[1] 12w The Use of Patient-Derived Induced Pluripotent Stem Cells for Alzheimer's Disease Modeling

[1] 94w This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

[1] 189w Alzheimer's disease (AD) is a chronic neurological disorder that is marked by progressive and irreversible neurodegeneration. It represents the most common cause of dementiaa clinical syndrome characterized by a pathological decline in multiple facets of cognitive functioning, including memory, language and behaviour (van der Flier and Scheltens, 2005). Age is by far the largest risk factor for AD, and with the average age of the human population rising, AD now poses a massive threat to public health (Guerreiro and Bras, 2015). The global prevalence of dementia is expected to rise from 50 million in 2018 to over 80 million by 2030, while worldwide cost is estimated to rise from 1 to 2 trillion USD (Patterson, 2018). Currently, there is no cure for AD or method to slow the rate of cognitive decline, and all treatments target symptomatic relief and improving quality of life (Weller and Budson, 2018). The abysmal success rate for new AD drugs despite their success in preclinical models critically illustrate the need for better disease modeling, with improved methods for drug screening and evaluation starting with the in vitro testing phase (Cummings, 2018;Zahs and Ashe, 2010).

[2] 307w Neuropathologically, AD is a heterogeneous condition characterized by two cardinal features: the formation of extracellular amyloid plaques and intracellular neurofibrillary tangles (NFTs) (Stefani and Dobson, 2003). Amyloid plaques are deposits composed mainly of -amyloid (A) peptides which are mostly released from neurons via proteolytic cleavage of the transmembrane amyloid precursor protein (APP) (O'Brien and Wong, 2011). NFTs are mostly composed of aggregates of hyperphosphorylated tau protein (Metaxas and Kempf, 2016). Despite amyloid plaques and NFTs being identified over 110 years ago, the underlying relationship between their presence and the etiology AD remains unclear (Bondi et al., 2017). A wealth of pre-clinical and clinical research strongly implicates A in the pathophysiology of AD, reflected in the A cascade hypothesis first J o u r n a l P r e -p r o o f conceived in 1984 (Glenner et al., 1984). Refined versions of this hypothesis suggest that the formation of A oligomers is the key pathological event in AD, leading to a cascade of plaque formation, neuronal dysfunction, and cell death (Selkoe and Hardy, 2016). Emerging longitudinal clinical data from the Alzheimer Disease Neuroimaging Initiative as well as the Dominantly Inherited Alzheimer Network indicate that A is the first pathology to accumulate in AD, followed by other abnormalities, including a rise in NFTs (McDade et al., 2018;Young et al., 2014). Despite strong evidence suggesting a causal relationship between A and AD, targeting brain amyloid has thus far not had a meaningful clinical impact in several phase 3 trials in patients ranging from pre-symptomatic at-risk, to Mild Cognitive Impairment and Mild-Moderate AD (Henley et al., 2019;Honig et al., 2018;Salloway et al., 2014;Selkoe, 2019). While there are many aspects of these failures to consider, one highlight is the enduring translational gap between preclinical disease modeling and clinical testing, especially with the prominent use of rodent models of disease.

[3] 48w In part due to the emerging challenges with A therapy, NFTs are now front and center for therapeutic development. Favoring NFTs as a drug target is the better correlation with AD disease stage and progression, which has notably been lacking for A plaque load (Nelson et al., 2012).

[4] 86w Preclinical data support the hypothesis that NFTs play a critical role in AD (Nelson et al., 2012), and numerous compounds targeting tau are currently in clinical testing (NCT 03375697, NCT 03019536, NCT 02579252). Nevertheless, the precise mechanisms and pathogenic roles of tau and its hyperphosphorylation in AD are not fully elucidated (Du et al., 2018). Moreover, data used to support current drug development are largely based on similar preclinical modeling strategies as for failed A drugs, again highlighting the need for better translational models of disease.

[5] 41w Beyond A and tau, there are numerous candidate AD processes being explored for drug development including glucose hypometabolism, immune pathways, the gut-microbiota, and other cellular processes implicated with AD (Du et al., 2018). However, our understanding of AD pathophysiology remains incomplete.

[6] 49w Diagnosis of AD is generally separated into autosomal-dominantly inherited familial cases (fAD) and sporadic cases (sAD) (Piaceri et al., 2013). sAD is by far the most prevalent form of the disease, with fAD making up less than 1% of all AD cases (Piaceri et al., 2013;Ryan et al., 2016).

[7] 182w Most cases of fAD can be attributed to germline mutations in one of three genes: APP, presenilin 1 (PSEN1) and presenilin 2 (PSEN2). sAD however, typically does not have a single distinct genetic cause, but does show genetic predisposition in terms of heritability and age of onset (Gatz et al., 2006). Polymorphisms in the APOE gene, encoding for apolipoprotein E (ApoE), is the J o u r n a l P r e -p r o o f major genetic risk factor for sAD, with a growing list of other genetic loci each affording minor risk (Karch and Goate, 2015). Polygenic hazard scores (PHS) compiled for sAD have revealed that patient-specific mutational profiles can greatly influence age of onset, rate of clinical decline, and neuropathophysiological hallmarks of AD (Desikan et al., 2017;Tan et al., 2019). The monogenic background of fAD has made it an easier model for recreating the disease both in vitro and in vivo, and as such, most research has focused on fAD. However, further study on both fAD and sAD must occur to further our understanding of the disease.

[1] 247w In order to identify the bulk of studies that have involved patient-derived iPSCs, an extensive literature review was completed, a workflow for this can be seen in Figure 1. PubMed and PubMed Central (PMC) were searched with specific parameters, while Google Scholar was searched with a variety of parameters. Studies that were published simply to establish iPSC cell lines were omitted from the final list. PMC was searched using the parameters "(Alzheimer[abstract] OR Alzheimer's[abstract]) AND (induced pluripotent stem cells OR hipsc OR ipsc)". This search yielded 930 results, out of which 292 were selected based on title and/or abstract. Full text review reduced this to 43 articles. Pubmed was searched using the parameters "(Alzheimer OR Alzheimer's) AND (induced pluripotent stem cells OR hipsc OR ipsc)" for papers published during or after 2011. This search yielded 416 results, out of which 98 were selected based on title and/or abstract; PMC articles were excluded during this step as well as it was assumed that they were already selected. Full text review reduced this to 17 articles. Google scholar search was conducted with a variety of terms and yielded 2 additional papers. Articles were separated based on their models investigating fAD-related mutations, sAD-related mutations or both (i.e. PSEN1 mutations were included in the fAD section despite not explicitly originating from fAD patients); these lists can be found on Table 1, Table 2, and Table 3, respectively. Research articles identified in the 3D model section were not searched with specific parameters.

