Using human induced pluripotent stem cells (hiPSCs) to investigate the mechanisms by which Apolipoprotein E (APOE) contributes to Alzheimer’s disease (AD) risk

Abstract

Although the biochemical and pathological hallmarks of Alzheimer’s disease (AD), such as axonal transport defects, synaptic loss, and selective neuronal death, are well characterized, the underlying mechanisms that cause AD are largely unknown, thereby making it difficult to design effective therapeutic interventions. Genome-wide association studies (GWAS) studies have identified several factors associated with increased AD risk. Of these genetic factors, polymorphisms in the Apolipoprotein E (APOE) gene are the strongest and most prevalent. While it has been established that the ApoE protein modulates the formation of amyloid plaques and neurofibrillary tangles, the precise molecular mechanisms by which various ApoE isoforms enhance or mitigate AD onset and progression in aging adults are yet to be elucidated. Advances in cellular reprogramming to generate disease-in-a-dish models now provide a simplified and accessible system that complements animal and primary cell models to study ApoE in the context of AD. In this review, we will describe the use and manipulation of human induced pluripotent stem cells (hiPSCs) in dissecting the interaction between ApoE and AD. First, we will provide an overview of the proposed roles that ApoE plays in modulating pathophysiology of AD. Next, we will summarize the recent studies that have employed hiPSCs to model familial and sporadic AD. Lastly, we will speculate on how current advances in genome editing technologies and organoid culture systems can be used to improve hiPSC-based tools to investigate ApoE-dependent modulation of AD onset and progression.

1. INTRODUCTION

Alzheimer’s disease (AD) is currently the 6th leading cause of death in the United States and affects an estimated 5.8 million Americans, with this figure expected to rise to nearly 14 million individuals by 2050 (“2019 Alzheimer’s disease facts and figures,” 2019). In 2018 alone, costs associated with the care of patients with Alzheimer’s and other dementias totaled nearly $234 billion. Worldwide prevalence is estimated to be as high as 24 million, with AD accounting for more than 50% of all dementia cases (Bekris et al., 2010; Mayeux and Stern, 2012). Prevalence rates of AD increase with age, increasing dramatically after age 65, with a documented nearly 15-fold increase in the prevalence of predominantly Alzheimer’s dementia in individuals between 60 and 85 years of age (Evans et al., 1989; Mayeux and Stern, 2012).

Animal models that overexpress Alzheimer’s disease (AD)-related proteins or have familial AD (fAD)-related mutations introduced into the genome have provided important insights into AD (Bettens et al., 2010; Drummond and Wisniewski, 2017; Esquerda-Canals et al., 2017). Unfortunately, these models do not display important pathological hallmarks of the human disease and have not adequately modeled the complex genetics associated with sporadic AD (sAD) (Duff and Suleman, 2004; Seok et al., 2013; Warren et al., 2015). In addition, the complexity of in vivo experiments makes it difficult to eliminate confounding variables and directly investigate the manifestation of molecular, biochemical, and cellular phenotypes. Studies of human neuronal cells have been restricted to experiments with cadaveric tissue samples, which are limited in supply and rapidly lose disease phenotypes with extensive ex vivo culture. Because of the limitations of current animal and human models of AD, the mechanisms that cause AD onset and progression remain poorly understood, possibly explaining the failure of recent experimental therapies (Qosa and Volpe, 2018).

Moving forward, accessible and reproducible human in vitro models are needed to compliment these existing models. Advances in cellular reprogramming have enabled the generation of in vitro central nervous system (CNS) disease models that can be used to dissect disease mechanisms on a cellular level and evaluate potential therapeutics (Goldstein et al., 2015; Robbins and Price, 2017). In this review, we will summarize how human induced pluripotent stem cells (hiPSCs) have emerged as viable system to model various aspects of AD pathogenesis. In particular, we will discuss how emerging technologies in genome engineering and organoid culture in conjunction with hiPSC-based models could provide new opportunities to investigate the risk-modifying effects of Apolipoprotein E (ApoE).

2. Alzheimer’s Disease: Pathophysiology, Molecular Mechanisms, and Genetics

Neuropathological hallmarks of AD include extracellular amyloid plaques, cerebral amyloid angiopathy, and neurofibrillary tangles in the form of intraneuronal aggregations of modified Tau proteins (Selkoe, 1991; Serrano-Pozo et al., 2011). In addition, cortical thinning can be observed early in the disease process by MRI, showing symmetrical atrophy of the cortex in the medial temporal lobes (Dickerson et al., 2011, 2009; Serrano-Pozo et al., 2011). Amyloid plaques seen in the brains of AD patients result from the extracellular accumulation of 40 or 42 amino acid amyloid-beta (Aβ) peptides, produced by proteolytic cleavage of amyloid precursor protein (APP) in neurons. Broadly speaking, amyloid plaques can be characterized as either diffuse or dense-core plaques. Although diffuse amyloid plaques can be detected in cognitively normal aging adults, dense-core plaques are commonly seen in the brains of individuals with AD, and are associated with damage to surrounding cellular architecture causing synapse loss, neuron death, and activation of astrocytes and microglial cells (Itagaki et al., 1989; Knowles et al., 1999; Vehmas et al., 2003). Neurofibrillary tangles are formed by aggregation of Tau, a microtubule-associated protein, leading to neuron death. Under physiological conditions, Tau regulates the stability of microtubule structure. However, in AD and other tauopathies, Tau is hyperphosphorylated causing microtubule disassembly and aggregation of phosphorylated Tau (p-Tau) filaments (Guo et al., 2017; Medeiros et al., 2011). Interestingly, although global amyloid burden is associated with decreased cognitive function and may predict longitudinal cognitive decline, recent neuroimaging studies suggest Tau positron emission tomography (PET) may be more sensitive than Aβ PET and increasing levels of Tau may be a better predictor of early decline in cognition than Aβ alone (Aschenbrenner et al., 2018; Farrell et al., 2017; Ossenkoppele et al., 2019).

2.1. The amyloid hypothesis

The Aβ peptide observed in the cortex of AD patients was first isolated and sequenced in 1984 and found to be homologous to the amyloid peptide seen in the brains of individuals with Down’s syndrome (Glenner and Wong, 1984a, 1984b). This finding was the first to suggest the genetic deficit in AD may be localized on chromosome 21. Subsequent cloning and characterization of a cDNA that encoded the amyloid peptide revealed a highly conserved β-amyloid precursor protein (APP) gene product that was mapped to human chromosome 21, further suggesting a causal relationship between trisomy 21 in Down’s syndrome and AD pathology (Goldgaber et al., 1987; Robakis et al., 1987). These discoveries set the stage for further investigation in the field built on the premise that Aβ accumulation is the primary pathogenic event leading to Alzheimer’s dementia.

Genetic investigations quickly linked mutations in the APP gene with early-onset AD, providing evidence that polymorphisms in the APP gene can drive early-onset amyloid pathology (Goate et al., 1991), although mutations across independent families were shown to be heterologous. Further characterization of APP mutants revealed that mutations within APP were generally localized in or flanking the region coding for the amyloid peptide that altered the processing of the gene product by proteases α-, β-, and γ- secretase, leading to increased production of Aβ peptide (Cai et al., 1993; Citron et al., 1992). Additionally, it was demonstrated that polymorphisms in the Presenilin 1 and Presenilin 2 (PSEN1, PSEN2) genes led to increased Aβ generation and altered APP processing through direct effect on γ-secretase (De Strooper et al., 1998; Levy-Lahad et al., 1995; Scheuner et al., 1996; Sherrington et al., 1995). These seminal studies pushed the field to focus research heavily on the proposed ‘amyloid cascade hypothesis’ that postulated AD pathogenesis begins with missense mutations in APP, PSEN1, or PSEN2 genes leading to the downstream cascade of altered APP processing, increased Aβ production, and deposition in the form of diffuse plaques that represent the catalyst for widespread neuronal dysfunction and the clinical presentation of dementia (Hardy and Selkoe, 2002). Additionally, mutations defined within the gene encoding the tau protein leading to frontotemporal dementia and neurodegeneration do not induce amyloid pathology, which has strengthened the hypothesis that Aβ seeding precedes neurofibrillary tangle formation and is the initial event leading to downstream neurodegeneration in AD (Hardy et al., 1998).

