegenerative diseases, and currently no effective clinical treatments exist for either, despite decades of clinical trials. The failure to translate preclinical findings into effective treatments is indicative of a problem in the current evaluation pipeline for potential therapeutics. At present, there are no useful animal models for AD and PD research that reflect the entire biology of the diseases, specifically, the more common non-Mendelian forms. Whereas the field continues to seek suitable rodent models for investigating potential therapeutics for these diseases, rodent models have still been used primarily for preclinical studies. Here, we advocate for a paradigm shift toward the application of human-induced pluripotent stem cell (hiPSC)-derived systems for PD and AD modeling and the development of improved human-based models in a dish for drug discovery and preclinical assessment of therapeutic targets.

Graphical Abstract

In light of the persistent failure to generate clinically effective disease-modifying therapeutics for neurodegenerative diseases, there is a pressing need to improve current research preclinical modeling systems for such diseases. MacDougall et al. advocate for a paradigm shift toward the application of state-of-the-art human-induced pluripotent stem cell (hiPSC)-derived systems for PD and AD modeling and the development of improved human-based models in a dish for drug discovery and preclinical assessment of therapeutic targets in an effort to improve the efficiency of the current drug-development pipeline.

Main Text

The Burden of Alzheimer’s Disease (AD) and Parkinson’s Disease (PD)

AD and PD are age-related neurodegenerative diseases (NDDs) characterized by pathological hallmarks of chronic neurodegeneration and proteinous aggregations. AD and PD are the two most prevalent age-related NDDs worldwide, affecting an estimated 29.8 million people¹ and 6.2 million people worldwide, respectively.² Age is one of the foremost risk factors for AD and PD, with global estimates showing that the AD incidence rate doubles every 5 years up to the age of 65 and then increases exponentially up to the age of 85,³ and this trend between advancing age and increasing incidence rates is similarly observed in PD.⁴ Moreover, with a rapidly increasing aged population, the global incidence rate for these diseases is projected to rise dramatically over the next three decades, with the United States alone estimated to have almost 14 million AD cases by 2050.⁵ Due to the chronic nature and current lack of effective treatment options for these diseases, patients are frequently left with debilitating symptoms that require ongoing care. These diseases pose not only a significant economic burden but place a considerable strain on the healthcare infrastructure. In 2019, the total lifetime cost of care for a single patient with dementia was estimated to be $357,000, and the total annual payments for long-term care was estimated to be $244 billion, a cost that is projected to increase to $1.1 trillion by 2050.⁶ Global projections suggest that incidence rates will continue to climb and may threaten to overwhelm the existing healthcare infrastructure.⁷ This problem is exacerbated by the current lack of clinically effective disease-modifying therapeutics (DMTs). The development of DMTs requires an advancement in our understanding of the underpinning pathogenic mechanisms and the causal factors of AD and PD. Forthcoming knowledge from basic science AD and PD research can potentially be translational into the discovery of DMTs and will facilitate clinical trials.

The Current Status of AD and PD Treatment

The current “gold-standard” treatments are aimed at improving symptom management rather than curative treatments for the underlying disease. To date, there are only 5 US Food and Drug Administration (FDA)-approved drugs available for AD, four of which are cholinesterase inhibitors (rivastigmine, tacrine, galantamine, and donepezil), with the fifth being an N-methyl-D-aspartate (NMDA) receptor antagonist (memantine).⁸ Moreover, the therapeutics available for PD are aimed at artificially supplementing dopamine (DA) or catecholamine neurotransmitters to physiological levels, without actually treating or preventing degeneration of these types of neurons. FDA-approved PD treatments include DA precursors (levodopa [L-DOPA]), DA agonists (rotigotine and ropinirole), catechol-O-methyltransferase (COMT) inhibitors (entacapone), and monoamine oxidase B (MAOB) inhibitors (selegiline and rasagiline).⁹ Whereas these agents may provide some level of temporary relief for patients, they are only effective in a subset of patients, become less effective over time, and perhaps most importantly, do nothing to slow or halt disease progression.¹⁰

In contrast, drugs under development intended to modify the pathological processes leading to AD or PD and to act on the underlying pathogenic mechanisms of the diseases (i.e., DMTs) are gaining increasing research focus and represent a highly attractive therapeutic avenue. Despite years of extensive research aimed at developing DMTs for AD and PD, there are currently no clinically effective treatments that prevent or delay the onset of these diseases or halt their progression.^11,12 To date, over 1,000 putative therapeutics have been assessed in clinical trials, with over 99% of agents evaluated in 413 clinical trials between 2002 and 2012, resulting in unsatisfactory outcomes.⁸ This high failure rate reflects the gaps in understanding the etiology of these diseases and is also indicative of utilization of unsuitable disease models in the preclinical proof-of-concept stage. The latter highlights the need to revise the design of preclinical studies and to improve the disease models utilized in the investigation of novel therapeutics toward a model system that fully, or at least better, recapitulates crucial aspects of the underpinning molecular processes of PD and AD.

Setbacks in AD and PD Clinical Trials: Lessons from Aducanumab

The extremely high failure rate of DMTs developed for NDDs, particularly those aimed at AD and PD, stands in clear contrast with other illnesses (e.g., cancer), with almost one-half of the tested therapeutics developed toward cancers demonstrating benefit in patients (https://clinicaltrials.gov/). Although this failure has been attributed to many aspects of the drug-discovery pipeline for these diseases, it primarily stems from a lack of understanding of the biological processes that lead to these diseases, which may fundamentally misinform target identification. For example, the majority of the AD clinical trials have been aimed at decreasing levels of beta-amyloid (Aβ) aggregates and plaques using drugs that reduce production and aggregation of Aβ and/or improve its clearance. Unfortunately, such trials have repeatedly failed to rescue the disease phenotypes, and many investigators and funding bodies now acknowledge the focus needs to shift to other potential culprits. Insufficiencies in preclinical disease models, particularly the currently prevalent animal models, have played a major role in propagating the misleading putative efficacy and safety outcomes in the preclinical investigations of these drugs. This is perhaps best exemplified in the antibody-based drug-targeting Aβ, aducanumab, developed by Biogen using Neurimmune’s proprietary Reverse Translational Medicine platform, which was until recently, the most promising candidate drug for AD (Table S1). Aducanumab exhibited a considerable reduction in the number of amyloid plaques present in the brains of transgenic (Tg) AD mice and improved clinical outcomes (positron emission tomography scans for amyloid at 1 year, neuropsychiatric assessments, quarterly MRI surveillance, and fluid biomarkers).¹³ The performance of aducanumab was so promising, particularly in its ability to reduce amyloid, that whereas preclinical- and early clinical-stage studies were ongoing, Biogen was simultaneously running two phase 3 clinical trials (EMERGE and ENGAGE). Thus, many patients, investigators, and pharmaceutical officials were greatly disappointed when it failed to maintain its therapeutic efficacy in the EMERGE and ENGAGE trials in 2019. Moreover, it revealed that Aβ is likely not a clinically relevant therapeutic target for creating an effective DMT for AD. Nevertheless, Biogen maintains that there were some positive benefits in the EMERGE trial, which prompted the company to apply for regulatory approval for aducanumab in the United States in July 2020.

A variety of additional therapeutics aimed at reduction of Aβ have similarly been met with persistent failure to translate into positive clinical outcomes. In 2016, Eli Lilly’s anti-Aβ antibody drug, solanezumab, failed to surpass placebo benefits in a phase 3 trial of 2,100 patients (Table S1). Drug giant Merck used a different approach to combat Aβ in people with late-onset AD (LOAD) by targeting the β-secretase 1 enzyme (BACE) with the inhibitor verubecestat (MK-8931). After successful preclinical and phase 1 clinical investigations, Merck began a phase 2/3 trial of MK-8931 in 2012 with the goal to complete the trial in late 2019. Nevertheless, Merck ultimately cancelled this trial in early 2017 after an independent study found that it had “virtually no chance” of working, and a second attempt to treat earlier stages of AD was also scrapped in early 2018. In 2014, AstraZeneca and Eli Lilly announced an agreement to co-develop lanabecestat (AZD3293), an oral BACE inhibitor. After very promising preclinical and phase 1 clinical trials, a phase 2/3 clinical trial of lanabecestat started in late 2014 but was halted in late 2018 before its planned conclusion due to similar ineffective clinical outcomes. Johnson & Johnson’s BACE inhibitor atabecestat, designed to slow cognitive decline in people at risk for AD, was also shelved in 2018 when liver enzymes spiked in study participants, a nonintended side effect. 557 participants in this trial are still being tracked as part of a safety follow-up. In addition, BACE inhibitors may have a significant side effect related to impaired motor coordination due to the fact that BACE enzymes, specifically BACE1, are necessary for the proper function of muscle spindles. Notwithstanding, BACE1 knockout mice show no pathological outcomes and are perfectly healthy. Of note, however, is the caveat that all anti-Aβ therapies have been extensively tested in patient populations suffering from sporadic AD (sAD) and LOAD, with no therapeutic studies having yet been conducted in patients with the less-common, early-onset, familial forms. Additionally, issues including poor study design, wrong stage of LOAD matched to a particular drug, limited statistical power of endpoint measures, and inclusion of ineligible participants may potentially contribute to the failures associated with LOAD clinical trials.¹⁴ Many of these caveats could possibly be circumvented through improvement of otherwise insufficient and/or incomplete preclinical assessment of putative DMTs.

Although successful drug development for AD-PD spectrum diseases is undoubtedly impeded by a current lack of understanding of the causes and early processes leading to disease onset, this review focuses on the shortcomings of preclinical research studies in the context of disease model systems. We provide an overview of current in vitro and in vivo systems for modeling AD and PD in preclinical studies, highlight their benefits and limitations, and discuss the need for improved models for developing successful DMTs. Consistent with this perspective, this review advocates for a careful and thoughtful transition away from animal models of age-related NDDs and toward human-induced pluripotent stem cell (hiPSC)-based models to facilitate preclinical studies with improved predictive validity for prospective DMTs. Finally, we provide an updated appraisal of the current hiPSC technologies and explore future directions of hiPSC models.

In Vivo Models: The Current State of PD and AD Animal Models

An Overview of In Vivo Modeling Approaches

A variety of model organisms have been utilized in the study of AD and PD ranging from simple invertebrates, such as C. elegans, to higher-order mammals, such as nonhuman primates.^9,15 Simple invertebrate models present a powerful framework for assessing novel pharmacologic and genetic intervention strategies, offering the advantages of low cost and the ability to screen hundreds of potentially therapeutic compounds in parallel; however, the invertebrate nervous system is not sufficiently complex to model the region-specific neurodegeneration observed in AD and PD, and invertebrates are not able to exhibit disease-related behavioral impairments. As such, invertebrates are typically employed for testing the initial viability of novel compounds to act on disease-relevant pathways, and those that are found to be effective move to mammal-based models for safety profiling and assessing therapeutic efficacy. The primary animal models used at this stage of preclinical testing are mice and rats, and experimental results from these rodent-based studies frequently serve as the basis for determining whether a potential therapeutic will progress to human clinical trials. Despite decades of potentially promising preclinical studies in rodents, there are currently no effective DMTs for AD and PD.