[2] 280w The monogenic cause of fAD makes it an attractive option for the modeling of AD from patientderived cells, and most iPSC work thus far has focused on establishing cell lines with fAD-related mutations. Understandably, all iPSC models of fAD have been established from mutations in one of APP, PSEN1, or PSEN2 (Table 1 and Table 3). APP can be proteolytically cleaved in a nonamyloidogenic or amyloidogenic manner; the A fragment is produced through the amyloidogenic pathway, wherein APP is sequentially cleaved by the proteases -secretase and -secretase (Figure 2). fAD-causing APP mutations occur within the A region or are adjacent to A region N-or C-terminus (Hatami et al., 2017;Weggen and Beher, 2012). Overall, mutations can increase the amount of A produced, increase aggregation propensity of A, promote formation of toxic conformations of aggregates, or alter processing to favor generation of A42the major A isoform associated with AD pathogenesis (Ryan and Rossor, 2010;Weggen and Beher, 2012). Pathogenic mutations in PSEN1 and PSEN2 alter the catalytic subunit of -secretase leading to J o u r n a l P r e -p r o o f a similarly increased A42/A40 ratio, but these genetic variants also impair the normal function of -secretase which has been hypothesized as an alternative and much less explored pathway to neurodegeneration and AD (Kelleher and Shen, 2017;Steiner et al., 1999). Additionally, trisomy of chromosome 21the polysomy causing Down syndrome (DS)also leads to the formation of amyloid plaques and NFTs likely due to an increased gene dosage as APP is located on chromosome 21. Overall, the monogenic nature of fAD provides a defined, manipulatable cause for observed pathophysiology, making it an ideal disease to model with patient-derived iPSCs.

[3] 602w Measurements of A levels as well as total and phosphorylated tau in iPSC-derived neurons, and how these levels differ from non-AD control lines, have become standard practice in the characterization of fAD iPSCs. Early models showed that fAD cell lines could recapitulate both increased A levels and tau hyperphosphorylation found in patients with AD. Yagi et al. reported the first generation of an iPSC model of AD using patient-derived cells, and differentiated neurons from PSEN1 and PSEN2 mutant fAD showed increased A secretion and an increased A42/A40 ratio (Yagi et al., 2011). Shortly after, Israel et al. reported the generation of fAD iPSCs from patients with either an APP duplication or sAD, with differentiated neurons from both conditions showing an increase in phosphorylated tau (Israel et al., 2012). While A40 secretion from differentiated neurons was also elevated, no change in A42 or the A42/A40 ratio was reported due to low detectability, highlighting potential differences for certain fAD mutations, sAD, and experimental protocol. Subsequent efforts using fAD iPSC-derived models have generally showed a trend of 1.2 -5 fold increase in the A42/A40 ratio and a 2 -4 fold increase in total and phosphorylated tau (Table 4), enabling future studies to better understand mechanisms by which these increases contribute to AD pathophysiology. Moore et al. used iPSCs to study the A/p-tau relationship in vitro by deriving cortical neurons from patients possessing mutations in PSEN1, APP, or trisomy 21 (Moore et al., 2015). In this work, they showed that AD mutations that increased APP dosage (V717l mutation and APP duplication) were directly linked to an increase in both total tau and phosphorylated tau levels. Furthermore, they showed that -secretase inhibition (GSI), significantly increased total tau, while modulation of -secretase via -secretase modulators (GSM)chemicals that selectively block APP processing activity of -secretasedecreased total tau. Separately, it has also been shown that treatment of fAD neurons with A antibodies decreases total tau (Muratore et al., 2014). Overall, this research implies a link J o u r n a l P r e -p r o o f between tau and A at the level of APP processing, with the former study suggesting the intriguing possibility that altered APP cleavage may impact tau independent of A levels. This link was again studied by Li et al., who investigated the molecular mechanisms of this relationship, finding that DS neurons showed a significant upregulation of the protein p44 (Li et al., 2015a). p44, an isoform of the tumour suppressor protein p53, increases in level with age and has been shown to cause late age-like cognitive impairment and increased tau phosphorylation when overexpressed in mouse models (Pehar et al., 2014). Upregulation of p44 as a result of APP duplication suggests a direct link between the two. Indeed, Li et al reported that the APP intracellular domainan APP cleavage product that is generated from the C-terminus of the protein following proteolytic cleavage by /-secretase and -secretase (Figure 2)regulates translation of p44 by binding to an internal ribosomal entry site of p53 mRNA (Li et al., 2015a). In another study by Hu et al. it was shown that secretomes from iPSC-derived neurons from fAD patients impaired long-term potentiation in rat models (Hu et al., 2018). Interestingly, A mediated the dysfunction caused by PSEN1 and APP mutation secretomes, but extracellular tau mediated this effect in trisomy 21 secretomes. This synaptic dysfunction could be alleviated with antibodies targeting prion proteinan A receptor that has been implicated in causing AD (Hu et al., 2018;Lauren et al., 2009), emphasizing potentially independent pathogenic mechanisms of A and tau, converging on a shared downstream target in human derived tissues.

[4] 154w As with fAD, early studies on iPSC sAD models focused on investigating the Aβ and tau pathology seen in derived neurons. As mentioned, Israel et al. reported the first generation of iPSCs and J o u r n a l P r e -p r o o f neurons from sAD patients (Israel et al., 2012). In their experiments, they found that one of their two sAD lines showed increased Aβ secretion when compared to healthy control samples (Israel et al., 2012). The same line displayed an increase in phosphorylated tau and GSK-3βa kinase which phosphorylates tau. However, the other sAD line did not show any difference in these traits when compared to controls, highlighting the phenotypic variation that can occur between lines from different individuals. Similarly, Kondo et al. reported no change in extracellular Aβ for their two patient-derived sAD neurons, but did find a decrease in intracellular Aβ (Kondo et al., 2013).