2.2. Familial Alzheimer’s disease (fAD)

Familial Alzheimer’s disease (fAD) is characterized by early age of dementia onset, ranging from 30 to 65 years of age, and accounts for 1–6% of overall cases (Bekris et al., 2010). The accelerated onset in fAD is driven by highly penetrant autosomal mutations in the coding sequence of PSEN1, PSEN2, or APP (Bertram and Tanzi, 2005; Chartier-Harlin et al., 1991; Goate et al., 1991; Janssen et al., 2003; Levy-Lahad et al., 1995; Rovelet-Lecrux et al., 2006; Scheuner et al., 1996). The Presinilins encode major components responsible for γ-secretase cleavage, and therefore modulate Aβ levels by altering cleavage of APP. Mutations in the APP and PSEN1 genes are associated with complete penetrance, while mutations in the PSEN2 gene show 95% penetrance (Goldman et al., 2011; Sherrington et al., 1996). Nearly 300 autosomal dominant pathogenic mutations within these three genes have been identified thus far, however novel disease-causing mutations continue to be revealed (Giau et al., 2019; “Mutations | ALZFORUM,” 2019) along with numerous variants of unknown pathogenicity.

2.3. Sporadic Alzheimer’s disease (sAD)

The late onset sporadic form of AD (sAD), associated with onset after the age of 65, accounts for more than 95% of all AD cases (Reitz et al., 2011; van der Flier and Scheltens, 2005) and is directed by genetic and environmental risk factors. A genome wide association meta-analysis study of 94,437 individuals diagnosed with late onset AD, identified 25 risk loci that regulate pathways of immunity, lipid metabolism, tau binding proteins, and amyloid precursor protein (APP) metabolism (Kunkle et al., 2019). Although numerous common and rare variants continue to be identified that may contribute to disease risk, polymorphism in the Apolipoprotein E (APOE) gene has been identified as the strongest risk factor for sAD (Corder et al., 1993; Liu et al., 2013; Saunders et al., 1993; Strittmatter et al., 1993), and continues to be replicated by GWAS analyses (Jansen et al., 2019; Kunkle et al., 2019).

3. The Role of Apolipoprotein E (APOE) in Alzheimer’s disease

Apolipoprotein E (ApoE) is a cholesterol transport protein secreted primarily by astrocytes in the CNS, conferring its effect on neurons primarily through low-density lipoprotein (LDL) family of receptors (Grehan et al., 2001; Holtzman et al., 2012; R. E. Pitas et al., 1987; Robert E. Pitas et al., 1987). Human APOE is present as three main variants- ε2, ε3, and ε4- with typical allele frequencies of ~8%, ~75%, and ~15% respectively (Farrer et al., 1997; Raber et al., 2004). The ApoE isoforms differ by single amino acid substitutions at amino acid position 112 and 158: ε2 (Cys112, Cys158), ε3 (Cys112, Arg158), ε4 (Arg112, Arg158). Compared to individuals with an APOE3/3 genotype, the presence of the ε2 allele confers ~40% decrease in risk of developing AD, while one or two copies of the ε4 allele introduce a 3-fold and up to ~12-fold increase in risk, respectively (Corder et al., 1994; Holtzman et al., 2012). Despite the relatively low population frequency of the ε4 allele, frequency among individuals with AD increases significantly to ~40% (Liu et al., 2013; Raber et al., 2004). In addition to Alzheimer dementia, the frequency of the ε4 allele has been shown to be elevated in pure synucleinopathies without Alzheimer pathology such as pure dementia with Lewy bodies (pDLB) and Parkinson’s disease dementia (PDD) (Tsuang et al., 2013). Interestingly, this suggests that ApoE may be capable of facilitating neurodegeneration through non-amyloidogenic mechanisms. Further, it has been shown in human and rodent models that the ε4 allele exacerbates tau-mediated neurodegeneration and decreases age of onset in patients with tauopathy, independent of amyloid pathology (Koriath et al., 2019; Shi et al., 2017). Recent evidence in a mouse model of tauopathy suggests a role of ApoE in modulating activation of microglia leading to neurodegeneration and tau pathogenesis (Shi et al., 2019).

The 299 amino acid ApoE protein is composed of two distinct domains, the N-terminal domain contains the receptor binding region whereas the C-terminal domain is responsible for lipid binding (Weisgraber, 1994). The LDL receptor family including LDLR and LDL receptor protein 1 (LRP1), as well as heparan sulfate proteoglycans (HSPG) are responsible for a majority of the ApoE and Aβ endocytosis in the CNS (Kanekiyo et al., 2014; Liu et al., 2013). ApoE and Aβ bound lipoproteins can be endocytosed by (i) direct interaction with LDL receptors (ii) the LRP/HSPG complex or (iii) HSPG alone (Mahley et al., 1999). The presence of Cys158 in the ApoE2 isoform impairs its ability to bind to LDL receptors (Kowal et al., 1990), making individuals homozygous for the ε2 allele susceptible to type III hyperlipoproteinemia (Weisgraber et al., 1982). The conversion of ApoE2 cysteine residues at the 112 and 158 sites to a positively charged residue increases its receptor affinity. In contrast to the low binding affinity of ApoE2 to LDLR (~1–2% of ApoE3/4 binding), the binding affinity of ApoE2 to LRP1 is less affected (~40% of ApoE3/4 binding) (Kowal et al., 1990) and the binding affinity to HSPG is not isoform specific (Mahley and Rall, 2000).

Several amyloid-dependent and -independent mechanisms have been postulated to explain the risk-modulating effects of various ApoE isoforms (Figure 1). Here, we will summarize the current data from various in vitro and in vivo studies that support each of these proposed mechanisms prior to focusing on hiPSC based studies in section 4.

Figure 1. ApoE Isoform Specific Effects in AD.

Amyloid-β (Aβ) is produced primarily by neurons in the brain (①) where it interacts with lipidated ApoE secreted by microglia and astrocytes (②) in an isoform specific manner. ApoE4 promotes the oligomerization and aggregation of Aβ (③) and the subsequent deposition of plaques (④) in vivo. Cell surface low-density lipoprotein receptor (LDLR), LDLR-related protein 1 (LRP1) and heparan sulfate proteoglycan (HSPG) receptors mediate the endocytosis of Aβ by astrocytes and microglia. In addition to promoting the production and aggregation of Aβ, ApoE4 impairs Aβ clearance by reducing cellular uptake and transport across the blood brain barrier in vivo (⑤). ApoE2 overexpression enhances the proteolytic degradation of Aβ by insulin-degrading enzyme (IDE) and neprilysin produced by microglia and astrocytes (⑥). The enhanced proteolysis of neuronal ApoE4 by a chymotrypsin-like serine protease produces neurotoxic fragments (⑦). ApoE4-astrocytes exhibit reduced synaptic pruning capacity triggering the synaptic accumulation of the complement-component 1q (C1q) protein, possibly inducing complement pathway mediated neurodegeneration in vivo (⑧). ApoE4 expression also elicits a prolonged increase in pro-inflammatory cytokine secretion by astrocytes and microglia leading to chronic neuroinflammation and neurodegeneration (⑨). ApoE isoform specific roles in these processes are indicated. Figure was generated with the assistance of Biorender.