Rodents present several opportunities to model NDDs. Most importantly, the rodents are complete organisms with an aging brain, and their nervous system exhibits a general hierarchical organization similar to that observed in humans. Specifically, the presence of an anatomically distinct substantia nigra (SN) that provides DA to support the function of a striatal-based motor system is essential for modeling PD behavior,¹⁰ whereas a basal forebrain-based cholinergic system that provides acetylcholine to the hippocampus (HP) and cortex to support learning and memory is essential for modeling AD.¹⁶ Another key feature of the rodent nervous system is the capacity to manifest the cellular pathologies observed in AD and PD, namely misfolded protein aggregates that form intra- and extracellular inclusions known collectively as proteopathies.¹⁷ Together, the presence of analogous neural architecture and cellular pathways makes rodents an attractive model for studying disease-relevant processes within the context of functional neural circuits and complex behaviors. In addition, rodents offer several economic and experimental advantages relative to higher-order mammalian, such as nonhuman primates. Specifically, rodent studies are relatively less expensive, span a shorter timeline, and require fewer approvals.

However, despite possessing the capacity to display these anatomical and pathological features to an extent, AD and PD rodent models that recapitulate all clinical and pathological features simultaneously have yet to be developed. Decades of research have revealed that the mechanisms driving disease features in rodent models are not necessarily common to those of the naturally occurring disease in humans.^18,19 This limitation is likely linked to the fact that rodents do not live long enough to naturally develop age-related NDDs, and consequently, these features must be experimentally induced on significantly compressed timescales, which is dissimilar to how they manifest in human populations. Similarly, laboratory animals are not exposed to environmental factors across decades of a lifespan like humans.

Rodent-based models of AD and PD can be broadly divided into two classes based upon the method employed to induce disease-related phenotypes: transgenic and chemical toxin-based models.

Transgenic Models of AD

The cause of AD is poorly understood; however, several pathological features are consistently observed in AD patients. Pathological hallmarks at the molecular level include extracellular senile plaques composed of insoluble Aβ peptides and intraneuronal neurofibrillary tangles (NFTs) composed primarily of the microtubule-associated protein tau, which has become hyperphosphorylated.^20,21 These cellular phenotypes lead to widespread neuronal loss in diffuse brain regions, including the HP, cerebral cortex, cingulate gyrus, thalamus, amygdala, basal ganglia, and some parts of the brainstem.²² Neuronal death causes cognitive and behavioral impairments, most notably in memory deficits. Thus, an ideal animal model of AD would exhibit these molecular and cellular pathological features and lead to cognitive impairments in memory and other behaviors.

The familial, or autosomal-dominant, form of AD results from mutations in the amyloid precursor protein (APP), presenilin 1 (PSEN1), or presenilin 2 (PSEN2) genes, all of which play a role in the metabolism of Aβ, a core component of extracellular plaques.¹⁹ Almost 200 AD mouse models have been developed to date (ALZFORUM Research Models Database; https://www.alzforum.org/research-models), with the most common manipulations targeting the APP or PSEN1 genes.²³ AD transgenic models carrying such mutations offer researchers the ability to induce clinical features by manipulating genes associated with the familial-linked form of the disease (Table 1); however, this form of AD accounts for less than 1% of all clinical cases, and the extent of involvement of these genes in the much more common, sporadic form remains unclear.²⁴ Most of these models are based on aspects of the amyloid hypothesis, with the primary goal to generate aberrant accumulation of Aβ peptides to reproduce the extracellular plaques characteristic of AD. Models based on point mutations and overexpression of APP or PSEN exhibit some deficits in learning and memory, and several generations of models showed an improvement in the ability to model senile plaque formation (Table 1). However, virtually none of these models are able to reproduce the accumulation of NFTs, and the neurodegeneration phenotype is generally modest with a pattern of neuronal loss that is distinct from that observed in AD.31, 32, 33 Other efforts to generate AD transgenic mouse models have focused on human MAPT expression, the gene encoding for tau, which is the primary component of NFTs.²⁸ Collectively, these models showed that manipulation of the MAPT gene can accelerate tau hyperphosphorylation and aggregation, resulting in NFT formation, which causes a clinically relevant pattern of neuronal loss; however, these models fail to reproduce the accumulation of Aβ and formation of plaques²⁹ (Table 1). Thus, transgenic models based on manipulation of AD-associated genes have so far failed to simultaneously reproduce the major hallmark features of AD, and their relevance to sporadic forms of the disease remains unclear. Furthermore, the relevance of these models is limited also due to overexpression of the mutated transgenes, which does not reflect physiological levels.

Table 1.

Transgenic Models of AD and PD

Gene	Model	Mutation/Transgene	Inheritance	Pathology	Neurological Deficits	Refs.
Alzheimer’s Disease
APP	PDAPP	V717F (Indiana) + PDGF promoter	dominant	Aβ plaques; synaptic loss; no NFTs	cognitive and memory deficits	30
	APPlon	APP V717I (London)		Aβ plaques; synaptic loss; no NFTs	learning and memory deficits	31
	Tg2576	hAPP751 (Swedish) + HamPrP promoter APP KM670/671NL (Swedish)		extensive Aβ plaques; synaptic loss; no NFTs	learning and memory deficits	32
	APP23	KM670/671NL (Swedish) hAPP751 (Swedish) + Thy-1 promoter		Aβ plaques; synaptic loss; no NFTs	learning and memory deficits	33,34
	J20	APP KM679/671NL (Swedish) + PDGF promoter		Aβ plaques; no NFTs	learning and memory deficits	35
PSEN1	PS1	PSEN1 A246E	dominant	none	None	36
APP/PSEN	APP/PS1	APP V717I (London), PSEN1 A246E	dominant	Aβ plaques; no NFTs	learning and memory deficits	25
	APPswe/PSEN 1dE9	APP KM670/671NL (Swedish), PSEN1:deltaE9		Aβ plaques; neuronal loss; no NFTs	learning and memory deficits	26
	APP/PS2	hAPP695 (Swedish), PSEN2 (N141L) + Thy1.2 promoter		Aβ plaques; neuronal loss; no NFTs	learning and memory deficits	27
tau	Htau	htau + tau promoter	recessive	widespread NFTs including hippocampus; no Aβ plaques	None	28
	THY-tau22	htau G272V P301s + Thy1.2 promoter		NFTs in hippocampus; synaptic loss; no Aβ plaques	learning and memory deficits	29
APOE/tau	P301S	P301S tau mice with E4 allele knock-in	recessive	increased tau accumulation; neural atrophy & neuroinflammation	N/A	37
Parkinson’s Disease
SNCA		point mutation: A53T, A30P, E35K, E46K, E61K; overexpression: duplication & triplication	dominant	widespread α-syn expression accumulation; no dopaminergic neurodegeneration	moderate motor deficits	39, 40, 41, 42, 43
LRRK2		point mutation: GS2019S, R1441C/G, T1348N, A2016T, D1994A, S190A, S935A, and others	dominant	affect inclusion formation; little dopaminergic neurodegeneration	most lack motor deficits	44, 45, 46, 47
UCH-L1		point mutation: I93M and S18Y	dominant	no α-syn expression formation; dopaminergic neurodegeneration	mild motor deficits	47,48
PRKN		point mutation: W402A, S65A, and others; overexpression: duplication & triplication	recessive	affect synuclein α-syn expression aggregation; most exhibit no dopaminergic degradation	lack clear motor deficits	49,50
PINK1		point mutation: G309D, exonic deletions				51
DJ-1		point mutation: L166P, exonic deletions				52

Aβ, β-amyloid; NFTs, neurofibrillary tangles; PDGF, platelet-derived growth factor; N/A, not applicable; Tg, transgenic.

Genome-wide association studies (GWASs) have revealed over 25 different genetic loci associated with risk for developing LOAD, the primary sporadic form of the disease; however, none of the familial-linked genes related to Aβ metabolism listed above has been identified as conferring risk for developing LOAD. Apolipoprotein E (APOE) is the strongest and most reproducible genetic risk factor for LOAD.⁵³ The APOE gene gives rise to three APOE allelic variants: ϵ2, ϵ3, and ϵ4, defined by two SNPs: rs429358 and rs7412.⁵⁴ Since its initial discovery, APOE has been studied extensively by multiple groups; however, its actual role in health and mechanistic explanations for its aberrant function in the context of AD have yet to be determined. Several APOE transgenic models have been developed to explore the interaction between APOE and autosomal dominant-linked genes, such as APP (Table 1). Introduction of human APOE has previously been shown to postpone deposition of Aβ in transgenic mice expressing familial AD (fAD) mutations (fAD-Tg mice), including 5xfAD-Tg mice, which overexpress mutant human Aβ (A4) precursor protein 695 (APP) with the Swedish (K670N, M671L), Florida (I716V), and London (V717I) fAD mutations, in addition to human PSEN1 containing the two fAD mutations, M146L and L286V.⁵⁵ Importantly, these models establish an experimental link between APOE and familial-linked AD genes by demonstrating that APOE differentially regulates multiple aspects of Aβ accumulation. The interaction between APOE and MAPT has also been a subject for investigation, whereby the presence of the APOEε4 allelic variant has been shown to significantly exacerbate tau-mediated neurodegeneration in a P301S tau transgenic mouse model, demonstrating that APOE can also affect tau pathogenesis and neurodegeneration independently of Aβ pathology.³⁷

Overall, the historical focus of transgenic AD modeling on autosomal-dominant forms of the disease, which represent less than 1% of AD cases, via manipulation of familial-linked genes, makes their relevance to the vast majority of clinical patients unclear. Taken together, the primary concerns with current transgenic models of AD are the number and complexity of genes involved and a lack of clear understanding of their roles in disease etiology, making it difficult to know how or when to manipulate these targets to faithfully reproduce AD pathologies, which is reflected in the current failure to generate an incumbent model that simultaneously reproduces all hallmark clinical features of the disease as they manifest in human patients. Many of the transgenic AD models are successful in overexpressing Aβ peptides or tau proteins independently and subsequent formation of senile plaques or NFTs, respectively, but virtually no models to date are able to reproduce both of these features in parallel,⁵⁶ much less faithfully reproduce the observed patterns of neuronal loss and subsequent behavioral impairments as a result.