[5] 285w Deeper comparison between sAD and fAD Aβ/tau physiology has also been investigated. Ochalek et al. compared AD patient neurons originating from either PSEN1 fAD patients or sAD patients (Ochalek et al., 2017). Analysis of Aβ showed increased Aβ42 and Aβ40 in both fAD and sAD lines compared to controls; however, only fAD lines displayed a significant Aβ42/Aβ40 ratio increase. Generally, a lack of significant differences Aβ42/Aβ40 ratio is a trend for many sAD iPSC models (Table 5 and Table 6). Analysis of APP and APP carboxy-terminal fragment also revealed increased levels for both fAD and sAD lines. Site-specific tau phosphorylation was also compared between the lines, revealing an increased ratio of phosphorylated tau/ total tau at 5 different phosphorylation sites (Ser262, Ser396, Ser202/Thr205, Thr181, and Ser400/Thr403/Ser404); increased GSK-3β activity was also observed. Both sAD and fAD lines also showed a reduced ability to survive oxidative stress and Aβ42 induced cell death, displaying an inability of both sAD and fAD cells to survive when exposes to stressors, which could possibly be explained by upregulation of stress-related genes. Interestingly, while cell lines showed some phenotypic heterogeneity, this difference was not as overt as had been found in cell lines from Israel et al. and Kondo et al., displaying the need for more research on the heterogeneity of sAD iPSC cell lines (Israel et al., 2012;Kondo et al., 2013;Ochalek et al., 2017). It is of importance to note that absence of Aβ and tau pathology may not nullify an sAD model. In particular, Birnbaum et al. showed that sAD neurons from some patients displayed increased reactive oxygen species production, DNA damage, and altered mitochondrial protein expression without displaying significantly altered Aβ or tau pathology (Birnbaum et al., 2018).

[6] 222w Despite advances in iPSC modeling, the extent to which AD patient derived iPSC's and iPSCderived neurons cultured in 2D can recapitulate the disease phenotype is still considered far weaker than animal study and distant from the patient phenotype (Liu et al., 2018). Three dimensional (3D) neural tissue engineering made large advancements in recent years with promise for applications in disease research (de la Vega et al., 2019;Zhuang et al., 2018). 3D neural tissue models refer to neural cell structures that utilize tissue engineering strategies to create a 3D microenvironment. Many different types of neural tissue models exist, including spheroids, cerebral organoids, microfluidics, and hydrogel scaffolds. Compared to standard 2D culture, 3D models are better able to recapitulate the complex microenvironment cells face within their native tissue, allowing cells to form more accurate cell-cell interactions and cell-extracellular matrix (ECM) interactions (de la Vega et al., 2019;Zhuang et al., 2018). Compared to in vivo animal study, 3D models also provide the advantage of being far more cost effective and manipulatable. In regard to AD pathophysiology and drug research, these approaches could lead to a platform for disease study that could supplement or bypass animal research, which would J o u r n a l P r e -p r o o f greatly facilitate drug discovery and improve understanding of disease pathophysiology (Figure 3).

[1] 256w Overall, the advent of iPSCs has been remarkable in allowing researchers access to AD patientderived brain tissue, including neurons, astrocytes, and microglia. More recent methodology has also allowed 3D tissue engineering, with resulting structures that may better recapitulate brain physiology compared to traditional 2D cultures. The ability to differentiate iPSCs to the major cell types in the brain, and ongoing work to create subpopulations of neurons, offer an unprecedented opportunity to study the molecular mechanisms of well known clinical and pathologic findings in AD in human tissue. The promise of this approach is to better model Alzheimer disease pathophysiology, defining new and more relevant targets for therapeutic development, and a platform to test emerging drug candidates. As can be gleaned from this review, the use of iPSCs in AD research is in its infancy, and the next decade will show whether the approach lives up to its potential. Indeed, while promising discoveries have been made, there are many areas in need of further research, includingbut not limited tocontinued modeling of sAD, investigation into AD glial cell modeling, and further development of 3D iPSC models of AD. These areas of research will also offer an important opportunity to look beyond our traditional view of AD pathophysiology, especially to address novel mechanisms upstream and downstream of A and tau, as well as relevant pathogenic events that might happen at the earliest stages of brain development. Investigations into these areas and more is vital to improve our understanding of AD and work towards better treatments for this devastating disease.

[1] 217w Microfluidic-based cell culture systems utilize microchambers and small liquid volumes to recapitulate in vivo processes. These devices can incorporate biocompatible scaffolds to create a 3D microenvironment for seeded cells. The high customizability of microfluidic devices allows for investigation of many different neural processes ranging from neuronal stiffness to biochemical gradients for neuronal differentiation (Osaki et al., 2018). While, to our knowledge, no research has been so far published using patient-derived AD iPSCs in a microfluidic system, these approaches still hold for AD modeling and drug screening. Kunze et al. created a co-pathological neural cell culture where disease state neurons (neurons induced to hyperphosphorylate tau) and healthy cells were cultured in separate compartments but connected through a neurite network (Kunze et al., 2011). This device allows for the investigation of neurodegenerative propagation between diseased and healthy cells. Another study by Choi et al. created a microfluidic system that generated spatial oligomeric gradients for the investigation of the toxicity of Aβ. Neural progenitor cells were seeded into a microfluidic channel and experienced the Aβ gradient, showing neuronal loss and destruction of neurites. It is possible that these, and other microfluidic J o u r n a l P r e -p r o o f devices reviewed elsewhere (Osaki et al., 2018), could be used to model patient-derived AD.

[2] 44w Desirable devices could investigate advanced in vitro drug screening by integrating patientderived neurodegeneration with a blood brain barrier transport model. Eventually, such a device could improve early-stage pruning of candidate drugs, supplementing or even replacing costly in vivo animal models (Oddo et al., 2019).

[3] 231w Hydrogels are 3D networks of hydrophilic, cross-linked polymer chains that absorb water and can provide structural support for seeded cells. Hydrogels are highly variable depending on their composition and can provide a microenvironment that mechanically and biochemically reflect the natural cell ECM (Willerth and Sakiyama-Elbert, 2008). The modifiable properties of hydrogels also allow them to be tuned to specific cellular targets, and their porous nature allows for the transport of nutrients and waste, making them ideal tools for engineering 3D neural tissue (de la Vega et al., 2019). Although not originating from patient-derived cells, Choi et al. reported creation of a 3D system using Matrigel as its hydrogel (Choi et al., 2014). Cells with lentiviral introduced fAD mutations showed a significant increase in Aβ40 and Aβ42. In 3D culture, it was found that Aβ formed many insoluble aggregates of 10 -50μm diameter and that treatment with a GSM decreased the number of these aggregates. Additionally, filamentous phosphorylated tau aggregates could be visualized in the soma and neurites of cells highly expressing the fAD mutations. In addition to Matrigel, 3D AD models have been developed using self-assembling peptides, collagen, and poly(ethylene glycol) (Labour et al., 2016;Papadimitriou et al., 2018;Zhang et al., 2014). Based on our literature search, no research has yet been done on hydrogels to model patient-derived fAD cells and no work at all has been done on modeling sAD explicitly.