3.1. Lipidation status and Aβ binding

Conflicting in vitro results obscure the extent of Aβ and ApoE interaction due to variations in purification and detection methods, source of the complex, and the molar ratio of the components (Tai et al., 2014). While some in vitro studies have indicated that the ApoE2 and E3 isoforms are more lipidated than the pathogenic E4 isoform (Bell et al., 2007; Fu et al., 2016), others have found ApoE4 is more lipidated (DeMattos et al., 2001; Kara et al., 2017). Isoform specific differences in the lipidation of ApoE have shown to cause dramatic changes in its structure, affecting its conformation and subsequent Aβ complex formation and receptor interaction (Saito et al., 2001). For instance, analysis of unpurified ApoE protein showed lipidated ApoE3 bound Aβ with higher avidity compared to ApoE4, but this isoform specific effect was abolished upon the purification of the protein, which involves its delipidation (LaDu et al., 1995). Although the isoform specific effect is lost, delipidated ApoE has been reported to bind to Aβ with ~5–10 fold higher affinity than the lipidated forms (Tokuda et al., 2000).

The glial ATP binding cassette proteins ABCA1 and ABCG1 are responsible for cholesterol and phospholipid transfer to ApoE in the CNS (Vance and Hayashi, 2010). It has been demonstrated that ABCA1 deficient astrocytes secrete smaller lipoprotein particles with significantly lower amounts of ApoE, effectively reducing cholesterol efflux (Hirsch-Reinshagen et al., 2004; Wahrle et al., 2004) that is crucial to neuronal health (Valenza et al., 2015, 2010; Zhang and Liu, 2015). In this vein, cholesterol efflux from ApoE3 astrocytes has been shown to be greater than that of ApoE4 astrocytes (Gong et al., 2002). Abca^−/− FAD mice displayed lower ApoE levels but increased amyloid deposition, pointing to the importance of ApoE lipidation in Aβ clearance (Corona et al., 2016; Koldamova et al., 2005; Wahrle et al., 2005). Remarkably, the expression of ApoE4 but not ApoE3 reduced Aβ clearance in this Abca1^−/+ FAD model (Fitz et al., 2012). These conflicting results of ApoE-Aβ binding experiments are likely due to the presence of Aβ in distinct pools in the monomeric, oligomeric and fibrillar forms both unassociated and complexed with ApoE (O’Brien and Wong, 2011; Steinerman et al., 2008). In addition, the size of these pools likely depends on Aβ levels and ApoE isoform specific binding affinity (Tai et al., 2013).

3.2. Aβ production, degradation, deposition and clearance

There is some data that has suggested that ApoE isoforms may differentially regulate Aβ production. For example, ApoE4 has been found to be associated with increased endocytosis of APP and production of Aβ relative to the E2 and E3 isoforms (He et al., 2007; Ye et al., 2005). Along similar lines, in several cell-based models ApoE has been found to increase APP transcription and subsequent Aβ production in the order E4>E3>E2 (Huang et al., 2017; C. Wang et al., 2018). This role of ApoE in APP transcription has been supported by studies that have shown reduced levels of mature full length APP in an ApoE deficient FAD mouse model relative to the wild type control (Dodart et al., 2002).

ApoE isoform-specific differences in Aβ binding may directly result in differential Aβ clearance and subsequent deposition in the brain (Bales et al., 2009; Buttini et al., 2002; Dodart et al., 2005; Dolev and Michaelson, 2004; Fagan et al., 2002; Holtzman et al., 2000). Two main mechanisms result in Aβ clearance from the brain to mitigate deposition—transport across the blood brain barrier and cellular uptake. It has been shown that the clearance of Aβ-ApoE2 and – ApoE3 complexes is more effective through the blood brain barrier (Holtzman et al., 2008). Specifically, these studies demonstrated that Aβ-ApoE2 and -ApoE3 complexes were cleared at a rapid rate by both LRP1 and VLDL receptor (VLDLR) but Aβ-ApoE4 was cleared only through the slow VLDLR. In terms of clearance of Aβ by cellular uptake, ApoE isoforms have been found to influence binding and endocytosis by cell surface receptors on astrocytes where Aβ is degraded in the lysosome (Li et al., 2018). For example, the pathological acidification of endosomes in astrocytes expressing ApoE4, but not ApoE3, lowered surface presentation of the LRP1 which impaired Aβ clearance in vivo (Prasad and Rao, 2018). Additionally, astrocytic LRP1 deficiency increased amyloid deposition in vivo (Liu et al., 2017a), while ApoE deficiency facilitated Aβ clearance (DeMattos et al., 2004). Thus, ApoE isoform specific differences in receptor affinity could affect Aβ clearance due to direct competition (Verghese et al., 2013) or ApoE-Aβ complex preference. The role of the ApoE2 isoform in these mechanisms is unclear and requires further investigation.

In addition to clearance across the blood brain barrier, the degradation of Aβ by proteolytic enzymes such as neprilysin and insulin-degrading enzyme (IDE) secreted by microglia and astrocytes, is essential to its clearance (Farris et al., 2003; Iwata et al., 2000). While neprilysin and IDE deficiency leads to increased Aβ peptide levels and amyloid deposition in vivo (Farris et al., 2007, 2003; Iwata et al., 2001; Miller et al., 2003), their overexpression leads to more efficient clearance of the peptide and reduced amyloid burden (Hemming et al., 2007; Iwata et al., 2004; Leissring et al., 2003; Spencer et al., 2008). Related to these processes, lipidated human ApoE increased Aβ degradation in the order E2>E3>E4 in ApoE deficient microglia by enhancing neprilysin and IDE activity (Jiang et al., 2008).

3.3. Neurotoxicity

Although astrocytes are the primary source for ApoE in the CNS under physiological conditions (Boyles et al., 1985), neurons produce ApoE in response to excitotoxic injury (Xu et al., 2006) and astroglial factors (Harris et al., 2004; Xu et al., 2008). As such, neuronal, but not astrocytic, ApoE4 has been demonstrated to be more likely to undergo proteolysis compared to ApoE3 and produce truncated neurotoxic fragments in AD brains as well as cultured neurons (Brecht et al., 2004; Harris et al., 2003; Huang et al., 2001; Rohn, 2013). More specifically, a chymotrypsin-like serine protease secreted by neurons has been implicated in proteolytic activity that causes neurotoxic ApoE4 fragments to escape the secretory pathway and translocate to the cytosol where they lead to mitochondrial dysfunction and disrupted neurite outgrowth (Harris et al., 2003; Mahley and Huang, 2012; Tamboli et al., 2014). In addition, it has been established that ApoE4 displays a higher propensity to form neurotoxic fibrillar oligomers compared to the E2 and E3 isoforms, which might be responsible the isoform specific differences in amyloid plaque nucleation (Hatters et al., 2006). In a related study, astrocytic ApoE4, but not ApoE3 enhanced plaque seeding by increasing the half-life of Aβ, amyloid deposition and reducing amyloid induced gliosis in mouse brains (Liu et al., 2017b).

3.4. Protection against oxidative stress

It is well established that Aβ induces astrocytic glutamate release and disrupts glutamate uptake by astrocytes and neurons (Li et al., 2009; Matos et al., 2008; Talantova et al., 2013). In vitro studies have suggested that ApoE protects against glutamate-induced toxicity by reducing oxidative stress (Lee et al., 2004; Zhou et al., 2013). Similarly, ApoE displayed an isoform-specific antioxidant activity (E2>E3>E4) in the presence of hydrogen peroxide (Miyata and Smith, 1996). Finally, ApoE deficient mice have elevated levels of oxidative stress and increased antioxidant production, suggestive of ApoE’s protective role in the brain (Shea et al., 2002). However, the isoform-specific effects of ApoE on oxidative stress observed with in vitro studies have not been recapitulated in vivo and have yielded conflicting results (Dose et al., 2016).