In addition to transgenic models developed by individual academic labs, the National Institutes of Health (NIH) is currently leading the Accelerating Medicines Partnership-AD (AMP-AD) Target Discovery and Preclinical Validation Project, which is a consortium of multi-institutional and multi-disciplinary grants that collectively intend to expedite the drug-discovery pipelines for AD (https://www.nia.nih.gov/research/dn/amp-ad-target-discovery-and-preclinical-validation-project). Particularly relevant to this review is the arm of AMP-AD, named the Model Organism Development and Evaluation for LOAD (MODEL-AD) initiative. MODEL-AD aims to generate novel transgenic mouse lines that accurately model AD phenotypes and serve as an improved platform for the drug-development pipeline and more translatable and biologically relevant models for preclinical DMT validation. The MODEL-AD initiative is comprised of the Center at Indiana University School of Medicine, The Jackson Laboratory, Sage Bionetworks, the University of Pittsburgh School of Medicine, and a center at the University of California Irvine. Together, this consortium intends to establish the next generation of in vivo AD models based on human data in a rigorous and highly controlled fashion. In doing so, it aims to align the pathophysiological features of AD models with corresponding stages of clinical disease using translatable biomarkers as a means of increasing the translatability of preclinical findings.⁵⁷ To date, 54 AD mouse models have been generated, including fAD, APOE, APP, MAPT/tau mice, as well as LOAD models based on Aβ accumulation and other previously identified LOAD-related variants (https://www.model-ad.org/strain-table/), which are available through the JAX/IU/Pitt AD Precision Models Center.

Chemical-Toxin Models of AD

In addition to the more recent transgenic models, AD rodent models were historically based upon exposure to toxic compounds that cause selective neurodegeneration leading to cognitive and behavioral impairments similar to those observed in AD patients. Whereas a variety of toxin-exposure models have been employed in the study of both AD and PD, AD models have largely been replaced by the more recent transgenic models discussed above. Chemical-based models of AD induce AD-like pathology by using either compounds that selectively and reversibly disrupt cholinergic neuron activity or toxins that permanently disrupt cellular homeostasis processes leading to cell death, such as okadaic acid and various classes of heavy metals⁵⁸ (Table 2). Although these models produce some neurological and behavioral deficits that resemble AD symptoms, including impaired learning and memory, they largely lack AD pathological lesions (i.e., the Aβ plaques and NFTs). Animal models that focus explicitly on disruption of cholinergic neurons are based on the now somewhat dated cholinergic hypothesis of AD.⁵⁹ In such models, muscarinic receptor antagonists, for example, the compound scopolamine, are delivered to targeted brain regions, such as the HP, to block endogenous acetylcholine activity and induce cognitive impairments in learning and memory.⁶⁰ Other chemical intervention models of AD involve the local or peripheral administration of neurotoxins leading to nonspecific neuronal cell death67, 68, 69, 70 (Table 2).

Table 2.

Chemical Models of AD and PD

Agent	Class	Biological Pathway	Pathological Features	Neurodegeneration	Neurological Deficits	Refs.
Alzheimer’s Disease
Scopolamine	biological tropane alkaloid	nonselective muscarinic receptor antagonist	block long-term potentiation; interfere with neuronal spine maturation & dendritic arborization	N/A	impaired learning, memory, attention, sensory discrimination, and locomotor activity	61, 62, 63
Okadaic acid	biological toxin	PP1 & PP2A inhibitor; oxidative stress; mitochondrial dysfunction	tau hyperphosphorylation and accumulation; no amyloid pathology	cholinergic neurons in forebrain	memory impairment	64, 71, 72
Heavy metals	inorganic compound	ROS and oxidative stress	accumulation of Aβ & NFTs; diffuse gliosis; pericellular edema in cerebral cortex; cholinergic dysfunction; increased ptau	cholinergic neurons in hippocampus	impaired memory and learning	67, 68, 69, 70
Parkinson’s Disease
6-OHDA	catecholamine analog	Inhibition of mitochondrial complex 1 produces H2O2 ROS.	lesion formation; no LB formation	dopaminergic loss (SN & VTA)	impaired motor activity	73, 74, 75, 76
MPTP	synthetic opioid analog	ETC disruption inhibits mitochondrial complex 1, causing free radical accumulation.	acute: none chronic: α-syn aggregates (SN, midbrain); no LB formation	acute: dopaminergic loss (SN & VTA) chronic: neuronal loss (diffuse)	acute: none chronic: motor deficits	77, 78, 79, 80, 81
Rotenone	Pesticide	inhibition of mitochondrial complex 1 and microtubule formation	α-syn aggregates (diffuse); LB-like inclusion formation	DA neuron loss (SN & VTA); cholinergic loss	impaired motor activity, abnormal posture, and slow movement	82, 83, 84
Paraquat	Herbicide	increased oxidative stress causing increase in ROS	α-syn aggregates; LB-like inclusion formation	DA neuron loss (SN)	impaired motor activity	85, 86, 87, 88
Preformed fibrils	recombinant monomeric protein	trigger hyperphosphorylation of endogenous α-syn proteins (S129 site)	LB-like inclusion formation; dysregulated dopamine release from the SN	neuronal loss in SN	disrupt ultrasonic vocalizations; no motor impairments	87,88

DA neuron, dopaminergic neuron; ETC, electron transport chain; H₂O₂, hydrogen peroxide; LB, Lewy body; PP1 and PP2A, protein phosphatase 1 and 2A, respectively; ROS, reactive oxidative species; SN, substantia nigra; VTA, ventral tegmental area.

Transgenic Models of PD

As is the case for AD, transgenic mouse models of PD enable the study of clinical disease features induced through manipulation of genetic loci associated with Mendelian forms of the disease, and hundreds of unique lines are now commercially available for purchase. Like AD, the majority of clinical PD cases are classified as sporadic, with heritable forms of the disease accounting for less than 1% of all cases.⁸⁹ Genetic alterations in the SNCA, LRRK2, and UCH-L1 genes are linked to autosomal-dominant forms of PD, whereas alternations in the PRKN, PINK1, and DJ-1 genes are associated with the autosomal-recessive forms⁹⁰ (Table 1). The two genes most commonly targeted in transgenic PD models are SNCA and LRRK2.^91,92 Mutations in these genes cause familial PD and were also associated with the sporadic form of PD via GWAS.⁹³ Additionally, the role of SNCA overexpression in familial (tri/duplication cases) and sporadic PD has been well established.⁹⁴ To date, most transgenic models manipulating SNCA not only introduce the familial missense mutations but also employ an overexpression strategy by inserting multiple copies of the human or mouse SNCA gene to mimic increased α-synuclein (α-syn) production, either alone or in combination, with a familial missense mutation.95, 96, 97

A missense mutation in LRRK2 is the most commonly reported mutation associated with familial PD,⁹⁸ and several transgenic mouse lines have been generated based on a variety of known LRRK2 mutations. The GS2019S mutation, for example, results in a gain-of-function mutation with enhanced kinase activity.⁹⁹ Other models, such as the R1441C/G, T1348N, and A2016T mutations, report changes in α-syn expression or aggregation and in some cases, nominal motor deficits; however, the precise deficits and clinical features vary substantially among models.^89,92,100 In addition to the SNCA and LRRK2 genes, multiple other transgenic lines have been generated in which other PD-associated genes are targeted (Table 1). Whereas many of these models successfully alter the expression and aggregation of α-syn, most exhibit little to no neurodegeneration, and consequently, motor functions remain relatively unimpaired⁹² (Table 1). Careful analysis of α-syn aggregates produced in many of these transgenic lines have further revealed that the inclusions formed do not resemble the naturally occurring Lewy bodies (LBs) observed in postmortem PD brains but instead, display a unique structural composition.¹⁰¹

The failure of transgenic models to reproduce both the molecular and neurodegenerative aspects of PD is likely due to a combination of several factors. Chief among these concerns is that the nature and extent of involvement of familial-linked genes, and the more common sporadic form of the disease remains unclear, calling into question the relevance of models based on manipulation of these genes to the vast majority of PD cases. Another limitation is that the transgenic models result in overexpression of the transgene that does not mimic physiological levels and thus, may introduce artifacts. Moreover, the sporadic form of the disease likely involves alterations in multiple genes, with susceptibility being further impacted by environmental components, suggesting that monogenic transgenic models may have limited utility. Overall, many of the same issues plaguing the AD transgenic models hold true for models of PD: there is a current lack of understanding of the precise role of the familial-linked genes in PD etiology and the extent to which they contribute to the sporadic forms of the disease. These issues have made it exceedingly difficult to produce a transgenic PD model, which faithfully reproduces α-syn accumulation and LB formation, in turn, leading to a pattern of neurodegeneration and behavioral impairments that match those observed in clinical populations. The current transgenic models recapitulate only some of these features and as a result, have proven to hold little predictive validity for assessing the effectiveness of potential DMTs. In contrast, chemical-toxin models of PD are still widely utilized due to their ability to faithfully reproduce the selective DA neurodegeneration characteristic of PD.

Chemical-Toxin Models of PD

Over the last half-century, a wide variety of toxic compounds have been employed to model the pattern of neurodegeneration observed in PD. Out of these, four chemical injury models have emerged as the most widely utilized for preclinical research, and their effects have been extensively characterized. These include 6-hydroxydopamine (6-OHDA), a catecholamine analog; 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP), a synthetic opioid analog; rotenone, a pesticide; and paraquat, a herbicide¹⁰² (Table 2). Another nontransgenic PD model is based on injection of preformed fibrils (PFFs; recombinant monomeric α-syn proteins converted into fibrillar aggregate form) into the brains of animals, which triggers hyperphosphorylation of endogenous α-syn proteins and elicits LB-like pathology in the SN pars compact (SNpc)^87,88 (Table 2). These toxin-based models are able to induce, with varying degrees, α-syn aggregation and selective neurodegeneration of DA neurons in the nigrostriatal pathway, a notable shortcoming of their transgenic model counterparts. The pathways involved in triggering cell death and the extent of DA neuron loss, however, vary substantially among models, which have been reviewed in detail elsewhere.¹⁰³ Ultimately, the result of DA cell loss in PD is that motor nuclei of the striatum are no longer able to maintain normal physiological function, causing a wide range of impaired motor abilities. However, loss of nigrostriatal DA neurons is, in and of itself, not a model of PD per se but instead mimics a more general Parkinsonism characterized by a broader spectrum of motor symptoms.¹⁰⁴ To this point, whereas some toxin-based models, including PFF, chronic MPTP, and rotenone administration, are able to induce α-syn aggregates and LB-like formation, they are distinct from the characteristic LB inclusions in PD, and the overall distribution of these aggregates does not reflect the pattern observed in PD brains.¹⁰² These models are also limited in their ability to model the progressive nature of PD and spread of the pathology to other brain regions, as PD patients also suffer neurodegeneration in subcortical and eventually, cortical regions in the later stages of the disease.¹⁰⁵

In summary, toxin exposure-based models demonstrate some similarities to the pathologies observed in PD; however, the mechanism of cell death induced by the toxins is likely distinct from those underlying neuronal loss in PD.^106,107 Of note, typical dosing schedules for chemical-toxin models of PD occur over days to weeks, whereas the course of PD develops over decades and progresses over years. These limitations suggest that toxin exposure-based models aimed at inducing selective DA neuron death in the nigrostriatal pathway present an incomplete picture of PD and provide an explanation as to why efforts to identify DMTs in such models have failed to be effective in clinical trials.¹²

Viral Vector Models for PD and AD

Advantages of Viral Vector Systems

Viral vector delivery systems offer several advantages that are difficult to achieve with other approaches.¹⁰⁸ First, viral vector systems provide robust control over the temporal expression of the gene of interest. Second, aging is the primary risk factor of AD-PD spectrum disorders, and the possibility to deliver the vector at any point in the animal’s lifespan represents a major advantage for study design. Third, viral vectors support local transgene delivery and expression, thereby allowing for accurate and specific targeting of brain regions of interest. Fourth, precise dosage manipulations can be readily achieved with the use of viral vectors. Fifth, the model of interest can be created in multiple animal species, ranging from small rodents to large nonhuman primates. Sixth, different variations of genes can easily be made to obtain a phenotype of interest. Seventh, the versatility and flexibility of different vector platforms used for viral-mediated gene transfer into the central nervous system are vast. The viral systems, including recombinant adeno-associated vectors (rAAVs) and lentiviral vectors (LVs), efficiently and robustly support short- or long-term gene expression, respectively, both locally and globally. Last, the viral vector-based systems are significantly more economical, both effort and cost wise, relative to other disease models.