[4] 40w Whether this is due to difficulty generating an AD phenotype in 3D hydrogel patient-derived models or caused by a lack of research into the area is unclear, but it is an unmet area that could benefit greatly from further research.

[5] 46w Outside of these challenges, future research could be directed towards the source and characteristics of the disease cells. As mentioned, the majority of AD iPSC research is focused on fAD cell lines (Table 1, 2, 3), and more research should be done using sAD cell lines.

[6] 227w Additionally, iPSCs could be used to look at rare phenotype cases within fAD or sAD. For example, Li et al. describes iPSCs generated from an fAD patient who also presented cerebellar ataxia (Li et al., 2018). Cerebellar ataxia is rare during the course of fAD (Anheim et al., 2007), and although it was not a focus of the study, research on this iPSC line could reveal underlying mechanisms causing cerebellar ataxia. The establishment of iPSC line from rare mutations can allow for further investigation onto why particular mutational profiles cause varying symptoms for the disease. The cell-type origin of AD iPSCs is also an interesting avenue of research. The majority of AD iPSCs research has been conducted using reprogrammed fibroblasts, while a minority of studies have used blood mononuclear cells. Both cell types have advantages: fibroblasts are more heavily researched, while blood mononuclear cells are less invasive to collect. Noteworthy research has also been conducted investigating alternative methods of cell collection. Rose et al. derived iPSCs from autopsy leptomeningeal cells, with the argument being that post-mortem collection allows for a better correlation of iPSC/derived cell biology with neuropathological changes (Rose et al., 2018). Similarly, Sproul et al. reported the generation of iPSCs from non-cryoprotected dura mater that had been stored for up to 11 years, showing the robustness of tissue to be reprogrammed (Sproul et al., 2014b).

[1] 209w The discovery of induced pluripotent stem cells (iPSC) in 2006 by Takahashi and Yamanaka launched a wave of research into the modeling of patient-derived cells (Takahashi et al., 2007;Takahashi and Yamanaka, 2006). Their research found that the introduction of four distinct exogenous transcription factors (Oct4, Sox2, cMyc, and Klf4) could reprogram murine and human fibroblasts into a pluripotent state, which could theoretically be differentiated into any cell type of the body. The discovery of iPSCs made it possible to model in vitro patient cells previously inaccessible in part due to restrictions on stem cells derived from embryos, such as the cells of the central nervous system (CNS). In standard protocols for modeling neurodegenerative disorders using iPSCs, a patient sample is takentypically fibroblasts via a skin biopsy or whole bloodand is reprogrammed using one of a variety of protocols (Malik and Rao, 2013). These cells are then differentiated into a neural fate and then used to investigate cellular pathophysiology or as tools to identify and screen potential therapies. Such models have been created for a number of neuropsychiatric and neurodegenerative disorders, including Parkinson's disease, Huntington's disease, schizophrenia, frontotemporal lobar degeneration, and AD, with the potential for better translation of preclinical findings to the relevant human population (Wu et al., 2019).

[2] 79w In this review, we will cover the use of patient-derived iPSCs for modeling and characterization of AD. We performed an in-depth literature review to identify the bulk of studies where patientderived iPSCs have been used, and discuss the collective experience thus far using these cells to model fAD and sAD. We further highlight the emerging use of 3D culture models for AD, and identify areas for further research and improvement in the growing field of iPSC modeling of AD.

[3] 265w Patient-derived iPSCs serve as a valuable tool for investigating how fAD-associated mutations could play a role in other cellular physiological mechanisms of AD pathogenesis. Moreno et al. investigated mechanisms for type 2 diabetes being associated with dementia (Moreno et al., 2018). They suggest that insulin resistance causes an increased Aβ42/40 ratio, but subsequently found that neurons with the N141I PSEN2 mutation did not show any increased resistance to insulin when compared to controls. However, they did report that fAD cell lines showed improved Ca +2 signalling upon treatment with insulin and suggest that further functional study could be beneficial to understanding the role of insulin in fAD. Another study by Martín-Maestro examined how mitochondrial recycling could be affected by fAD mutations (Martín-Maestro et al., 2017). A246E PSEN1 mutant neurons demonstrated significant mitophagy dysfunction as a result of an impairment in lysosomes ability to degrade ubiquitin-tagged mitochondria, lending an explanation for the significant mitochondrial dysfunction shown in AD. A separate group also found that neurons harbouring APP and PSEN1 mutations had impaired general autophagy and lysosome function which could be rescued upon inhibition of -secretase with a gamma-secretase inhibitor J o u r n a l P r e -p r o o f (GSI), also suggesting a direct causal role for fAD mutations in autophagy (Hung and Livesey, 2018). Interestingly, extracellular vesicles isolated from neurons originating from patients with mutations in PSEN1 had high Aβ42/40 ratios and caused mitochondrial dysfunction of healthy neurons. Furthermore, impairment of autophagy via lysosomal impairment resulted in an elevation of high Aβ42/40 ratio pathogenic extracellular vesicles (Eitan et al., 2016).

[4] 352w Drug discovery is an area of major research focus for iPSC fAD models. GSIs are one of the most common drugs screened during establishment and testing of AD iPSCs, and GSIs are beneficial for elucidating physiological mechanisms of AD as shown by previously mentioned studies (Hung and Livesey, 2018;Moore et al., 2015). GSIs can be used to validate an iPSC line's ability to respond to potential drug treatments due to their mechanism of inhibiting Aβ formation (Mertens et al., 2013;Yagi et al., 2011). Additionally, iPSCs have been utilized to investigate therapeutic potential of GSIs, specifically that of modern, second-generation GSIs (Liu et al., 2014). Gamma secretase modulators (GSMs) have also been screened for therapeutic potential utilizing iPSCs and have yielded promising in vitro results (Liu et al., 2014;Mertens et al., 2013;Yagi et al., 2011), although success has been limited in other studies due to poor drug-like properties (Bursavich et al., 2016). Non -secretase-manipulating drugs have also been screened with relative success utilizing patient-derived iPSCs. Muratore et al. have used A antibodies to successfully reduce tau of early APP mutant neurons (Muratore et al., 2014). N-butylidenephthalide, a small molecule derived from the chloroform extract of Angelica sinensis, decreases total tau and phosphorylated tau levels in DS neurons via activation of the Wnt signalling pathway, although a significant decrease in A42 or the A42/A40 ratio was not reported (Chang et al., 2015). Additionally, Apigenina natural polyphenol found in many plantsdemonstrated neuroprotective qualities against microglial-induced inflammatory stress when administered to both fAD and sAD neurons (Balez et al., 2016). Cholesterol metabolism has also been identified as a possible druggable target in fAD, as cholesterol esterases are increased as a result of APP fAD mutations and have been shown to regulate both A and tau (van der Kant et al., 2019). While patient-derived iPSCs for the most part cannot predict safety and tolerability in humans, they are clearly a useful tool for novel drug discovery and target engagement. As these methods yield novel drugs for AD, it remains to be seen to what extent iPSC-derived CNS cell types afford better translational outcomes than current pre-clinical models.