3.5. Modulation of neuroinflammation

Post-mortem analysis of brain tissue from AD patients has revealed chronic neuroinflammation (Gomez-Nicola and Boche, 2015). Mechanistically, it has been shown that Aβ induces an inflammatory response in the form of reactive gliosis (Canning et al., 1993; Heurtaux et al., 2010), which has been hypothesized to result in subsequent neurodegeneration (Lucin and Wyss-Coray, 2009). Several studies indicate that ApoE exhibits both anti- and pro-inflammatory activities. In one such study, exogenous ApoE in the presence of Aβ reduced the inflammatory response in glia cells (Guo et al., 2004). However, in the absence of Aβ, ApoE induced the secretion of pro-inflammatory cytokines, with ApoE4 eliciting a greater response than ApoE3. Additional research with in vitro models has shown that this ApoE isoform effect on neuroinflammation is cell-type dependent (Maezawa et al., 2006a, 2006b). Specifically, the inflammatory response in microglia derived from ApoE targeted replacement (TR) mice was greater in E4 microglia (E4>E3>E2) (Maezawa et al., 2006b) but cytokine secretion by astrocytes followed the reverse order, with the highest response elicited in E2 astrocytes (E2>E3>E4) (Maezawa et al., 2006a). By comparison, in vivo studies have not fully recapitulated these context-specific effects of various ApoE isoforms. Instead, in vivo studies have consistently showed that ApoE4 promotes higher levels of inflammation relative to ApoE2 and ApoE3 (Shi et al., 2017; Zhu et al., 2012). Consistent with these findings, compared with patients without any ApoE4 alleles, AD patients with at least one copy of ApoE4 displayed significantly higher levels of the pro-inflammatory cytokines IL1β and IL6 (Olgiati et al., 2010).

3.6. Synaptic plasticity and integrity

Soluble Aβ oligomers have been shown to cause synaptic dysfunction and impair long-term potentiation (LTP) prior to the amyloid deposition (Selkoe, 2002). In healthy brains, the extracellular matrix protein Reelin promotes LTP by binding to the Apoer2-VLDLR complex in postsynaptic neurons and activating NMDA receptors to strengthen synaptic networks (Wasser and Herz, 2017). Moreover, Reelin antagonizes Aβ-mediated LTP suppression that occurs in AD (Durakoglugil et al., 2009). As it relates to the effects of ApoE on this process, in a study with primary cortical neurons ApoE4 reduced the recycling and cell surface expression of Apoer2, impairing Reelin-mediated LTP (Chen et al., 2010). These results were supported by in vivo work where ApoE genotype influenced Aβ-induced LTP suppression in the order E4 > E3 = apoE-KO > E2 (Trommer et al., 2005). Other studies have shown that ApoE can indirectly affect synaptic integrity through the action of astrocytes. For example, Chung and colleagues demonstrated that ApoE controlled the rate of synaptic pruning of astrocytes in an isoform-specific manner E2>E3>E4 (Chung et al., 2016). In turn, it is hypothesized that this defective phagocytic capacity of ApoE4 astrocytes may increase the rate of C1q-coated senescent synapses, thereby leading to increased synaptic vulnerability to complement-cascade mediated neurodegeneration (Chung et al., 2016).

4. Human Induced Pluripotent Stem Cells as a Tool to Model Alzheimer’s Disease

In 2007, Yamanaka and colleagues first described the generation of human induced pluripotent stem cells (hiPSCs) from human somatic cells through retroviral overexpression of transcription factors KLF4, c-MYC, OCT4, and Sox2 (Takahashi et al., 2007). More recently, advances in reprogramming technologies have allowed for the generation of integration-free hiPSCs using sendai viruses (Fusaki et al., 2009), mRNA (Warren et al., 2010), and episomal vector expression (Junying et al., 2009; Okita et al., 2011). HiPSCs provide a nearly unlimited starting material from which specialized cell types that mimic those found in the adult human body can be generated. As such, directed differentiation protocols, largely based on the modulation of the chemical microenvironment, have been developed to generate the various neurons and supporting cell types of the central nervous system (CNS) from hiPSCs. Specifically, robust protocols have been developed for the differentiation of hiPSCs to neurons (Bardy et al., 2015; Begum et al., 2015; Zhang et al., 2013), astrocytes (Shaltouki et al., 2013; Tcw et al., 2017; Zhao et al., 2017), microglia (Abud et al., 2017) and oligodendrocytes (Wang et al., 2013). More recently, overexpression of specific transcription factors in the context of developmentally relevant signaling molecules has allowed for the differentiation of hiPSCs to relatively homogenous neural cell types (Nehme et al., 2018; Sun et al., 2016; Zhang et al., 2013).

Numerous studies have shown that neural cells derived from both fAD and sAD hiPSCs display disease-relevant phenotypes such as elevated secreted Aβ42 levels, increased Aβ42/40 ratio, and hyperphosphorylation of tau at various epitopes (Arber et al., 2019; Israel et al., 2012; Muratore et al., 2014; Ochalek et al., 2017). Together, these pioneering studies laid the groundwork for hiPSC modeling of AD by validating that an accessible, human cell-based platform can be used to study potential mechanisms by which identified mutations modulate detectable disease-relevant phenotypes. Here, we will discuss how hiPSC-based models provide a valuable resource for studying AD-related mechanisms and elucidating the processes by which defined risk factors, such as ApoE, influence disease onset and progression (Table 1).

Table 1.

Summary of studies using hiPSC-derived brain cells to model effects of fAD and sAD mutations and risk factors.