AD and PD Viral Vector Models: AAVs and LVs

The viral vector approach to model AD and PD phenotypes in vivo has been utilized by several different groups using either rAAVs or LVs. Since their first use at the beginning of the millennium, a variety of viral-based models have been developed and utilized for AD. The initial AAV-tauopathy models exploited AAV2/2 to express human tau P301L protein in the brains of mice and rats.^109,110 In these studies, the AAV-tau-P301L vector was injected into the basal forebrain of adult rats leading to the increased levels of the tau for at least 8 months post-transduction. Significantly, hyperphosphorylated tau and aggregates resembling NFTs were found at 3−4 weeks post-transduction.^110,111 These studies provided the first proof of concept that injection of AAV-P301L tau could result in persistent expression of the protein and its aggregation in mice and rats. Consistently, injection of AAV-wild-type (WT) or triple-mutant APP resulted in the formation of Aβ plaques but no overt neurodegeneration. These findings are in contrast with the above observations that AAV-WT tau or AAV-P301L tau was capable of causing significant neurodegeneration of HP pyramidal neurons.¹¹² In more recent experiments, AAV2/5 was used to deliver WT tau and the GFP control vector into the dorsal HP of aged Fischer 344 rats.¹¹³ Both viruses were robustly expressed in HP neurons and led to aberrant axonal structure, axonal degeneration, and fragmentation of tau in HP axons compared to control GFP-expressing axons that appeared normal. In another model, the tau P301L expressed from the human synapsin I promoter was delivered by the AAV2/9 vector into the lateral entorhinal cortex, which allowed the authors to detect multiple forms of phosphorylated and aggregated tau variants prior to the loss of perforant path synapses and neurodegeneration.¹¹⁴ These data support the usefulness of AAV-based approaches in AD, particularly in modeling early pathological changes, such as axonal degeneration and degradation.

A number of studies used viral-based models aimed at inducing other pathological aspects of LOAD.^115,116 For instance, AAV-BRI-Aβ42 and AAV-BRI-Aβ40 vectors were used for controlled expression and secretion of Aβ peptides in rodent HP.¹¹⁶ Notably, AAV-BRI-Aβ42-injected animals demonstrated that these vectors were able to induce plaque formation, but this was not seen with rAAV2/1-BRI-Aβ40 expression. In another study, AAV2/2 was used to deliver Aβ40 and Aβ42, C-terminal fragment of APP containing the Aβ peptides (C100), and a V717F mutant of C100 to the HP and cerebellum of mice. Interestingly, Aβ42 and V717F mutant demonstrated greater induction of microgliosis and disruption of the blood-brain barrier (BBB) compared to the Aβ40 forms but did not activate plaque formation, astrocyte induction, or neurodegeneration. The use of AAV and LV technology to express various forms of Aβ and its precursors will continue to be an essential method of studying their role in LOAD.¹¹⁷ More recently, LV has been used to express TDP-43, Aβ42, or both in the motor cortex of rats, which uncovered that the loss of TDP-43 implicates microglia as a cause of the synaptic degeneration observed in LOAD.¹¹⁸ Furthermore, lenti-TDP-43 triggered alterations in the APP processing associated with Aβ42 production and induced caspase activity and neuroinflammation, which was similarly observed with the lenti-Aβ42. The combined expression of both TDP-43 and Aβ42 protein resulted in an outcome similar to TDP-43 alone but with the added feature of neuronal loss, suggesting a possible synergistic effect between the two factors. Another study using LVs to express TDP-43 in the motor cortex found altered amino acid metabolism, oxidative stress, and neuronal death.¹¹⁹ These results together demonstrate the utility of LV-based AD model systems to elucidate key pathogenic factors related to protein dysfunction implicated in AD, whereas Aβ and tau have been extensively tested in LOAD, consistent with the long-held notion that Aβ is a key suspect in AD pathogenesis. Nevertheless, the continuing failures in drug trials aimed at lowering amyloid levels demand refocusing the effort to other potential culprits.

Multiple recent studies demonstrated that APOEε4 may contribute to toxicity associated with the AD phenotype, and to date, APOE remains the most significant and replicable genetic risk factor for AD.¹²⁰ Given the role of APOEε4 in AD pathogenesis and the potential protective effect of the ε2 isoform, Hu and colleagues¹²¹ generated mouse models using AAV8-APOE isoforms driven by GFAP promoter specifically expressed in astrocytes in all brain regions, which resulted in an overall increase in APOE levels throughout the mouse brain. The viral-mediated overexpression of APOEε4 in the APOEε4-TR mice increased poorly lipidated APOE lipoprotein particles and decreased APOE-associated cholesterol in APOEε4-TR mice. Conversely, APOEε2 overexpression in APOEε4-TR mice enhanced APOE lipidation and associated cholesterol. Furthermore, overexpression of APOEε4 elevated the levels of endogenous Aβ, whereas APOEε2 overexpression trended to lower endogenous Aβ. Based on these data, authors suggest that increasing APOEε2 in APOEε4 carriers is a beneficial strategy to treat AD, whereas increasing APOEε4 in APOEε4 carriers is harmful. In another study, direct intracerebral injection was performed using AAV to express APOEε2. The study reported that APOEε2 overexpression markedly reduced brain-soluble (including oligomeric) and -insoluble Aβ levels, as well as amyloid burden, in two mouse models of brain amyloidosis in which pathology is dependent on either the expression of mouse or human APOEε4 isoforms.¹²² Zhao et al.¹²² further showed that a widespread reduction of brain Aβ can be achieved through a single injection of virus via intra-thalamic delivery of AAV expressing APOEε2. Collectively, these data suggest that AAV gene delivery of APOEε2, using an AAV, can rescue the detrimental effects of APOEε4 on brain amyloid pathology; thus, this approach may represent a viable therapeutic avenue for treating or preventing LOAD.

Similar viral vector approaches have been assessed in various models of PD, with the earliest renditions employing AAV2/2 or LVs.123, 124, 125, 126 These vectors injected into adult rat brains were used to deliver WT, A30P, or A53T mutants of human α-syn for disease induction. The authors demonstrated efficient expression of α-syn in nigral DA neurons, accompanied by molecular and cellular pathologies and nigral DA degeneration that evolved gradually over time. Nonetheless, the first generation of rAAV2/2 vectors used in these studies displayed progressive neuronal loss, although the neurodegeneration of the tyrosine-hydroxylase-positive neurons, as well as the time course, was quite variable (25%–80% and 6 weeks to 1 year, respectively).^123,124 In contrast, LVs encoding WT-, A30P-, or A53T-mutated α-syn were capable of inducing neuronal cell loss in rats in a more delayed and more consistent manner (25%–40% and 5 months, respectively).¹²⁶ Both studies demonstrated progressive loss in neurite length and swollen perikarya in remaining DA neurons. Furthermore, massive cytoplasmic accumulations of α-syn were found in both cell bodies and neurites. These results have been replicated recently in multiple studies that utilized both viral platforms.127, 128, 129, 130, 131

LVs have also been utilized in a variety of mice PD models. For example, Lauwers and colleagues¹²⁵ showed that injection of LV carrying WT or A30P mutant of α-syn in the striatum, amygdala, or SN of mice was capable of inducing neurodegenerative changes associated with PD in a time-dependent manner and included α-syn-dependent neuritic enlargement and cytoplasmic inclusions. Further, this study demonstrated that nigral overexpression of A30P-α-syn resulted in about 25% cell loss at 10–12 months.¹²⁵ Delivery of A30P, E57K, and E35K to the SN of the rat using LV demonstrated a similar extent in DA neuron loss (about 50%) compared with the WT (30%), whereas the faster fibril-forming mutant A53T used in the same study did not show a significant decrease in DA cell number. Interestingly, the overall neuropathological features observed in mice (i.e., onset and the severity) appeared less severe compared to the rats, although these differences may be an artifact of viral purity or production titers of the injected virus and/or other related technical discrepancies.

Important Considerations for AD and PD Viral Vector Models

The development of novel serotypes, including AAV2/1, AAV2/5, AAV2/6, AAV2/8, AAVrh10, DJ, and DJ8, has broadened cellular targets and improved AAV transduction efficiencies. The AAVs serotyped with these new capsids have been tested in rats and nonhuman primates to overexpress α-syn.132, 133, 134, 135, 136, 137 Overexpression of the S129A form delivered via AAV2/5 and AAV2/6 consistently displays enhanced toxicity on the neurodegeneration process.^134,138 These experiments also demonstrated that careful dosing of the overexpression transgene is crucial, as even control vectors carrying only fluorescent reporters delivered at high titers can demonstrate nonspecific toxic effects.¹³⁵ This caveat should be taken into careful consideration when designing rescue experiments based on AAV. For example, it has been shown that the therapeutic AAV harboring small interfering RNA (siRNA) targeting SNCA resulted in a high level of toxicity and caused a significant loss of nigrostriatal DA neurons. This study further highlighted the limitation of AAV applications in animal studies. However, it cannot be excluded that the neurotoxicity was caused by the robust reduction of SNCA levels in rat models, as α-syn plays a crucial biochemical role in DA neurons, and its robust reduction could decrease cell viability.¹³⁹ Taken together, this underscores the importance to develop a tool for fine tuning the expression of disease-associated genes for designing gene-therapy approaches. In this regard, the more recent CRISPR-dCas9-based system may present a more advantageous system for modulating gene expression in a precise and fine-tuned fashion. Along this line, we recently developed the LV-CRISPR-Cas9 system to achieve precise and fine-tuned regulation of SNCA gene expression. The study provided proof of concept that the manipulation of gene expression (e.g., reversing overexpression) through epigenome editing is a valuable therapeutic strategy for neurological disorders, such as PD,¹⁴⁰ caused by gene dysregulation. As such, the viral vector PD models using LV and AAV for the overexpression of α-syn protein clearly display PD pathology associated with DA neurodegeneration, and the use of novel AAV serotypes improves α-syn expression in DA neurons compared to AAV2/2. Overall, both AAVs and LVs are capable of inducing transgene expression and demonstrate a high level of tropism for DA neurons, resulting in similar levels of neurodegeneration. Further improvement of these viral systems will be required to circumvent a slow progression of the cellular PD phenotypes and relatively low levels of behavioral impairments.