[5] 308w Advances in genome-editing technology have made it possible to target and manipulate single gene polymorphisms (Soldner and Jaenisch, 2018). The monogenic profile of fAD has led multiple researches to use genome-editing technology to target fAD mutations in patient-derived iPSCs to better the individual impact of each gene on the overall AD pathophysiology. The standard method for studying cellular variation in patient-derived iPSCs caused by fAD-associated mutations requires the generation of isogenic controls. Patient-derived isogenic controls are generated by utilizing genome-editing technology to replace the mutant allele of APP, PSEN1, or PSEN2 with a healthy/wild type allele. These cell lines are considered the gold standard for comparative controls, as any variation seen between the two cell lines will only be caused by the gene in question (Kim et al., 2014). Currently, clustered regularly interspaced short palindromic repeats (CRISPR)/Cas9 is the most common genome-editing technique for fAD iPSCs. Multiple establishment studies have been performed for isogenic iPSC cell lines with the PSEN1 mutation using CRISPR/Cas9, although few have investigated physiological differences upon differentiation into a neuronal fate (Frederiksen et al., 2019a;Oksanen et al., 2017;Pires et al., 2016;Poon et al., 2016). Similar establishment studies have been reported for APP mutant cell lines (Frederiksen et al., 2019b). Oksanen et al. used CRISPR/Cas9 to generate isogenic controls from two PSEN1 mutant fAD patients and found significant disease pathology when iPSCs were differentiated into a glial-fate (Oksanen et al., 2017). fAD PSEN1 astrocytes showed a significantly higher Aβ42 amount and Aβ42/Aβ40 ratio when compared to isogenic controls, and that isogenic controls were similar in these levels to non-mutant controls, implying the AD disease phenotype could be rescued upon correction of only the PSEN1 gene. PSEN1 mutant cells also showed impaired Ca 2+ signalling and metabolism when compared to isogenic controls, although comparison to non-mutant controls was not investigated. Two isogenic control lines for PSEN2

[6] 243w were generated in a study by Ortiz-Virumbrales et al. where the diseased cell line derived neurons showed an increased Aβ42/Aβ40 ratio (Ortiz-Virumbrales et al., 2017). Interestingly, only one of the diseased cell lines showed sensitivity to Aβ42 when compared to the isogenic controls (Ortiz-Virumbrales et al., 2017), despite previous findings suggesting that fAD mutations display increased susceptibility to Aβ42 toxicity (Armijo et al., 2017). This result suggests that non-fAD implicated genes may play a role in fAD cellular pathophysiology and shows the value isogenic control lines possess in elucidating this role. In addition to patient-derived gene edited iPSCs, research has also been performed on healthy iPSCs edited to contain fAD mutations. The Goldstein group has published multiple studies using E9 cellshealthy iPSCs that have been genetically altered with TALENs to possess the PSEN1 mutation E9 (Woodruff et al., 2016;Woodruff et al., 2013). Similar to other isogenic lines, they found a relative increase in Aβ42/Aβ40 J o u r n a l P r e -p r o o f ratio in both E9 neural progenitor cells and neurons in their initial study (Woodruff et al., 2013). This increased in a gene dose-dependent manner, with E9/E9 genotype yielding a higher ratio than wildtype/E9 and the E9/null genotype yielding similar results to E9/E9. Overall, the generation of isogenic gene controls allows for elimination of axillary polymorphic gene influences on AD pathology, giving an opportunity for better understanding of how the fAD mutations lead to AD.

[7] 201w The polygenic nature of sAD makes modeling of this disease variant far more challenging when compared to fAD, and thus less research has been conducted on sAD iPSCs despite it comprising the vast majority of AD cases (Table 2). The 4 variant of the APOE gene is the major genetic risk factor for sAD (Karch and Goate, 2015). Humans have 3 polymorphic alleles for APOE: 2, 3, and 4. APOE3 has the highest frequency, being found in 77.9% of the population, APOE4 and APOE2 follow with 13.7% and 8.4%, respectively (Farrer et al., 1997). However, APOE4 is significantly enriched in AD patients, showing a frequency of 36.7% with APOE3 and APOE2 following with 59.4% and 3.9%, respectively. ApoE, a fat-binding protein, mediates lipid transport between cell and various tissues. In the CNS, astrocytes and microglia produce ApoE, and its function is to transport cholesterol to neurons which express ApoE receptors on their surface (Liu et al., 2013). The exact mechanism by which APOE polymorphisms contribute to AD pathogenesis remains unclear, but it is thought that ApoE polymorphisms can influence Aβ aggregation and clearance, lipid metabolism, synaptic plasticity, neuroinflammation, neurogenesis, and other gain-of-toxic function or loss-of-normal function factors (Liu et al., 2013).

[8] 137w APOE4 has been found to be associated with increased amyloid plaques and aggregated Aβ in both healthy and AD individuals (Barthel et al., 2011;Kok et al., 2009;Polvikoski et al., 1995;Reiman et al., 2009;Schmechel et al., 1993), and has been implicated in multiple other potential mechanisms related to gain-of-function toxicity and loss-of-function neuroprotection (Liu et al., 2013). Since the discovery of APOE as a risk factor for sAD, multiple lesser genetic risk factors have been identified. Among these, CLU, encoding the apolipoprotein clusterin (CLU), has also been implicated in sAD pathogenesis. SNPs for CLU have been associated with both protection from and association with sAD, although there is currently no functional explanation for these findings (Karch and Goate, 2015). TREM2 encodes a microglial receptor that has the dual function of stimulating phagocytosis and supressing cytokine production/inflammation (Rohn, 2013).