fAD Models
Mutation	Cell Type Analyzed	Phenotype	Reference
PSEN1A246E	Neurons	Increased Aβ42 levels; Decreased by γ-secretase inhibitors	Yagi et al. 2011
PSEN1A246E	Neurons	Increased susceptibility to Aβ toxicity	Armijo et al. 2017
PSEN1A246E	Neurons	Increased Aβ42 levels; Increased Aβ42/40; Robust response to γ-secretase modulators	Liu et al. 2014
PSEN1ΔE9	Neurons	Increased Aβ42/40 through reduction of Aβ40; Altered γ-secretase activity; Increased APP CTFs; No difference in Tau phosphorylation	Woodruff et al. 2013
PSEN1ΔE9	Neurons	Altered sub-cellular distribution of APP; Accumulation of APP CTFs in the soma; Reduced endocytosis and transcytosis of APP and LDL; Impaired Recycling of LRP1	Woodruff et al. 2016
PSEN1ΔE9	Astrocytes	Increased Aβ production, altered cytokine release, dysregulated Ca2+ homeostasis	Oksanen et al. 2017
PSEN1ΔE9	Microglia	Decrease in cytokine release, increase in chemokinesis	Konttinen et al. 2019
PSEN1ΔS169	Neurons	Increased Aβ42 levels; Increased Aβ42/40; Increased levels of Tau phosphorylation; Premature neural differentiation from hNPCs	Yang et al. 2017
PSEN1V89L	Neurons	Increased Aβ42 levels; Increased Aβ42/40; Increased levels of Tau phosphorylation	Ochalek et al. 2017
PSEN1M146L	Neurons	Increased Aβ42 levels; Increased Aβ42/40; Robust response to γ-secretase modulators	Liu et al. 2014
PSEN1M146V	Neurons	Three-fold increase in secreted Aβ42/40 ratio in homozygous and two-fold increase in heterozygous PSEN1M146V mutant knock in lines compared to isogenic controls	Paquet et al. 2016
PSEN1L150P	Neurons	Increased Aβ42 levels; Increased Aβ42/40; Increased levels of Tau phosphorylation	Ochalek et al. 2017
PSEN1H163R	Neurons	Increased Aβ42 levels; Increased Aβ42/40; Robust response to γ-secretase modulators	Liu et al. 2014
PSEN1L166P	Neurons	Partial loss of γ-secretase function; Decreased production of endogenous Aβ40 and an increased Aβ42/40	Koch et al. 2012
PSEN1D385N	Neurons	Catalytically inactive PSEN1; Accumulation of unprocessed full-length protein; Strong decrease in both Aβ42 and Aβ40	Koch et al. 2012
PSEN1 (various mutations)	Neurons, matched patient CSF, postmortem brain tissue	Reduced γ-secretase carboxypeptidase-like activity; Mechanisms include decreased γ-secretase activity, decreased protein stability, and reduced PSEN1 maturation	Arber et al. 2019
PSEN2N141I	Neurons	Increased Aβ42 levels; Decreased by γ-secretase inhibitors	Yagi et al. 2011
PSEN2N141I	Basal forebrain cholinergic neurons (BFCN)	Increased Aβ42/40; Electrophysiological deficits; Phenotypes abolished by CRISPR/Cas9 gene correction	Ortiz-Virumbrales et al. 2017
APPdp	Neurons	Increased Aβ40; Increased levels of Tau phosphorylation and active GSK3β levels; Accumulation of RAB5-positive endosomes	Israel et al. 2012
APPTP	Neurons	Neuron-specific Aβ peptide production; Increased Aβ42 aggregate formation; Altered Tau phosphorylation and localization; Increased cell death	Shi et al. 2014
APPTP	Neurons	Inactivating one copy of the APP gene in a trisomy 21 background normalized APP, secreted levels of Aβ42, and Aβ42/Aβ40; Alteration did not decrease Tau phosphorylation	Ovchinnikov et al. 2018
APPV717I	Forebrain neurons	Increased Aβ42 levels; Increased levels of Tau phosphorylation; Altered APP subcellular localization; Altered APP processing	Muratore et al. 2014
APPV717L	Cortical neurons	Increased Aβ42 levels; Increased Aβ42/40	Kondo et al. 2013
APPV717F	Neurons	Defective transcytosis of APP and lipoproteins	Woodruff et al. 2016
APPswe	Neurons	Defective transcytosis of APP and lipoproteins	Woodruff et al. 2016
APPswe	Microglia	Decrease in cytokine release, increase in chemokinesis	Konttinen et al. 2019
sAD Models
Risk factor	Cell Type Analyzed	Phenotype	Reference
N/A	Neurons	Increased Aβ40 levels; Increased Tau phosphorylation; Increased active GSK3β	Israel et al. 2012
N/A	Neurons	Elevated Tau hyperphosphorylation; Increased amyloid levels; increased GSK3β activation	Ochalek et al. 2017
APOE4	Neurons, astrocytes, microglia	Neurons: increased Aβ42, increased synapse number Astrocytes: impaired Aβ uptake, cholesterol accumulation Microglia: reduced Aβ phagocytosis	Lin et al. 2018
APOE4	Neurons	Increased APP transcription and Aβ synthesis mediated by non-canonical MAP kinase pathway in the rank order ApoE4>ApoE3>ApoE2	Huang et al. 2017
APOE4	Neurons	Increased synaptogenesis mediated by CREB activation in the rank order ApoE4>ApoE3>ApoE2	Huang et al. 2019
APOE4	Neurons	Increased Aβ levels; Increased Tau phosphorylation; GABAergic neuron degeneration	Wang et al. 2018
APOE4	Neurons	Calcium dysregulation and associated neurodegeneration; higher levels of cellular and secreted hyperphosphorylated Tau	Wadhwani et al. 2019
Mutation	HiPSC-derived Cell Type Analyzed	Phenotype	Reference
APOE4	Microglia	Decreased Metabolism, phagocytosis, and migration in APOE4 microglia-like cells	Konttinen et al. 2019
APOE4	Astrocytes	Decreased neurotrophic function	Zhao et al. 2017
TREM2, APOE	Microglia	Phenotypic switch from homeostatic to neurodegenerative phenotype via TREM2-APOE pathway	Krasemann et al. 2017

4.1. Genetic modification of hiPSCs for disease modeling

To date, numerous hiPSC lines have been developed from patients with diverse genetic variants and repositories have been established to provide researchers access these cell lines for disease modeling (e.g. CIRM, WiCell, NYSCF, Coriell). However, genetic diversity among patient samples may make it difficult to tease apart observed phenotypic differences because of their inherent genetic and epigenetic diversity. To this end, the development of genome engineering tools has greatly increased the utility of hiPSCs by allowing targeted modifications without altering the genetic background (Brookhouser et al., 2017a). Specifically, the CRISPR/Cas9 system has provided major advances in efficient genome engineering of mammalian cells allowing for editing at single base pair resolution (Cong et al., 2013; Paquet et al., 2016). It is estimated that over 10,000 human diseases are caused by monogenic mutations (World Health Organization), allowing the possibility of inducing or correcting disease-relevant gene mutations and comparing to the parental, or isogenic, cell line. As we will discuss, the use of such isogenic cell lines with identical genetic background has become the gold standard in modeling and analyzing the effects of AD-related mutations and risk factors.

4.2. Modeling of fAD with hiPSC-based models

Numerous studies have employed hiPSCs derived from patients with fAD-related mutations in PSEN1/2 and APP to demonstrate the utility of such models in recapitulating disease-relevant phenotypes in vitro. As it relates to PSEN1/2, hiPSC-derived neurons from fAD patients harboring various mutations in PSEN1 (ΔE9 (Woodruff et al., 2013), V89L (Nemes et al., 2016), M146L (Liu et al., 2014), M146V (Paquet et al., 2016), L150P (Ochalek et al., 2017), H163R (Liu et al., 2014), L166P (Koch et al., 2012), ΔS169 (Yang et al., 2017), A246E (Armijo et al., 2017; Liu et al., 2014; Yagi et al., 2011), D385N (Koch et al., 2012)), and PSEN2 (N141I (Yagi et al., 2011; Yu et al., 2010)) have been analyzed for disease-relevant phenotypes. While analysis of cells from all PSEN1/2 mutations revealed an increased Aβ42/40 ratio compared to control lines, only the L166P and ΔE9 increased this ratio by decreasing Aβ40 whereas for all other mutations an increase in Aβ42 was responsible for an elevated ratio. As such, through the study of neurons from these various hiPSC lines it has been suggested that PSEN1/2 mutations might be context-specific with some mutations causing a toxic gain-of-function through elevated Aβ42 production and others inducing a loss-of-function through decreased Aβ40 generation. In addition, further phenotypic characterization of cells revealed other disease-specific phenotypes including higher levels of phosphorylated tau (Ochalek et al., 2017; Yang et al., 2017), sensitivity to neurotoxic stimuli (Armijo et al., 2017), and electrophysiological deficits (Ortiz-Virumbrales et al., 2017). Importantly, in some of these studies genetic engineering approaches were used to correct these disease-causing mutations to establish direct causal links between genotype and phenotype (Ortiz-Virumbrales et al., 2017; Woodruff et al., 2013).

With regards to familial mutations in APP, neurons generated from hiPSCs with various APP mutations including missense (V717I (Muratore et al., 2014), V717L (Shirotani et al., 2017)), truncation (E393Δ (Kondo et al., 2013)), duplication (DP (Israel et al., 2012)), and triplication (TP (Ovchinnikov et al., 2018; Shi et al., 2012)) have been extensively phenotypically characterized. For example, Israel et al. (Israel et al., 2012) demonstrated that neurons generated from hiPSCs with a duplication in the APP gene (APP^dp) exhibited higher levels of Aβ, hyperphosphorylated tau, and increased activity of the GSK3β kinase compared to healthy control lines. Critically, this study validated the production of detectable disease phenotypes in hiPSC-derived neurons within weeks of culture, despite the relatively long time required for phenotypes to become evident in AD patients. Moreover, treatment of APP^dp neurons with β-secretase (BACE) inhibitors reduced tau phenotypes and reduced GSK3β activity (Israel et al., 2012).