In Vitro Cell-Culture Models

Overview of Current In Vitro AD and PD Models: Limitations and Opportunities

Over the past four decades, a variety of AD and PD in vitro models have been developed to examine the interactions between key aspects of their relative disease pathophysiology and subsequently utilized for preclinical therapeutic screening. Current AD and PD in vitro models primarily aim at recapitulating molecular and cellular disease hallmarks, such as mitochondrial dysfunction, oxidative stress, cell survival, and the presence of intracellular and/or extracellular protein aggregates. For example, the most prominent and widely used cellular models of AD are based on induction of Aβ40−42 aggregates or tau hyperphosphorylation and aggregation^141,142 (Table 1). Similarly, in vitro models of PD are based on the aforementioned toxin-induced mitochondrial dysfunction (e.g., MPTP, rotenone, paraquat, 6-OHDA) and disrupted proteostasis (e.g., thapsigargin, ionomycin, tunicamycin), α-syn overexpression, and expression of the mutated PD familial genes (e.g., α-syn, Parkin, PINK1)⁹² (Table 2). Whereas these cell-culture models have been invaluable in progressing our understanding of pertinent neurobiology, they have not yet resulted in viable DMTs aimed at prevention, delayed onset, or slowed disease progression, calling into question their relevance and translational validity.

There are several clear limitations in these models that are important to acknowledge in order to improve current preclinical practices. For example, models based on overexpression and knockout/down of pertinent disease-associated genes are notoriously variable in the quantitative changes to protein levels and the methods used to induce changes to gene expression (i.e., viral-mediated, breeding-based, constitutive, or transient expression) that may lead to contradictory data and inferences.⁹² This lack of consistency in eliciting disease-associated phenotypes introduces a caveat to preclinical studies using these models to test potential DMTs. Moreover, variability in cellular phenotypes and methodology observed in preclinical NDD models (e.g., fold change of protein overexpression or knockdown) strongly suggest a greater genetic heterogeneity that is not able to be fully explored in current cell models. This is further complicated by the prevalence of cell models based on genetic variants identified in familial cohorts, despite familial variants representing a small minority of overall NDD cases.^109,143 Due to the uncovered genetic causes, it remains difficult to accurately represent sporadic NDD cohorts or disease subtypes with non-Mendelian genetics in preclinical models, creating a significant barrier in elucidating pertinent disease mechanisms, identifying potential therapeutic targets, and drug development.

Given these challenges, researchers have been compelled to evaluate these models in an effort to identify areas of improvement and appraise the opportunities presented with novel cell-culture techniques. In recent years, we have built upon these models, incorporating novel gene-editing technology and evaluating important genetic disease associations obtained in GWASs in improved tissue-culture models.^144,145 The increased incorporation of hiPSC technology represents a major progression in the field of NDD in vitro modeling and provides previously unavailable opportunities to utilize disease-relevant cell types derived from patients to explore disease pathogenesis in the context of the human genome. The full spectrum of advantages and limitations of hiPSC-derived NDD models, particularly those pertaining to AD and PD, are explored in full in the next section.

hiPSC-Derived Models of NDDs: Versatile and Genetically “Faithful”

Modern advances in cell-culture technology represent one of the most exciting breakthroughs in the research of NDDs, providing opportunities for improved disease modeling, drug screening, and personalized medicine. Until recently, postmortem brain examination was considered the gold-standard human-based biological material for obtaining insights into pathological processes of human neurodegenerative conditions. However, modern stem cell technologies, including the analysis of patient-specific neurons and glial cells, have opened up new avenues for investigating the pathological mechanisms of NDDs and have been increasingly harnessed in the molecular dissection of NDDs with enhanced genetic complexity, such as AD and PD. Through a simple, noninvasive biopsy of patient skin samples, in addition to a variety of alternative biological samples, the generation of hiPSCs enables researchers to reveal molecular pathways and disease mechanisms that underly NDDs in a way that was impossible with previous cell-culture methods. At present, the most prevalent donor cells are fibroblasts, which are utilized in over 80% of all published reprogramming experiments;¹⁴⁶ however, hiPSCs can be obtained from a variety of sources, including embryonic cord stem cells, umbilical cord blood, corneal epithelial cells, and blood cells, such as peripheral blood mononuclear cells147, 148, 149 (Figure 1). Consequently, hiPSC-derived models represent a versatile alternative to existing cellular disease models and are obtained from patient somatic tissues that are coerced into the pluripotency through use of a variety of reprogramming tools, such as plasmids, vectors (e.g., episomal), and viral transduction (e.g., adenovirus, Sendai virus, lentivirus; these methods are comprehensively reviewed in Brouwer et al.¹⁵⁰ and Malik and Rao¹⁵¹). These cells are self-renewing and offer an alternative to embryonic stem cells (ESCs), which are accompanied by ethical considerations and social concerns. There are currently several hiPSC repositories that collect and distribute hiPSC lines derived from patients with NDDs and health age-matched controls in an effort to centralize samples and make them readily available to researchers (Table S2).

Figure 1.

Schematic of Modeling Neurodegenerative Diseases Using Human-Induced Pluripotent Stem Cells (hiPSCs) Derived from Biological Samples

A variety of biological samples, such as primary keratinocytes, embryonic cord stem cells (EMBCs), umbilical cord blood (UCBs), corneal epithelial cells (CECs), and blood cells, such as peripheral blood mononuclear cells (PBMCs), have been utilized as starter tissue for chemical reprogramming into desired cell types for preclinical research. Additionally, fibroblasts obtained from a simple, noninvasive skin biopsy can be chemically reprogrammed into hiPSCs, which represent highly versatile cells in their ability to adopt different cellular identities. To date, hiPSCs have been successfully reprogrammed into various progenitor cells, including neural (NPCs), oligodendrocyte (OPCs), and glial (GPCs) progenitor cells, according to references provided. These, in turn, have been chemically coerced into forming authentic singular cultures comprised of neurons, oligodendrocytes, astrocytes, and microglia, which display true characteristics with high fidelity and generative highly informative preclinical NDD models. Additionally, combinations of neuronal and glial cell populations can be used in 2D co-cultures and 3D brain models.

One of the major applications of hiPSCs is the differentiation into specific cell types, which have consequently been utilized in a variety of biological contexts, including neurodevelopmental¹⁵² and neurodegeneration studies,¹⁴⁴ ex vivo transplantation,^153,154 disease modeling, target validation, and drug discovery.¹⁵⁵ To date, several cellular differentiation protocols have been developed that generate not only generic neurons but also specific neuronal cell-type populations, including excitatory, cholinergic, DAs, and inhibitory GABAergic neurons.156, 157, 158 A variety of specific glial subpopulation cells have also been derived from hiPSCs, including astrocytes,159, 160, 161 oligodendrocytes,^162,163 and microglia.164, 165, 166 A number of novel gene-editing techniques, including CRISPR-Cas9, zinc-finger nucleases (ZFN), and transcription activator-like effector nucleases (TALENS), introduced an additional dimension to the hiPSC-derived system—the creation of isogenic models that possess the same genetic background and only differ in the chosen mutation.¹⁶⁷ The diversity of cell differentiation protocols has facilitated mechanistic studies, and the availability of isogenic lines has supported genetic analyses of gene function and its pathogenic role in NDDs. hiPSC-derived models are suitable for these investigations, as they exhibit key disease-related features. For example, fAD- and sAD-derived hiPSC neuronal models carrying PS1 and PS2 mutations (A246E and N14II, respectively)¹⁶⁸ and APP duplication mutation¹⁶⁹ display important AD biochemical features, including Aβ secretion, increased Aβ42:Aβ40 ratio, and elevated hyperphosphorylated tau. It is important to note that the use of hiPSC models that generate Aβ-related phenotypes represent a more accurate means of exploring the role of Aβ expression in AD neurodegeneration when compared to traditional ectopic expression models or models that require supplement and/or induction of Aβ aggregates, as these adopt the “natural” occurring pathological processes related to Aβ deposits.

hiPSC-derived in vitro models of PD have also been highly informative in elucidating the pathophysiological progression linked to familial PD mutations in patient-derived hiPSC DA neuronal models. For example, hiPSC-derived neurons from a patient with the LRRK2 GS2019S mutation exhibited disease-related cellular phenotypes, including increased vulnerability to oxidative stress¹⁷⁰ and decreased neurite length and number.¹⁷¹ Other examples include hiPSC-derived DA neurons obtained from patients possessing the PARK2 mutation that exhibited mitochondrial dysfunction and oxidative stress morphology¹⁷² and hiPSC-derived neurons with the A53T mutation and the SNCA triplication that exhibit hallmark elevation and accumulation of α-syn aggregates.¹⁷³ Moreover, several studies ameliorated these disease phenotypes by interventions aimed at correcting the mutated gene, including LRRK2,¹⁷⁴ PINK1,¹⁷⁵ SNCA,^140,176 and GBA1.¹⁷⁷ Such models demonstrate the utility of hiPSC technology in the identification and validation of novel drug targets and provide proof of concept for gene therapy approaches. hiPSC-derived models for different NDDs have since been expanded to a variety of different co-culture 2D and 3D configurations, broadening the modularity and potential basic research and preclinical applications.

Current Configurations of hiPSC-Derived NDD Models

2D Single and Co-Culture Cell Models

A significant amount of work utilizing 2D cultures comprised of hiPSC-derived singular cell types has been published within the last decade. These single cell-type models have been informative in uncovering the role of individual cell populations in NDDs, such as spinal muscular atrophy,^178,179 AD, PD, and amyolateral sclerosis (ALS)¹⁸⁰ (Figure 1), and furthermore, have enabled a greater understanding of how disease-associated genes act on the biological level to affect cellular perturbations. hiPSC cellular models have been recruited in AD research to better understand the role of genetic polymorphisms and mutations in individual cell populations and how these may contribute or impact disease. In an hiPSC-based AD model, Kondo et al.¹⁸¹ observed increased Aβ oligomer accumulation linked to increased endoplasmic reticulum dysfunction and oxidative stress in hiPSC-derived neurons and astrocytes obtained from sAD and fAD patients containing the APP E693Δ mutation. hiPSC-derived 2D individual models have also shown great value in studying the genetic contributions of APOEε4, the greatest and most replicable genetic risk factor of AD, in a variety of neural cell models.¹⁸² Lin and colleagues¹⁸³ used 2D cultures to uncover that isogenic hiPSC-derived APOEε4 astrocytes exhibit large-scale changes in gene expression pertaining to cholesterol accumulation and internalization of Aβ42, both of which are critical homeostatic functions of astrocytes, relative to isogenic APOEε3 cells. Furthermore, isogenic ε4 astrocytes showed a reduced ability to express and secrete APOE, resulting in diminished lipidation. Enhanced neurite growth in both length and number in neurons, in addition to enhanced Aβ secretion, in isogenic APOEε4 neurons was also found to be a pathogenic contribution in these models.¹⁸³ In generating astrocyte cultures obtained from neurotypical individuals isogenic for the APOEε3 and APOEε4 alleles, Zhao et al.¹⁶¹ found that APOEε4 astrocytes exhibited a lower APOE lipidation status compared with the healthy APOEε3 controls, further highlighting the insight gained from such models.