[9] 453w Mutations in TREM2 have been associated with AD, frontotemporal dementia, Parkinson's J o u r n a l P r e -p r o o f disease, and amyotrophic lateral sclerosis (Karch and Goate, 2015). Specifically, the R47H missense variant has been shown to have a 1.7-3.4-fold increase in AD risk (Guerreiro and Hardy, 2013;Pottier et al., 2013). SORL1 encodes the endocytosis protein sortilin-related receptor L (SORL1), and assists in vesicular trafficking from the cell surface to the Golgi and endoplasmic reticulum. Genome-wide association studies (GWAS) and meta-analyses have confirmed a link between SORL1 variants and sAD (Lambert et al., 2013;Reitz et al., 2011). SORL1 is a receptor for ApoE and this interaction is thought to be the root of the SORL1-sAD link (Yin et al., 2015). In addition to these genes, GWAS have identified associated risk between sAD and the immune genes CR1, CD33, MS4A; the endocytosis genes BIN1, PICALM, CD2AP, and EPHA1; the lipid metabolism gene ABCA7; and other genes that have been reviewed previously (Harold et al., 2009;Karch and Goate, 2015;Lambert et al., 2009;Lambert et al., 2013). More recently, lumping numerous risk alleles each conferring a small risk of AD into a composite score has gained traction for predicting AD. Recent work integrated single-nucleotide polymorphism (SNP) data at 31 risk genes from sAD and control patients to generate a PHS, which was subsequently validated to predict the age-specific risk of developing sAD (Desikan et al., 2017). The composite genetic risk profile varied age of onset by over 10 years, while follow up work showed that PHS varies Aβ and total tau levels and that high PHS individuals showed the fastest rates of cognitive and clinical decline (Tan et al., 2019). The PHS displays the significant influence risk genes have on sAD progression, revealing a need for further understanding on how AD-associated genes contribute to sAD progression, both individually or the cumulative effect of a polygenic profile. There remains no true animal model of the disease because of sAD's unclear genetic etiology (Drummond and Wisniewski, 2017). In addition, genetic alteration of healthy iPSCs is mostly limited to investigating individual or a small number of risk alleles for sAD. While these individual risk alleles are of the highest relevance to current AD research as their pathogenic mechanisms remain elusive, such modified iPSCs cannot yet be used to model the full genetic complexity of sAD. Patient-derived sAD iPSCs would fill this gap, and therefore represent the most accurate method of recapitulating the disease in vitro, with the important caveats that given the high genetic variability among individuals, shared pathophysiologic downstream pathways are favored over the specific contributions of individual genetic alleles, and that many important environmental contributions to sAD cannot be accounted for.

[10] 472w iPSC models can also elucidate how risk genes contribute to the etiology of sAD. Recently, multiple groups have used genomic editing to alter the APOE gene in iPSCs. Wadhwani et al. utilized CRISPR to replace APOE4 with homozygous APOE3 in two iPSC cell lines (Wadhwani et al., 2019). It was shown that edited APOE3 cells did not display significantly altered Aβ levels, but did display reduced tau phosphorylation and reduced phosphoactivation of the ERK1/2 J o u r n a l P r e -p r o o f enzymethe increase of both of which has been implicated in AD progression (Wadhwani et al., 2019). Previous studies have also investigated CRISPR editing of APOE in healthy iPSCs (Lin et al., 2018;Wang et al., 2018). These studies have reported an increase in phosphorylated tau for APOE4, but also an upregulation of Aβ in iPSC-derived neurons, creating some discrepancies between these findings and those of Wadhwani et al. It is possible that this is due to heterogeneity seen between small sample sizes, but more work should be done to investigate the effect of APOE in patient-derived sAD cells. Outside of APOE, Young et al. investigated how SORL1 contributes to the AD phenotype. Reduced or loss of SORL1 expression has been found in sAD patients (Dodson et al., 2006;Scherzer et al., 2004), but Young et al. did not find a difference in SORL1 expression between sAD iPSCs and healthy controls (Young et al., 2015). It was noted that cells homozygous for the so called "risk" variations (R/R) in SORL1 could not be induced to increase SORL1 expression, whereas cells homo-or heterozygous for the "protective" variations (P/P and R/P) could be induced to increase expression. Furthermore, they found that increased SORL1 expression significantly reduced levels of Aβ40, revealing a possible mechanism for the link between SORL1 and sAD (Young et al., 2015). TREM2 and CLU have also been investigated in iPSC models, although these were not entirely patient-derived. Xiang et al. utilized pooled healthy/mild cognitive impairment/AD iPSCs heterozygous for the TREM2 R47H risk variant to compare to R47H knock-in mice (Xiang et al., 2018). While knock-in mice showed reduced TREM2 mRNA caused by atypical splicing, iPSC microglia-like cells and patient brains showed normal mRNA levels and splicing, displaying the limitations of mouse models in recapitulating human sAD phenotypes. In a study on iPSCs derived from a healthy individual, CLU was investigated for its role in mediating Aβ toxicity (Robbins et al., 2018). It was found that CLU knockout neurons were resistant to Aβ-induced neuronal degeneration while wildtype cells were susceptible, and that intracellular levels of CLU are increased following treatment. Additionally, genes downstream of APP were found to be dysregulated in CLU knockouts. Previous studies have found that deletion of CLU in mouse models significantly reduces Aβ deposits, supporting these iPSC results (DeMattos et al., 2002).

[11] 296w Given that sAD accounts for over 99% of all AD cases, drug testing is an important area of research for patient-derived sAD iSPC models (Joshi et al., 2012;Piaceri et al., 2013;Ryan et al., 2016). Few studies have been done on drug identification for sAD. Like with fAD, GSIs have been tested as validators of drug screening in sAD neurons. Hossini et al. used a GSI on two sAD cell lines, finding a reduction in phosphorylated tau in only one of the lines, again displaying the J o u r n a l P r e -p r o o f heterogeneity in pathophysiology seen in sAD cell lines (Hossini et al., 2015). Contrary to this, Israel et al. reported in their foundational study that β-secretase inhibitors but not GSIs reduced phosphorylated tauand additionally reduced GSK-3β activity, a trait shared with fAD cell lines (Israel et al., 2012). Other processes have also been investigated for drug discovery. van der Kant et al. reported that cholesterol metabolism inhibitors were found to reduce phosphorylated tau in sAD neurons, although this was similar to the decrease found in healthy iPSC cell lines (van der Kant et al., 2019). Young et al. investigated stabilization of the retromer complex as a possible therapeutic target (Young et al., 2018). The retromer complex is a highly conserved assembly comprised of multiple proteins that has an important role in the sorting and trafficking of the endosomal network and previous mouse model studies have found that paralogical chaperones can be used to enhance function and reduce pathogenic Aβ processing (Hu et al., 2015b;Mecozzi et al., 2014). Young et al. found that treatment with the chaperone R33 significantly reduced both Aβ40 and Aβ42 levels in sAD lines, and phosphorylated tau ratio in both sAD and fAD lines.