In a similar study, neurons generated from APP^V717I hiPSCs displayed elevated Aβ42/40 levels with a corresponding increase in phosphorylated tau that was reduced through γ-secretase inhibition (Moore et al., 2015). Likewise, treatment of APP^V717I hiPSC derived neurons with Aβ-specific antibodies ameliorated elevated tau levels (Muratore et al., 2014). Together, these studies establish a causal relationship between APP processing and tau-related phenotypes. Alternatively, a recent study highlighted that genome modification to restore a euploid karyotype in neurons generated from an individual with Down syndrome rescued Aβ phenotypes due to APP gene dose, however, was insufficient to reduce tau hyperphosphorylation (Ovchinnikov et al., 2018). These data challenge the proposed model that increased Aβ is directly responsible for downstream tau pathology, suggesting that a more complex mechanism may govern disease progression.

HiPSC-based systems have also allowed for the investigation of the precise molecular mechanisms that lead to these mutation-specific phenotypes. For example, neurons derived from APP^V717I mutant hiPSCs demonstrated abnormal subcellular localization of APP within acidic early endosomes containing active BACE, modulating APP processing by β- and γ-secretase and potentially driving increased levels of Aβ and phosphorylated tau (Muratore et al., 2014). In another study, it has been shown that APP^dp hiPSC-derived neurons exhibit a similar early endosome accumulation phenotype that may be driven by increased β-CTF levels (Israel et al., 2012). Similarly, analysis of neurons from fAD-hiPSC with mutations such as PSEN1^ΔE9, APP^V717F, and APP^swe revealed increased β-CTF generation and impairment of lipoprotein endocytosis and transcytosis to the axon, suggesting that fAD mutations may result in convergence on common molecular pathways (Woodruff et al., 2016). On the other hand, a recent phenotypic screen of seven diverse fAD hiPSC lines revealed that APP mutations altered γ-secretase cleavage site preference whereas PSEN1 mutations altered amyloid peptide profile through several mechanisms including decreased γ-secretase activity, decreased protein stability, and reduced PSEN1 maturation (Arber et al., 2019).

Non-neuronal cells, specifically astrocytes and microglia, play an important role in maintaining brain homeostasis and aberrant function of these effector cells has been shown to be a major contributor to brain dysfunction in AD (Colombo and Farina, 2016; Dzamba et al., 2016; Sofroniew and Vinters, 2010). Because hiPSCs can be directed into precise cell types, hiPSC-based models have been employed to investigate cell type-specific effects in fAD. For example, Oksanen et al. demonstrated that hiPSC-derived astrocytes with a PSEN1^ΔE9 point mutation exhibited AD-relevant phenotypes including increased levels of Aβ production and decreased Aβ clearance, as well as altered cytokine secretion, mitochondrial metabolism, and Ca²⁺ homeostasis (Oksanen et al., 2017). Importantly, these phenotypes were reversed upon isogenic correction of the PSEN1 mutation, demonstrating a direct genotype-to-phenotype relationship. In another study, astrocytes from several isogenic hiPSCs lines harboring mutations in APP were examined (Fong et al., 2018). Notably, astrocytes derived from hiPSCs homozygous for the APP Swedish (APP^Swe/Swe) mutation displayed Aβ endocytosis defects due to increased β-secretase cleavage whereas astrocytes with other APP mutations, such as V717F, did not display these deficits.

The use of hiPSCs to study ApoE in the context of these fAD related mutations has been limited. To date, the majority of studies have primarily employed lines homozygous for the ε3 allele or of an unreported genotype. In the future, the use of genome editing technologies to generate fAD hiPSC lines with various ApoE isoforms will allow for the determination of the extent to which the introduction of various APOE alleles modulates disease-related phenotypes.

4.3. Investigating effects of APOE genotype in AD with hiPSC models

Broadly speaking, the presence of disease-related phenotypes in cells generated from sAD hiPSCs have been highly variable (Duan et al., 2014; Israel et al., 2012). For example, one of the earliest iPSC models of sAD compared neurons derived from two sAD patients with iPSC-derived neurons of two fAD and non-demented control individuals (Israel et al., 2012). Neurons derived from one of the two sAD lines used in this study exhibited higher levels of Aβ40, p-Tau and active glycogen synthase kinase-3β (GSK3β), similar to the phenotype observed in fAD derived neuronal cultures. In a related study, analysis of neurons derived from several independent sAD hiPSC lines revealed elevated Aβ42/40 ratios only in a subset of the lines (Duan et al., 2014). Despite the variability in observed phenotypes in sAD hiPSC models, these studies provided proof-of-principle for using hiPSC-based models to investigate the role of various AD risk factors, such as ApoE, in a simplified and accessible system.

4.3.1. Phenotypic characterization of hiPSC lines with various ApoE genotypes

Over the past several years, numerous hiPSC lines have been generated from individuals with various ApoE genotypes (Brookhouser et al., 2018, 2017b, 2017c; Peitz et al., 2018; C. Wang et al., 2018; Y. Wang et al., 2018; Zhao et al., 2017; Zollo et al., 2017; Zulfiqar et al., 2016a, 2016b). Detailed biochemical analysis and phenotypic characterization of neural cells derived from these hiPSC lines have revealed several interesting ApoE isoform-specific effects. In one such study, neurons were derived from hiPSCs homozygous for APOE4 and APOE3 (C. Wang et al., 2018). This analysis revealed that Aβ secretion, p-Tau levels, and GABAergic neuron degeneration were elevated in APOE4/4 neurons when compared to APOE3/3 neurons. Moreover, treatment of these cells with β- or γ-secretase inhibitors decreased Aβ40 and Aβ42 levels in the culture medium while phosphorylated tau levels were unaffected, suggesting p-Tau accumulation occurs in an Aβ-independent manner. A related study attempted to use hiPSC-derived neurons to examine the mechanism behind this isoform-specific effect on Aβ levels (Huang et al., 2017). Specifically, Huang et al. demonstrated that ApoE binding to neuronal ApoE receptors activated a non-canonical MAPK pathway which led to cFos phosphorylation. In turn, the transcription factor activating protein-1 (AP-1) became activated, which increased expression of APP and, subsequently, Aβ levels in the rank order ApoE4>ApoE3>ApoE2. In a complementary study, Zollo et al. demonstrated that ApoE modulated sortilin-receptor (SORL1)-related APP trafficking and processing (Zollo et al., 2017). Specifically, the authors showed that in APOE4 neurons there was increased Aβ/SORL1 localization along the degenerated neurites when compared to neurons without an APOE4 allele. Moreover, SORL1 binding to APP was compromised in these APOE4 neurons which contributed to elevation of secreted Aβ.

Additional studies have revealed other neuronal phenotypes associated with various ApoE isoforms. For example, when co-cultured with neurons, APOE3/3 hiPSC-derived astrocytes provided neuroprotective effects when compared to neurons cultured with APOE4/4 astrocytes (Zhao et al., 2017). In addition, APOE genotype influenced synaptic integrity as demonstrated by the increased expression of synaptic proteins in neurons co-cultured with APOE3/3 astrocytes relative to those cells co-cultured with APOE4/4 astrocytes (Zhao et al., 2017). Mechanistically, a parallel study of hiPSC-derived neurons treated with recombinant ApoE suggests that this altered synaptogenesis is mediated by CREB activation in the rank order ApoE4>ApoE3>ApoE2 (Huang et al., 2019).

HiPSC-based models have also been used to examine the effect of APOE genotype in other neural cell types. For example, Zhao et al. characterized the lipidation status of ApoE secreted from astrocytes differentiated from APOE3/3 and APOE4/4 hiPSCs (Zhao et al., 2017). This analysis revealed that ApoE produced by APOE3/3 astrocytes was more lipidated than that produced by APOE4/4 cells. The authors speculate that this reduced cholesterol transport from APOE4/4 astrocytes could be detrimental to neuronal health. In a separate study, microglia differentiated from a library of hiPSCs with various fAD mutations and APOE genotypes were examined. These experiments showed that microglia derived from APOE4 iPSCs demonstrated aberrant functional phenotypes including decreased chemokinesis, reduced phagocytosis and aggravated cytokine response to inflammatory stimuli compared to microglia derived from hiPSCs harboring fAD mutations (Konttinen et al., 2019).