The use of 2D co-cultures has also enabled the study of APOEε4 in the context of individual cell populations and cell-cell interactions and how these may contribute to AD pathogenesis. Extending on the above experiment, upon co-culturing hiPSC-derived APOEε4 astrocytes with neurons, the morphological and functional integrity of neurons was compromised and demonstrated inhibition of maturation compared to their isogenic ε3 controls.¹⁶¹ The interplay between different neuronal and glial cells plays a critical role in the development and progression of AD and PD, thus leveraging on the hiPSC technology to establish in vitro models that mimic cell-cell interactions, and the cellular microenvironment has proven to be highly informative.

The benefit of 2D singular and co-culture models has similarly been realized in studying the contribution of several PD-causing mutations to disease-related cellular phenotypes, such as decreased cell viability and mitochondrial dysfunction. This was first demonstrated in a study by Nguyen et al.,¹⁷⁰ who showed that PD patient-derived hiPSC neuronal models with the LRRK2 G2019S mutation showed increased α-syn levels, upregulation of genes involved in oxidative stress response, as well as increased susceptibility to H₂O₂-induced oxidative stress and cell death. Similarly, Sánchez-Danés et al.¹⁷¹ evaluated changes in autophagy in DA neurons derived from PD patient hiPSCs lacking the LRRK2 G2019S mutation and observed elevated α-syn and autophagic dysfunction. A greater understanding of how cellular interactions underpin pathophysiology has been achieved with hiPSC-derived co-culture model systems. For example, hiPSC-derived astrocytes were shown to exert a protective benefit in co-cultured neural progenitor cells (NPCs) obtained from PD patients after exposure to the mitochondrial toxins, rotenone and potassium cyanide, demonstrated by the rescue of differentiation deficits and mitochondrial dysfunction.¹⁸⁴ Another study utilized neuron-astrocyte co-cultures to investigate the impact of the PD-causing LRRK2 G2019S mutation on neuron-astrocyte crosstalk. The authors reported that co-culture of control hiPSC-derived DA neurons with patient-derived astrocytes carrying the G2019S mutation induced neurodegeneration and accumulation of α-syn in the DA neurons.¹⁸⁵ This study exemplified the utility of co-cultures in the examination of different combinations of the gene variant/mutation by cell types to uncover cell-cell interaction and the cell type/s that exert the pathogenic effect of disease-associated variants. However, the 2D co-cultures form a microenvironment for interactions between different brain cell types and are still innately limited in their ability to model the complexity of the brain as a three-dimensional tissue and as such, possess inherent obstructions in translation relevance.

3D hiPSC-Derived Models

Emerging technologies have enabled researchers to begin generating and assessing 3D hiPSC-derived models of NDDs, including co-cultures and organoids, which are both biochemically and physiologically robust.

Organoid Models. 3D human brain organoid models are aimed at capturing the complex spatial organization of the human brain as closely as possible in an in vitro model. Obtained from hiPSCs, these brain organoids are self-organized and self-patterning 3D. In addition to their overall epigenomic and transcriptional signatures,^186,187 the physiological properties of 3D organoid models, such as neuronal responsiveness to glutamate, instinctive Ca²⁺ signaling, and development of cellular maturation phenotypes, have been also shown to closely mirror human brain tissue.188, 189, 190 Recent studies have generated and characterized AD and PD 3D hiPSC-derived models, which have been shown to exhibit pathological features in a highly reproducible manner. For example, Gonzalez et al.¹⁹¹ developed an AD brain organoid model that exhibited progressive accumulation of the Aβ peptide with plaque-like structures and amyloidogenic properties, as well as the appearance of phosphorylated tau and NFTs. Lin et al.¹⁸³ expanded on these findings through introducing isogenic APOEε3- or APOEε4-carrying microglia-like cells into previously established cerebral organoids cultivated from hiPSCs containing the AD-associated APP duplication mutation and observed a significantly long process formation in the APOEε4 microglia-like cells compared to the APOEε3 controls, suggesting a limited ability in sensory and responsive capabilities to extracellular Aβ. Most recently, Zhao et al.¹⁹² generated cerebral organoid models using hiPSCs derived from AD patients and healthy controls carrying different APOE genotypes. The APOEε4 organoid model exhibited disease-related phenotypes compared to the APOEε3 organoid derived from a different donor and the isogenic APOEε3 organoid model created by genome editing of the APOEε4 line. The adverse effect of the APOEε4 genotype in the organoid model recapitulated key AD pathological features, including loss in synaptic integrity, apoptotic cell death, Aβ and phosphorylated tau accumulation, and elevated soluble APOE levels, thus demonstrating the suitability of the system for modeling non-fAD. Furthermore, these results provide further support for the pathogenic role of theAPOEε4 in AD and its potential as a druggable target for AD.¹⁹² Similarly, Smits et al.¹⁹³ generated a functional midbrain DA organoid model of PD derived from PD patient tissue containing the LRRK GS2019 mutation. This organoid model produced and secreted DA and successfully exhibited associated hallmark pathologies in organoids, such as diminished quantity and complexity of midbrain DA neuronss¹⁹³ However, a GT approach correcting the LRRK mutation was insufficient for amelioration of disease phenotypes in this model. Bolognin et al.¹⁹⁴ also established a 3D DA neuronal model derived from a PD patient with the LRRK2 G2019S mutation, which exhibited signs of mitochondrial dysfunction and altered morphology, decreased DA differentiation and branching complexity, and elevated premature cell death in young neurons when compared to isogenic control lines; some of these perturbations were rescued by administration of the LRRK2 inhibitor 2 (Inh2).

However, it is important to note the various limitations that have been reported in the use of 3D organoid models, particularly pertaining to the spontaneous and uncontrolled manner of cell growth and differentiation that contributes to interexperimental variability. The reliance of spontaneous self-assembly in organoid models can lead to model-to-model variation, including heterogenous cell proportions, spatial orientation, and disease phenotype, making reproducibility a prominent concern in 3D co-culture and organoid models.¹⁹⁵ Recent advances are, however, showing improvements in reproducibility in 3D organoid models,¹⁹⁶ variability in terminal cell identities remains,¹⁹⁵ and continued progress is anticipated.

Co-Culture Systems. Recent biotechnological advancements have enabled the generation of sophisticated three-dimensional scaffolding that allows hiPSC-derived cells to grow in co-culture-based models, which facilitate a more controlled approach compared to the organoids discussed above, particularly with respect to cell-type, maturation, composition, and homogeneous exposure of the cells.¹⁹⁷ Existing co-culture-based models include models of motor neurons and the neuromuscular junction derived from ALS patients;¹⁸⁰ DA neurons from PD patients with known mutations (e.g., LRRK2¹⁹⁴); tri-cellular models of neurons, astrocytes, and microglia modeling the neuroinflammatory AD environment;¹⁹⁸ and skeletal muscle tissue.¹⁹⁹ These models provide researchers with improved tools to investigate the role of spatial tissue organization—cell-cell and cell-matrix connections, which play a major role in NDD pathogenesis. A recent study by Blanchard et al.²⁰⁰ constructed a hiPSC-derived 3D co-culture model of the BBB comprised of brain endothelial cells, astrocytes, and pericytes to model AD vascular-related pathologies with high fidelity. This 3D model revealed a key role of pericytes in APOEε4-derived cells, which exhibited elevated BBB accumulation of Aβ peptides and fibril amyloid as a result of increased NFAT-calcineurin pericyte signaling.²⁰⁰ Park et al.¹⁹⁸ also developed a hiPSC 3D tri-culture AD model comprised of neurons, astrocytes, and microglia in a microfluidic platform, which successfully recapitulates hallmark AD characteristics, including accumulation of phosphorylated tau of Aβ aggregates, as well as neuroinflammatory features, such as microglial recruitment, nitric oxide release, and subsequent oxidative stress.

Although 3D co-culture systems aim to overcome the aforementioned limitations observed in organoid models, they also inherent setbacks. Current hiPSC-derived models generate neuronal networks that lack critical features of the brain tissue that they aim to represent, including a lack of vascularization and requisite circulation of nutrients and removal of metabolic waste,²⁰¹ and a variability in generating biologically representative proportions of neuronal and glial cell populations, features that are arguable strengths in using 3D organoid cultures. In addition, 3D co-culture models require artificial scaffolding and sequential seeding of cells, whereas organoid models developed more “naturally.” Nonetheless, the use of hiPSC-derived co-cultures not only allows researchers to investigate the active role of each cell type in the generation of AD and PD disease characteristics but also the interaction of cells derived from different genotypes, highlighting the revolutionary toolkit that hiPSCs represent in the molecular study of NDDs. The advantages and limitations of hiPSC-based in vitro models are explored in greater detail in the following section.

Advantages of hiPSC-Derived Cellular Systems and Disease Models

Improved Disease Modeling and Drug-Discovery Platforms

The potential for hiPSC-based in vitro models to be highly informative for NDDs with a human genetic component investigation is high. The ability to correlate NDD candidate genes to cellular phenotypes, such as RNA and protein expression profiles, morphological changes, and biochemical signatures in hiPSC-derived cells, facilitates studies addressing how the human genetic components of complex NDDs, such as AD and PD, are expressed on a cellular level. Moreover, hiPSCs allow the incorporation of the genetic complexity of underlying patient genetic backgrounds in the model, which is often negated (e.g., chemical models) or oversimplified (e.g., monogenetic variant animal models). It is well acknowledged that the onset, progression, and severity of neurodegeneration are determined by complex genetic factors and the interplay of several genetic variants with relative minor effects. This innate genetic landscape captured in patient-derived hiPSC models allows us to investigate genetic determinants and biological factors, paired with the ability to generate modular, high-fidelity in vitro NDD models, to build upon our current understanding of disease biology.

hiPSC technology currently represents the only available human-based models in which the full genetic landscape of the patients is captured. Also, the possibility to apply genome-editing technologies, such as CRISPR-Cas9, using hiPSCs enabled the generation of isogenic models for evaluation of precise genetic variants. In addition to scalability, hiPSC are self-renewable. Thus, hiPSC-derived models present a forefront promising system for preclinical studies. Moreover, the human genetic diversity captured in patient-derived hiPSC capacity makes them suitable for the evaluation of “personalized medicine” approaches. The ability to screen effectively individual reactions to treatments and to identify nonresponders or poor responders prior to administration in a clinical setting further emphasizes the suitability of hiPSC-derived models for preclinical assessments (Figure 2). Collectively, the potential information generated through preclinical studies in hiPSC models is robust, relevant to human subjects, and essential to implement in the design of clinical trials (e.g., stratification of the study subjects) to improve the accuracy of results.