[12] 41w Overall, iPSC modeling provides one of the only methods for drug discovery specifically in sAD; however, while these results are promising, the scarcity of studies focusing on drug screening using sAD iPSCs reveals a need for further research on the subject.

[13] 25w Neural spheroids are self-or forced-assembled aggregates of neural cells that generally lack exogenous scaffolds and are cultured attached to treated surfaces or suspended in media.

[14] 111w Spheroids have the advantage of requiring fewer resources and less expertise when compared to other methods of engineering 3D neural tissue, while still providing complex models for neural networks and neuropathy (de la Vega et al., 2019). Spheroids have had success modeling a number of neurological disorders, including brain cancer, neurodevelopmental disorders, and neurodegenerative disorders (Birey et al., 2017;Daniele et al., 2018;Lee et al., 2018a;Lee et al., 2016) Lee et al. developed a 3D spheroid model for AD utilizing iPSCs taken from sAD patients (Lee et al., 2016). In this protocol, ultra-low adherence culture plates were used to generate selfassembled aggregates of iPSCs, spheroids were then differentiated into a neuronal fate.

[15] 13w Spheroids produced both Aβ40 and Aβ42, although this was variable between cell lines.

[16] 89w Treatment of the 3D spheroids and 2D culture with GSIs and β-secretase inhibitors led to a reduction in both Aβ40 and Aβ42; however, reduction in the 3D spheroids was relatively lower than what was seen in the 2D culture. This could be attributed to the discrepancy in bioavailability of the drugs due to the spheroid structure, with cells in the outer layers receiving a higher dose than the internal cells. This presents an interesting, perhaps more physiologically relevant, circumstance that may elevate the value of spheroids for drug discovery.

[17] 398w Cerebral organoids are highly complex, in vitro, self-organizing, miniature organs that resemble the brain. Currently, multiple different protocols exist for the generation of cerebral organoids. The first reported protocol involved generating an embryoid bodya 3D aggregate of iPSCsand then growing it within an ECM gel such as Matrigel (Lancaster et al., 2013). During the development process, large buds of neuroepithelium form, which then mature into discrete brain regions (Lancaster and Knoblich, 2014). Compared to spheroids, cerebral organoids are far more complicated in structure and more closely resemble brain tissue. Recent studies have used cerebral organoids to model AD with success. Raja et al. developed an fAD cerebral organoid model by culturing EBs long-term with a variety of fate determining growth factors (Raja et al., 2016). The organoids showed hallmarks of AD pathogenesis, including increased Aβ40, Aβ42, and phosphorylated tau. Additionally, it was found that Aβ levels were age dependent, with J o u r n a l P r e -p r o o f organoids showing a progressive increase in Aβ throughout maturation. Organoids also displayed abnormal endosomal morphology and trafficking, further solidifying accuracy of the disease pathophysiology. Raja et al. also noted that recapitulation of the disease phenotype could be achieved regardless of the fAD mutation, as they found consistent results with two APP duplication lines and two PSEN1 mutant lines. Interestingly, it was also found that a compound treatment of GSI and β-secretase inhibitor administered during development did not impair neural maturation but did decrease Aβ and phosphorylated tau levels. More recently, Meyer et al. utilized cerebral organoids to investigate whether APOE4 affects epigenetic regulation at mature neural stages (Meyer et al., 2019). Organoids were generated using the Lancaster method and originated from cells isogenic for APOE3 or APOE4. Transcriptional analysis revealed that APOE4 organoids displayed an upregulation of genes associated with nervous system development, synaptic transmission, and neurogenesis; furthermore, APOE3 organoids were composed of relatively more neural progenitor cells. Overall, this suggests that early differentiation caused by APOE4 may impact the AD phenotype. While cerebral organoids are a promising model, drawbacks do exist; mainly, the self-forming nature of organoids makes reproducibility a major issue, as it is essentially impossible to generate the exact same structure multiple times (Velasco et al., 2019). While this is an issue, the inherent complexity of cerebral organoids makes them a unique and useful method of investigating AD in vitro.

[18] 289w The vast majority of patient-derived iPSC modeling and general iPSC modeling has focused on differentiating iPSCs into a cortical neuron fate (Table 1, 2, 3). While neurodegeneration is the vital step in AD progression and it is understandable that most work up until this point has focused on modeling that, it is important to understand how glial and epithelial cells contribute to the J o u r n a l P r e -p r o o f progression or slowing of AD. Astrocytes constitute the majority of the cells in the brain and have vital functions in homeostasis and synaptic signalling, making them integral in the progression of AD (González-Reyes et al., 2017). Importantly, ApoE is produced mainly by astrocytes in the CNS, meaning any iPSC model of sAD that investigates APOE4s involvement in disease progression should incorporate differentiated astrocytes in addition to neurons. Few studies have investigated patient-derived astrocytes, but work that has been done has reported astrocytes showing increased Aβ production, altered cytokine profiles, dysregulated Ca 2+ signalling, immature morphologies and abnormal localization of astroglial markers (Jones et al., 2017;Liao et al., 2016;Oksanen et al., 2017). Such models could benefit greatly from co-cultures to better understand astrocytes role in AD pathogenesis. In addition to astrocytes, microglia also play a role in AD progression. Microglia are the immune cells of the CNS and are therefore thought to have a critical and complex role in the development of AD. Evidence exists for microglia being both neuroprotective and detrimental in AD progression (Hemonnot et al., 2019). Specifically, microglial TREM2 is involved in the phagocytosis of Aβ, and it is thought that other SNPs implicated in sAD contribute to an inhibited clearance of Aβ by microglia (Hansen et al., 2018).

[19] 138w Unlike neurons and astrocytes, microglia are difficult to differentiate from iPSCs, making in vitro modeling a challenge; currently, there is no established protocol for the derivation of a pure population of confirmed microglia cells, and microglia-like cells are the closest models that can be created (Abud et al., 2017). As mentioned previously, microglia-like cells have been modelled with some success in iPSCs to investigate mutations in TREM2 (Xiang et al., 2018). A recent study by Xu et al. generated microglia-like cells from sAD patients and found increased phagocytosis and increased cytokine secretion in AD microglia-like cells compared to controls, implying significant inflammatory characteristics (Xu et al., 2019). While this is an interesting finding, further work should be done to confirm that this is a ubiquitous response and how AD microglia cells react in the presence of AD neurons.