4.3.2. Engineering hiPSCs to elucidate APOE genotype-to-phenotype relationships

Although the analysis of hiPSC lines derived from patients with various APOE genotypes has revealed some potential isoform-specific effects, the analysis of these phenotypic effects has been somewhat confounded by the genetic and epigenetic differences inherent in individual hiPSC lines derived from distinct patients. To that end, advances in genome editing technologies has allowed for the generation of isogenic hiPSC lines from both NDC and AD patients that will only differ with respect to their APOE genotype and not their genetic background.

In a landmark study using CRISPR/Cas9-edited hiPSCs, Lin et al. examined the effect of APOE genotype in hiPSC-derived neurons, astrocytes, and microglia (Lin et al., 2018). Relative to the APOE3/3 counterparts, APOE4/4 neurons had a higher number of synapses, more early endosomes, and increased Aβ42 secretion. In addition, APOE4/4 astrocytes demonstrated less efficient Aβ clearance, lower ApoE levels, and accumulation of cholesterol. Likewise, APOE4/4 microglia had impaired Aβ42 uptake while also displaying an altered inflammatory profile. Moreover, transcriptome profiling of all three cell types was consistent with these phenotypes as the ApoE variant modulated the expression of several pathways associated with synaptic function, lipid metabolism, and immune response. Finally, isogenic conversion of sAD hiPSCs from APOE4/4 to APOE3/3 reversed these defects, as observed by the reduced number of synapses in neurons, as well as restored ApoE levels and Aβ uptake capacity in astrocytes and microglia. A separate study employing isogenic lines also confirmed many of these isoform-dependent effects (C. Wang et al., 2018). Specifically, conversion of APOE4 hiPSCs to an APOE3 genotype increased the levels of full-length ApoE in neuronal lysates and decreased Aβ42 secreted into the culture media. In addition to these effects, this analysis also revealed that this conversion resulted in fewer phosphorylated tau-positive GABAergic neurons. Finally, the introduction of the APOE3 alleles rescued the degeneration of GABAergic neurons observed in in APOE4/4 neuronal cultures, as measured by increased levels the synaptic marker GAD67.

The use of genetically modified hiPSCs lines have also allowed for the identification of novel target pathways that are influenced by APOE genotype. For example, Meyer et al. report that accelerated differentiation and reduced progenitor cell self-renewal was observed in neural cultures derived from APOE4/4 hiPSCs (Meyer et al., 2019). In addition, the authors demonstrated that the function of the transcriptional repressor REST was impaired in APOE4/4 cells. Moreover, the loss of REST function was attributed to reduced nuclear translocation and chromatin binding. Importantly, gene editing of these cells to an APOE3/3 genotype reversed these observed phenotypes. Because REST plays a role in the repression of neuronal genes during early nervous system development, the authors speculated that ApoE4-depdent dysregulation of REST may reduce hippocampal adult neurogenesis through depletion of the endogenous neural progenitor pool. In turn, this effect may accelerate the onset of AD by reducing this regenerative capacity.

5. Future Prospects and Emerging Trends: Improving hiPSC-Based Models

As discussed in this review, advances in cellular reprogramming have enabled the generation of in vitro models of AD that can be used to dissect disease mechanisms and evaluate potential therapeutics. However, neurons generated from these hiPSCs have shown some, but not all, of the early molecular and cellular hallmarks associated with the disease. Additionally, phenotypes and pathological hallmarks associated with later stages of the human disease have not been observed with current hiPSC-based systems. In general, current hiPSC-based models of AD are limited by the use of immature, heterogeneous neuronal populations in a 2-D microenvironment that does not mimic that of native brain tissue. Here, we will discuss emerging trends that will significantly improve the use of hiPSC-based models to investigate the role of ApoE in modulating AD onset and progression.

5.1. Cell-selective vulnerability

The central nervous system (CNS) is composed of many different neuronal subtypes, each with defined anatomical locations (e.g. forebrain, midbrain, hindbrain, spinal cord) and specialized functions (e.g. sensory, motor, communication, computation). Many neurodegenerative disorders are characterized by selective neuronal death—neurons of one regional subtype are completely unaffected while neurons of another regional subtype become diseased (Jackson, 2014; Saxena and Caroni, 2011). Specifically, the late-stage cognitive consequences of AD have been directly related to the loss of synapses and neurons in the cortex and in particular basal forebrain cholinergic neurons (BFCNs), which are the predominant source of cortical input and play a critical role in spatial learning and memory (Mesulam, 2004; Mufson et al., 2002). In fact, research has shown that the ability to use hiPSC-derived cells to model neurodegenerative disease is enhanced by the ability to generate neurons of the affected subtype as disease-related phenotypes are absent or significantly diminished in unaffected neuronal subtypes (Muratore et al., 2017). Specifically, Muratore et al. found that APP is differentially processed in hiPSC-derived forebrain neurons when compared to those of other regional identities (Muratore et al., 2017). In turn, phenotypes such as elevated Aβ42 secretion, increased levels of phosphorylated tau, and sensitivity to neurotoxic stimuli were only observed in fAD hiPSC-derived neurons of forebrain identity. Moving forward, the ability to analyze neurons of cortical identity will be critical to using hiPSC-derived neurons to study the age-related mechanisms of ApoE in the context AD.

5.2. Three-dimensional (3-D) culture

It is well established that in vivo cells reside within a complex three dimensional (3-D) microenvironment that plays a significant role in regulating cell behavior (Schwartz and Chen, 2013). Signaling and other cellular functions, such as gene expression and differentiation potential, differ in 3D cultures compared with 2D substrates (Cukierman et al., 2002, 2001; Pineda et al., 2013). Nonetheless, previous studies using AD hiPSCs have relied on 2-D neuronal culture models that do not reflect the 3-D complexity of native brain tissue, and therefore, are unable to replicate all aspects of AD pathogenesis. Several methods have emerged have emerged to model complex 3-D neural tissue physiology in vitro (D’Avanzo et al., 2015; Griffith and Swartz, 2006; Lancaster et al., 2013; Li et al., 2012). In one such study, Choi et al. developed a Matrigel-based 3-D system to model AD by overexpressing APP and PSEN1 mutations in immortalized neural stem cells (NSCs) isolated from the ventral mesencephalon region of the human fetal brain (Choi et al., 2014). In this 3-D model, the authors were able to induce robust extracellular deposition of amyloid-β, incluidng amyloid-β plaques, as well high levels of detergent-resistant aggregates of p-Tau. As an alternative, organoid-based methods have emerged as a robust and adaptable platform to model human brain development and complexity (Amin and Paşca, 2018; Huch and Koo, 2015; Lancaster et al., 2013; Paşca et al., 2015). Critically, 3-D cerebral organoids have shown to be advantageous in allowing for greater cell maturation over longer in vitro culture periods (Amin and Paşca, 2018). For example, initial studies revealed that astrocytes within cortical organoids began to resemble postnatal primary astrocytes within 9–10 months in culture and continued to mature within the organoids up to 20 months in culture (Sloan et al., 2017). It may be possible to leverage these features of 3-D hiPSC-based culture systems to develop physiologically relevant models to study the effects of risk factors, such as APOE. Additionally, further optimization of these systems to develop vascularized models will allow for better modeling of the role of ApoE in blood-brain-barrier (BBB) interactions and Aβ clearance (Mantle and Lee, 2018).

5.3. Introduction of age-related features into hiPSC-based models

One of the main limitations of hiPSC-based systems to model age-related diseases such as AD is that the age-associated features of patient cells are lost during the reprogramming process (Freije and López-Otín, 2012; Mahmoudi and Brunet, 2012; Miller et al., 2013). Consequently, genome-wide expression analysis performed by our laboratory and others (Mariani et al., 2012) demonstrated that hiPSC-derived neurons more closely resemble fetal rather than adult neurons. A recent study by Mertens et al. revealed that bypassing the hiPSC state and directly reprogramming patient fibroblasts into induced neurons (iN) results in cells that retained age-related phenotypes (Mertens et al., 2015). In the future, this powerful approach to generate iNs from AD patient samples may provide an in vitro model of the end-stage of the disease and allow for study of disease progression.