Figure 2.

Schematic of Improved DMT Drug-Development Pipeline for NDDs through Increased Incorporation and/or Substitution of hiPSC-Derived Models

Development of any putative DMT begins with the discovery phase, which encompasses both identification of pertinent gene targets and mechanisms of disease. Subsequent disease models are devised in the early preclinical phase to explore the cellular disease pathophysiology and validate drug targets, as well as early screening and optimization of early CGT-IPs/DMTs. Progressive development of DMTs, including further model development and target validation, as well as investigation into off-target effects, occurs during the mid-preclinical phase. As DMTs progress, further information regarding pharmacokinetic and pharmacodynamic activity is obtained in the late preclinical phase. The increased use of hiPSC-derived models in these phases represents an attractive improvement to the existing drug-development pipeline for several reasons; namely, they are cost effective, versatile, and most importantly, overcome many of the innate limitations of existing animal models of NDDs. At this stage, IND applications may be submitted for putative DMTs, which currently contain 4 critical objectives that must be met with any emerging DMT, particularly CGT-IPs: (1) target selection, (2) lead compound development and optimization, (3) initial and escalating dosing regimen, and (4) establishment of feasibility and ROA. Each of these can be addressed accurately and appropriately with substitution of hiPSC-based models in preclinical studies. DMTs may then receive FDA approval to commence clinical trials. The incorporation of hiPSC-based models in an additional phase 1a may be included, which facilitates improved candidate screening in the recruitment phase and identification of potential nonresponders or poor responders based on preclinical genetic studies. DMTs with demonstrable safety in phase 1a then progress through phase 1, 2, and 3 clinical trials, after which, those with a significant positive effect receive New Drug Application (NDA) regulatory approval and are released for consumer use with continuing pharmacovigilance to identify any previously unidentified adverse outcomes over time.

Overcoming Limits of Animal Models

hiPSCs offer a way to overcome many of the issues present in current animal-based models, as previously discussed. First, hiPSC-derived models provide a human-based platform for evaluating potential therapeutics in the actual genetic landscape of the human disease. Second, by deriving cells from patients, many technical issues regarding the best way to model human disease mutations are circumvented because the model contains the unique genetic background of the patient. Third, hiPSC cells are easily accessible to genetic and pharmacological manipulation, enabling high-throughput structural and functional assays that assess disease phenotype. To this end, we recently developed an hiPSC-derived DA neuronal model obtained from a PD patient with SNCA triplication in order to assess an all-in-one LV GT aimed at reducing SNCA levels. Our LV intervention successfully reduced SNCA levels and rescued disease-related phenotypes through reactive oxidative species (ROS) generation and preservation of cellular viability.¹⁴⁰ Fourth, compared with standard rodent model, generation of in vitro hiPSC-based models can be achieved in a significantly shorter timeline, in the order of 30 days as opposed to animal models that can range widely from months to years. Fifth, and in keeping with the previous point, hiPSC-based models are considerably less expensive and energy consuming than rodent models, particularly when considering on-going animal housekeeping obligations and infrastructure, regulations, as well as and training and compliance requirements. Increasingly economic preclinical disease modeling increases accessibility for research groups to assess putative therapeutics and improve translatability and ultimately improve DMT development efficiency.²⁰² Altogether, these advantages give hiPSC-derived models a distinct advantage in the study of neurodegeneration, and such models have the potential to generate novel discoveries that were inaccessible in previous cell-culture and animal-disease models. The studies reviewed here demonstrate the necessity and value in the incorporation of 2D and 3D hiPSC-based models in NDD mechanistic studies elucidating factors involved in disease causation and pathogenic pathways, as well as in translational and drug-discovery studies, such as therapeutics target identification and validation. Together, hiPSC-derived NDD models represent a leap forward in medical research through improved disease modeling that is considerably feasible, with relatively shorter experiments that are cost effective and perhaps most importantly, accurate and suitable for modeling human diseases.

Limitations of hiPSC-Derived In Vitro Models

Time and Reproducibility

The culturing technologies of hiPSCs emerged over the last decade, and their utility for disease modeling and preclinical assessment of putative DMTs are relatively novel. As with any emerging technology, there are several limitations that need to be considered when attempting to accurately recreate biological characteristics of NDDs. First and foremost, the cultivation of hiPSCs and generation of accurate and reliable cell models can be expensive, labor intensive, and time consuming. On average, it costs $10,000−$25,000 USD to validate hiPSC lines at a suitable standard for medical research.²⁰³ The processes of reprogramming, differentiation, and maturation are prolonged. Cellular reprogramming of starter tissue into hiPSCs can take at least 20−30 days,¹⁵¹ and subsequent cellular differentiation and maturation vary depending on the desired cellular type and method used; for example, generation of mature neurons, astrocytes, and microglia can require anywhere from 6 to 15 weeks,²⁰⁴ 4 to 9 weeks,^205,206 or 5 to 9 weeks,^164,207 respectively. As a result, researchers often prefer the use of progenitor cells as stable intermediates for experimentation, as they are broadly representative of the target cell type and can be generated in a faster time frame.²⁰⁴ Thus, the time constraints and costs required for establishing the hiPSC model system may present a concern for many research groups. In addition, the time required for the development of key disease-associated cellular phenotypes in hiPSC-derived cells is also a consideration in modeling NDDs. Whereas it takes decades of a patient’s life to present clinical and pathological symptoms, disease-related molecular phenotypes in hiPSC-derived models were detected about 2 months postmaturation.²⁰⁸ Nonetheless, the constraints stem from the limited lifespan of hiPSC-based models in culture that may not be sufficient for developing a complete picture resembling the disease on cellular and tissue levels. Strategies aimed at expediting the differentiation and maturation process, such as the incorporation of Notch and γ-secretase inhibitors to shorten maturation time,²⁰⁹ and neurogenin-2 (Ngn2) or NeuroD1 overexpression¹⁵⁸ have been partially successful and introduced other concerns raised from ectopic expression.

Other limitations are related to reproducibility of the systems in repetitive experiments and the inherent model-to-model variability, including setbacks in regard to the purity of the hiPSC-derived cultures and the presence of undesirable heterogenous cellular populations.²¹⁰ Thus, it is important to account for heterogeneity in cell populations when using hiPSC-derived models of disease, particularly in DMT screening and efficacy assessment. Moreover, hiPSC cultures have been reported to exhibit genomic instability and are prone to acquire genetic aberrations and mutations during cellular expansion and reprogramming.²¹¹ Care also needs to be taken in the usage of isogenic lines generated by various genome-editing techniques, as potential off-target effects can occur (e.g., unintended clonal variability across isogenic lines and off-target mutagenesis).^212,213 Therefore, it is important to periodically evaluate the genomic stability of hiPSCs to assure rigorous DMT screening.

Challenges in Recapitulating Sporadic NDD Pathophysiology and Cellular Phenotypes

The complexity of NDDs and the lack of comprehensive characterization and mechanistic understanding of their cellular phenotypes pose challenges in the ability to recapitulate them with cellular models, including hiPSC-based systems. The causes of NDDs are complex and multifactorial, including polygenic risk factors (i.e., multiple genes and variants), epigenetics marks, aging, sex, and environmental factors, such as oxidative stress triggers, some of which are difficult to induce, specifically in an endless list of various combinations, in a laboratory system. Furthermore, the majority of hiPSC-derived AD-PD models derived from patients with familial mutations that represent only a very small proportion of overall cases.^214,215 Since the mechanisms underpinning the familial versus sporadic forms may be distinct, there is a need to establish models derived from hiPSCs and fibroblasts obtained from sAD or sporadic PD patients. Toward this goal, initiatives of patients’ cell repositories have emerged in recent years; these include Applied StemCell (ASC), California Institute for Regenerative Medicine (CiRA), Cedars Sinai Induced Plluripotent Stem Cell Core, European Bank for Induced Pluripotent Stem Cells (EBiSC), Korean National Stem Cell Bank ( KSCB), National Institute of Agine (NIA), National Institue of Neurological Disorders and Stroke (NINDS), National Institute of General Medical Sciences (NIGMS), and the WiCell Research Institute (WiCell) (Table S2). The expected growth of these collections will facilitate the establishment of hiPSC-derived models more suitable for research of the common sAD and sporadic PD.

Nonetheless, hiPSC models derived from sporadic NDD patients produce appropriate cellular phenotypes characteristic of AD¹⁸¹ and PD²¹⁶ upon environmental manipulations, such as exposures to neurotoxins and chemical induction of oxidative stress. Also, a recent study revealed alterations in mitochondrial protein expression and elevated oxidative stress in hiPSCs generated from sAD patients, despite a marked lack of Aβ and tau pathology,²¹⁷ suggesting that hiPSC models may reveal valuable insight into the nuanced pathophysiology specific to sporadic NDD subtypes. Furthermore, a recent study by Meyer et al.²¹⁸ utilized hiPSCs derived from sAD patients and found that MAPT, encoding the tau protein, was significantly increased in sAD-derived cells compared to healthy controls. Thus, despite these challenges, current hiPSC-derived models are excellent surrogates that provide the opportunity to mimic disease pathophysiology and establish gold-standard disease characteristics with the full genetic underpinnings in sporadic patients, which can be harnessed as outcome measures to improve subsequent disease models.

Modeling Aging “in a Dish”

Inducing Age-Related Morphological and Functional Cellular Signatures. Whereas modeling neuro-developmental disorders (e.g., spinal muscular atrophy, primary herpes simplex encephalitis) with hiPSCs has been successful, it is much more difficult to accurately recapitulate disease phenotypes of age-related NDDs, such as AD and PD. The reprogramming of somatic cells to generate hiPSCs erases the molecular-age signatures of the donor cells and reverts them back to fetal developmental stages.^2,219,220 As such, the field has devised a variety of inventive strategies to accelerate cellular aging and/or maturation as a means of overcoming this limitation, including ectopic overexpression of aging-related proteins, such as progerin,²²¹ induced artificial shortening of telomeres,²²² and exposure to various toxins causing mitochondrial dysfunction and oxidative stress.^223,224 Another strategy to induce aging has been developed by the Chiba-Falek lab¹⁴⁴ and employs multiple passaging of the NPCs to imitate the natural aging processes. This strategy represents an attractive alternative compared to previous methods described above, as it reduces the potential artificial phenotypes, a consequence of toxin exposure or ectopic expression, and is based on a more natural process to induce aging in cellular models of NDDs.

“Direct Conversion” of Somatic Cells to Neuronal Models. Alternatively, methods known as “lineage/direct conversion” or “transdifferentiation” were developed to retain the donor age and preserve the “aged” molecular signatures. This approach is based on the direct reprogramming of human somatic cells, such as fibroblasts, into neurons through provision of various cell lineage-specific transcription factors without an intermediate de-differentiation stage into hiPSCs.225, 226, 227 An example is the generation of DA-induced neurons (iNs) directly from mouse embryonic fibroblasts²²⁸ and human astrocytes²²⁹ as a means of producing a more appropriate model system to study age-related human neurodegeneration. Indeed, these methods allow preservation of the aging effects on the genome.