Alternatively, hiPSC models of premature aging disorders have revealed that despite the reset of age-related phenotypes in hiPSCs, aging-associated characteristics can reemerge in several differentiated cell types such as mesenchymal stem cells, vascular smooth muscle cells, and fibroblasts (Ho et al., 2011; Liu et al., 2012, 2011a, 2011b; Studer et al., 2015; Zhang et al., 2011). Based upon these studies, overexpression of progerin, the mutant form of lamin A that causes Hutchinson-Gilford progeria syndrome (HGPS), was found to induce age-associated phenotypes in Parkinson’s disease hiPSC-derived dopaminergic neurons (Miller et al., 2013). The continued development of these methods to efficiently induce aging in vitro may be instrumental in addressing this major limitation of studying age-associated disease using hiPSC-based model systems.

Interestingly, several studies have shown an association between APOE status and age in both familial and sporadic forms of the disease. Corder et al. first reported that an increase in the number of APOE ε4 alleles not only increased the risk for sporadic late onset AD, but also reduced the mean age of onset from 84 to 58 years (Corder et al., 1993). Further, a population study of individuals from the Columbian PSEN1 E280A cohort demonstrated that 50% of the APOE4 carriers who also harbored the FAD mutation in PSEN1, developed AD 3 years earlier than non-APOE4 carriers (Pastor et al., 2003). Additionally, a longitudinal study in cognitively normal subjects found that memory decline began earlier and proceeded with greater acceleration in ε4 carriers relative to noncarriers (Caselli et al., 2009). In the future, in order to dissect the contribution of the APOE4 genotype to earlier disease onset, time course studies of isogenic hiPSC-derived organoids may be employed to study mechanisms by which APOE4 and age influence disease progression.

5.4. Sex differences in AD and hiPSC-based models

Sex and gender difference is another risk factor of AD, with about two-thirds of the diagnosed adults in the United States being women (“2018 Alzheimer’s disease facts and figures,” 2018; Hebert et al., 2013). Increased longevity only partially explains the sex and gender effect on frequency and lifetime risk of AD in women (Andersen et al., 1999; Corder et al., 2004; Johansson, 1989; Nebel et al., 2018). Additionally, studies have shown female APOE ε4 carriers are more likely to develop mild cognitive impairment (MCI) and AD relative their male counterparts (Altmann et al., 2014; Ungar et al., 2014). Fleisher et al. found that hippocampal atrophy was significant in women who had either one or two ε4 alleles relative to non-carriers, but the presence of two ε4 alleles was required for a significant effect in men (Fleisher et al., 2005). Interestingly, female APP/PS1/tau triple-transgenic AD mice (3×Tg-AD mice) had increased amyloid deposition, neurofibrillary tangles and neuroinflammation compared to male 3×Tg-AD mice (Yang et al., 2018). Moving forward, the inclusion of hiPSC lines derived from both male and female patients is therefore necessary to model AD and gain mechanistic insight into the role of ApoE in disease pathology.

5.5. Advances in gene editing approaches

Current CRISPR-based approaches to generate isogenic hiPSC lines require the introduction of deleterious double stranded DNA breaks (DSBs) followed by inefficient homology directed repair (HDR). In addition, extensive research demonstrates that the introduction of DSBs can result in chromosomal translocations, apoptosis, and acquisition of potentially oncogenic mutations (Chapman et al., 2012; Haapaniemi et al., 2018; Ihry et al., 2018; Kosicki et al., 2018). The recent development of base editing technologies that employ a nicking Cas9 endonuclease fused to a cytidine or adenosine deaminase allow for the generation of cell lines free of potentially deleterious DSB-mediated genome editing and does not require the use of HDR (Gaudelli et al., 2017; Komor et al., 2018, 2016). In fact, these approaches were recently applied to modify the APOE allele in hiPSCs (Standage-Beier et al., 2019). A remarkable study of an individual bearing the PSEN1 E280A fAD mutation recently reported that the onset of mild cognitive impairment (MCI) was delayed to her seventies, unlike other mutants who developed MCI symptoms by age 44 (Arboleda-Velasquez et al., 2019). The delayed onset of AD symptoms, despite an unusually high Aβ plaque burden, was attributed to homozygosity for the rare Christchurch (R136S) mutation in the receptor binding domain of ApoE3. Continued expansion of the CRISPR toolbox to precisely modify hiPSCs at such loci will allow for investigation of the complex genetic mechanisms by which ApoE modulates disease-related processes. In addition, multiplex CRIPSR systems could potentially be employed to edit multiple independent genomic loci to study the interaction between ApoE and other risk variants such as ABAC1 (Rodríguez-Rodríguez et al., 2007), CR1 (Keenan et al., 2012), and TREM2 (Wolfe et al., 2018)).

5.6. Identification of therapeutic interventions

Various ApoE-targeting therapeutic strategies have been examined to reduce the risk conferred by ApoE in AD including structure correction (Brodbeck et al., 2011; Chen et al., 2012), blocking its interaction with Aβ (Kuszczyk et al., 2013; Pankiewicz et al., 2014; Sadowski et al., 2006), modifying its expression levels (Bien-Ly et al., 2012; Cramer et al., 2012; Huynh et al., 2017), lipidation state (Rawat et al., 2019; Wahrle et al., 2008) and targeting ApoE aggregates (Liao et al., 2018). To this end, studies have shown that both fAD and sAD hiPSC-derived neurons have been useful as a tool to screen compounds for disease-modifying agents in preclinical development (Kondo et al., 2017, 2013). For example, Inoue and colleagues recently developed a platform utilizing hiPSC-derived neurons for the high-throughput screening of pharmaceutical compounds on Aβ production (Kondo et al., 2017). Using this system, more than two dozen compounds were identified as hits for Aβ-modulating compounds. Subsequent analysis identified a synergistic cocktail of compound hits that significantly reduced amyloid production in both fAD and sAD cultures. As it relates to ApoE, given its importance in clearing Aβ from the brain, coupled with evidence that the binding affinity of ApoE to Aβ is inversely correlated with AD risk (Liu et al., 2013), augmenting the functions of ApoE has been explored as a potential therapeutic strategy to treat AD (Fan et al., 2016; Finan et al., 2016). In this vein, a recent study employed hiPSC-derived neurons to evaluate the effects of ApoE modulating molecules on AD-related phenotypes (C. Wang et al., 2018). To that end, one small molecule, PH002, that modified ApoE4 to structurally and functionally mimic ApoE3 was able to reverse the detrimental effects of ApoE4 in hiPSC-derived neurons. More precisely, treatment of APOE4/4 hiPSC-derived neurons with PH002 decreased Aβ42 production, reduced phosphorylated tau levels, and increased GABAergic neuron numbers in a dose-dependent manner. In the future, similar screens and subsequent analysis of the pathways modulated by hit compounds may shed light onto potential mechanisms by which ApoE contributes to disease onset and progression.

6. Concluding Remarks

In sum, hiPSC technology has been an invaluable tool for studying the complex genetics associated with AD that cannot be recapitulated in animal models (Drummond and Wisniewski, 2017). Alternatively, hiPSCs provide an accessible human system for studying disease characteristics in all major brain cell types through directed differentiation protocols and co-culture systems. Continued use and improvement of such models will reveal the direct effect of the presence of specific APOE alleles on the manifestation of AD-related phenotypes as well as potential signaling pathways and transcriptional targets that are independently influenced by APOE genotype and disease status. Overall, the ability to identify definitive relationships between APOE genotype and AD-related phenotypes will have a significant translational impact on the design of molecularly targeted therapies to treat the many patients suffering from AD.