Ultimately, guidelines for addressing the biological and technical limitations related to hiPSC work, particularly differences among hiPSC donors, authentication of the derived culture purity, and genetic stability, should be considered to control for variability in hiPSC-derived models and to ensure robust outcomes and a high degree reproducibility.

Future Directions in hiPSC Applications in the Context of NDDs

Personalized Medicine and Drug Screening

AD and PD exhibit high variability and heterogeneity in symptoms; thus, these pathologies may each represent a group of subtype diseases,^230,231 a critical consideration in clinical trial design when progressing putative DMTs through the drug-development pipeline. Kondo et al.²³² showed that hiPSCs derived from fAD and sAD patients exhibited a different drug responsiveness to treatment with docosahexaenoic acid (DHA), a drug that previously failed in clinical AD trials. These data are indicative of different subtypes of AD that possess different clinical responses to treatment. Thus, there is a need to stratify the study participants in order to achieve a benefit in clinical trials, implying that several therapeutics that failed in clinical trials may still have the potential to be effective in certain defined subtypes of AD. The improvement of the current preclinical models will facilitate accurate assessment of existing and new drugs by defining better outcome measures and/or appropriate patient populations.

Kondo et al.²³² further demonstrated the utility of hiPSC-derived AD models for high-throughput screening of a panel of putative therapeutic compounds using previously established disease phenotypes as outcome measures, namely, the capacity to reduce the Aβ42:Aβ40 ratio in neurons derived from 13 AD patient hiPSC lines. Six lead compounds were able to significantly reduce neuronal Aβ42:Aβ40; however, one of the compounds containing a “cocktail” of cromolyn, topiramate, and bromocriptine only rescued this disease phenotype in hiPSC neurons derived from fAD patients.²³² Patient genetic susceptibility to various environmental factors has also been explored in hiPSC-derived PD models, which was reversed with GT,^174,233 providing proof-of-concept validation for a personalized approach to therapeutic intervention. Overall, further development of hiPSC technology may ultimately allow researchers to stratify patients based on specific genetic background of risk factors in order to assess genetic subpopulations’ differing response to treatment.

Regenerative Approaches to Neurodegeneration

The capacity of brain neurons to regenerate in response to injury is minimal; with adult neurogenesis restricted to a select few brain regions and in diseases characterized by chronic neurodegeneration, such as AD and PD, the resulting cognitive and behavioral impairments are irreversible. The most exciting avenues emerging out of hiPSC technology are the tissue-engineering strategies aimed at regenerating lost neuronal tissue through transplant of hiPSCs. These revolutionary therapeutic strategies leverage the capacity of hiPSCs to be differentiated into multiple subclasses of neural and glial cells. Whereas the concept of transplantation of hiPSCs to stimulate tissue regeneration is still at an early stage, it is a rapidly developing field, and recent evidence has demonstrated that transplantation of hiPSC and ESCs can be beneficial for multiple diseases, including age-related macular degeneration^234,235 and spinal cord injury.²³⁶ With more relevance here, a recent clinical trial conducted by Song et al.,²³⁷ which involved transplanting autologous hiPSC-derived DA neurons into PD patients, has effectively restored motor deficits for the first time. This groundbreaking study provides proof of concept that such hiPSC-based transplantation techniques provide the perfect platform for successful regenerative therapy in NDDs.

Nonetheless, it is important to acknowledge that hiPSC transplantation is accompanied by various challenges, including the risk of tumorigenicity and graft rejection by the patient.^238,239 To mitigate this obstacle, a variety of immune-evasive strategies to overcome immune rejection are currently being developed.^240,241 Combined with their self-regenerative properties, hiPSCs possess many highly desirable features for personalized-regenerative medicine and make them an ideal candidate for regeneration-based therapies in CNS disorders.242, 243, 244, 245 The transplantation of hiPSCs opens therapeutic strategies to address not only the recovery of lost tissue but also the need for a DMT by using genetically corrected cells. Derived from patients, hiPSCs can undergo a gene-modifying treatment in vitro, for example using CRISPR-Cas9 technology, for targeted correction of the mutated gene. Subsequently, the genetically corrected cells could be replaced in the clinical patient. Such DNA-modifying constructs would ideally remain episomal and be self-inactivating by the time the tissue was transplanted to the patient in order to reduce the probability of off-target effects.

A Roadmap for the Submission of New Investigational Products (IPs) to Treat PD and AD

Overview of the Current Development Pipeline for IPs

The preclinical studies for new PD and AD therapeutics are important in the development pipeline of IPs. The objectives of a comprehensive preclinical program for cellular-therapy and GT IPs (CGT-IPs) include the following: (1) establishment of biological plausibility (target selection); (2) identification of biologically active dose levels (lead compound development and optimization); (3) selection of potential starting dose level, dose-escalation schedule, and dosing regimen for clinical trials; and (4) establishment of feasibility and reasonable safety of CGT-IP-proposed clinical route of administration (ROA).

These steps are crucial in preparation of the products for Investigational New Drug (IND) application and Biologics License Application (BLA) submissions. In fact, (1) low-target engagement and efficacy and (4) unsatisfactory safety profile are among the most common reasons for the failures of the CGT-IPs for PD and AD therapeutics (Figure 2). Furthermore, a recent retrospective review of delayed or denied approval of new medical entities by the FDA between 2000 and 2012 indicates that uncomprehensive formulation of target efficacy and determination of the drug-safety profile represent major factors underlying the failure of drug-development programs for NDDs, including AD and PD.²⁴⁶ This study also informed the concern of fairly low translatability in transitioning from animal-based preclinical studies to human clinical trials. It is important to note that the FDA is well-aware of these shortcomings. In fact, the FDA recently created guidelines that recommends minimizing animal use for IP validations. For example, in the context of low species-specific compatibility of the clinical product, the FDA permits testing of alternative validation systems with demonstrated analogous and suitable characteristics. The emergence of AD and PD patient-derived hiPSCs and the related platforms outlined in this review could provide such an alternative experimental system for developing putative CGT-IPs. Consistent with these recommendations, the conduct of in vitro-based studies covering safety determination, dosage optimization, functional assays, immunophenotyping, morphologic evaluation, early toxicology data, pharmacokinetics, and pharmacodynamics of the IP may satisfy the FDA requirement for establishing the biological relevance of the product. Furthermore, with the use of in vitro studies, hiPSCs would primarily be beneficial for identification of potential safety issues and mechanism of action (MOA) of investigational CGT-IPs. These considerations are well aligned with FDA guidance that recommends incorporation of the principles of the “3Rs,” to cultivate test method protocols that encourage reducing, refining, and replacing animal use, as the follow-up to the applicable provisions of the Animal Welfare Act Amendments of 1976 (7 U.S.C. 2131 et seq.).

The concept of replacing animal modes supports the utility of AD and PD in vitro modeling programs. The suitability of these efforts should be considered more widely with the assessment of the “reshifting effect,” if any, on the ability of the preclinical testing program to provide necessary data regarding the safety and activity of the CGT-IP product. The discussion above suggests the possibility to generate an early portfolio of the therapeutic program for IND advancement using predominantly in vitro testing and validation, particularly in the context of preclinical testing, as well as a phase 1a clinical trial checkpoint (Figure 2). Furthermore, advanced analyses, including dose-limiting safety and toxicity assessments, could potentially rely, at least in part, on in vitro measurements. These new clinically relevant safety, pharmacokinetic, pharmacodynamic, tolerability, and biomarker data would then inform subsequent clinical decisions regarding new IPs. Thus, conducting necessary animal studies to address any outstanding issues should be considered only after development of a CGT-IP product progresses to later-phase preclinical study. For example, if manufacturing/formulation changes occur, such that the comparability of the later-phase CGT product to the product used in early phase (tested in vitro) is uncertain, then additional in vivo preclinical studies may be needed to bridge the two products. Such bridging studies allow data collected with the early-phase product to support later-phase development or licensure.

As such, we suggest that prior to initiation of the definitive preclinical studies in animal settings, in vitro studies should be performed for identification of potential safety issues and MOA of an investigational CGT-IP product (Figure 2). With that said, we suggest supplementing the in vitro studies with assessments of a battery of physiological and functional phenotypes of the product following in vivo administration. The assessments will test specifically the amelioration of gold-standard disease phenotypes established in hiPSC-based disease models, as previously demonstrated.^{140,174,193,233}

Accordingly, the preclinical testing program should incorporate a stepwise, multifactorial approach to achieve an understanding of the biological plausibility for use of the investigational CGT product in the intended AD and PD patient populations. To clearly dissect the process of certifying a GT product for IND submission, we advise to reach out to the FDA at the earliest steps of product development, throughout a pre-pre-IND mechanism, which is a nonbinding, informal discussion between reviewers from the Pharmacology/Toxicology Branch of the organization and the investigator. The advice given by the Center for Biologics Evaluation and Research (CBER)/Office of Cellular, Tissue and Gene Therapies (OCTGT) would be extremely valuable to consider when preparing final protocols for definitive preclinical studies, as well as in preparing various sections of the briefing document for the pre-IND meeting. The animal species selected for assessment of bioactivity and safety should demonstrate a biological response to the investigational CGT product, similar to that expected in humans, in order to generate data to guide clinical trial design. Considerations for determining the relevant species include the following: (1) comparability of physiology and anatomy to that of humans; (2) permissiveness/susceptibility to infection by, and replication of, viral vectors or microbial vectors for GT; (3) immune tolerance to a human CGT product or human transgene expressed by a GT product; and (4) feasibility of using the planned clinical delivery system/procedure.

Conclusive Remarks

The pursuit of clinically effective DMTs for age-related NDDs, such as AD and PD, has been met with significant failure, despite over 40 years of research. Multiple factors are responsible to the absence of AD and PD DMTs, and inadequate preclinical disease models constitute a major limitation. Nonetheless, clinical trials of putative DMT candidates cannot be halted while waiting for the “perfect” NDD model. Current IND regulations could potentially hinder current and future research on NDD treatments by over-reliance on data generated from inadequate animal models, further slowing the progress toward achieving a DMT. With the emergence of patient-derived hiPSCs, the field entered a new era in NDD research, enabling preclinical investigations in the context of each individual patient’s genetic background, which is a key in the development of DMTs for complex genetic diseases. Paired with novel gene-editing and delivery techniques, hiPSCs represent an emerging and promising preclinical toolkit that facilitates disease modeling and drug discovery. In conclusion, whereas the utility of rodent models is valuable in several aspects of preclinical evaluation, such as assessment of safety profiling, systemic bioactivity, and functional outcomes, we need to (1) acknowledge the benefits of NDD models generated from hiPSCs and consider their utility as a complementary platform for evaluating therapeutic efficacy in preclinical studies, (2) work on improving current animal and cell-based models—and customized models that will be applicable for evaluation of precision medicine therapies, and (3) define the measurable treatment outcomes of the therapeutics intervention and accordingly, select the most suitable model system/s.