[20] 210w Patient-derived iPSCs themselves pose numerous challenges in regard to general and ADspecific disease modeling. Variability is a large issue when developing models for disease. Where animals can be bred with identical genetic backgrounds, human cell lines from different individuals provide no such luxury, and genetic variability across the population can lead to large differences in phenotypes. Three recent large-scale genetic analysis of iPSCs originating from hundreds of individuals revealed that genetic influence is the largest cause of observed heterogeneity (Carcamo-Orive et al., 2017;DeBoever et al., 2017;Kilpinen et al., 2017;Yamasaki et al., 2017), J o u r n a l P r e -p r o o f and such variability could explain the finding that pathophysiology of AD patient-derived neurons varies from line to line (Israel et al., 2012;Kondo et al., 2013;Ochalek et al., 2017). For fAD research, the generation of gene-corrected isogenic controls in addition to healthy controls should be considered standard practice in order to minimize heterogeneity. sAD iPSCs pose a far greater challenge, as healthy isogenic controls cannot be produced from them. Ideally, protocols for the generation of iPSCs would have stringent genetic control to minimize variation, and control/AD cell lines would be genetically matched in order to mitigate heterogeneity; however, this is technically and temporally challenging.

[21] 25w In addition to genotype heterogeneity, epigenetics and neuronal maturation has also been highlighted as a challenge in iPSC neurodegenerative disease modeling (Berry et al., 2018).

[22] 271w Reprogramming patient cells through the iPSC stage wipes out age-related epigenetic markers that may impact disease phenotype. It has been suggested that removal of such epigenetic alternations limits the ability of cellular models to truly recapture disease processes in age-specific onset neurodegenerative disorders, including Huntington disease, Parkinson's disease, and AD (Mertens et al., 2018). Multiple approaches have been suggested to restore age-related alterations or avoid their loss. Specifically, induction of cellular stress in iPSC-derived cells has been used to induce disease phenotype in Huntington-and Parkinson's disease, while expression of a specific LMNA variantthe gene variant causing Hutchinson-Gilford progeria syndromeand telomerase inhibition have been investigated as methods to induce age-related markers (Mertens et al., 2018;Miller et al., 2013;Vera et al., 2016). Direct reprogramming of patientderived cells to a neural fate is an option for avoiding the loss of some age-related alterations altogether. Vierbuchen et al. showed the expression of Ascl1, Brn2, and Myt1l is able to drive patient fibroblasts to an induced neuronal lineage (iN) (Vierbuchen et al., 2010). Since this finding, non-transcription factors approaches have been developed such as the use of discrete cocktails of small molecules, micro RNA, or short hairpin RNA which have been used to produce iNs of varying lineages (Hu et al., 2015a;Li et al., 2015b;Tian et al., 2016;Xue et al., 2013;Yoo et al., 2011). Compared to iPSCs, iNs maintain multiple important age-related marks, making this method useful for investigating neurodegenerative disorders such as AD (Hu et al., 2015a). While direct reprogramming is an attractive area of investigation, current iN protocols do not provide any intermediate expansion cells stage, limiting the number of experientable cells and reproducibility.

[23] 37w Overall, accounting for age-related epigenetics is a major hurdle for modeling AD, and while methods addressing this have been investigated, more needs to be done to create neural models that truly recapitulate the age-related complexities of AD.

[24] 270w Recent advances in tissue engineering technologies have opened the door for 3D bioprinting as a method of engineering 3D neural tissue models, making them an attractive method for generating AD models (de la Vega et al., 2019). 3D bioprinting refers to the practice of combining 3D printing principles with cells and biomaterials to produce tissue-like structures. There are multiple methods of 3D bioprinting, but generally they encapsulate cells within bioinksbiologically relevant, printable materialsand printing them in a layer-by-layer method. These bioinks are often hydrogel-based to give mechanical and biochemical support (de la Vega et al., 2019). In regard to neurological disorders, bioprinting can create highly complex and customizable structures to recapitulate disease phenotype. Previous studies have found success bioprinting neural tissue. Gu et al. bioprinting neural mini tissue constructs using human neural stem cells (Gu et al., 2016). These constructs showed the ability to differentiate into functionally mature J o u r n a l P r e -p r o o f neurons and establish networks. Other groups have printed iPSC-derived neural progenitor cells (Fantini et al., 2019;Salaris et al., 2019), with one study showing high viability and ability to differentiate into mature neurons (de la Vega et al., 2018). 3D printing has also been utilized to generate models of brain cancer, showing its value in recreating disease phenotypes (Dai et al., 2016;Lee et al., 2019). So far, no research has been done on bioprinting models of AD, but the success shown printing healthy and disease cells as well as success shown when using hydrogels as detailed in Section 4.4 means this is an exciting avenue for further research.

[25] 81w Table 1: Studies that have utilized patient-derived induced pluripotent stem cell models of familial Alzheimer's disease. Reported quantitative results for AB 1-42/1-40 ratio, total tau levels, phosphorylated tau levels, and days in vitro when these were recorded. Result refers to comparison with healthy controls or isogenic lines (↑ = <0.05, ↑↑ = <0.01, ↑↑↑ = <0.001, n.s. = not significant). If multiple results were given for the same mutational category (APP dup , PSEN1, etc.) the most significant result is reported.

[26] 964w Title Reference fAD/DS (Mutation [fAD only]) 3D Origin Cell Type Differentiated Cell Type AB42/40 tTau pTau Days cultured Familial Alzheimer's disease patientderived neurons reveal distinct mutation-specific effects on amyloid beta. (Arber et al., 2019) fAD (APP and PSEN1) Yes Fibroblasts Cortical Neurons and cerebral organoids ↑↑↑ (approx. 2x) --100 -104 Altered γ-Secretase Processing of APP Disrupts Lysosome and Autophagosome Function in Monogenic Alzheimer's Disease (Hung and Livesey, 2018) fAD (APP and PSEN1) and DS No Fibroblasts Cortical Neurons ↑↑ (approx. 2x) --40 -90 PPARβ/δ-agonist GW0742 ameliorates dysfunction in fatty acid oxidation in PSEN1ΔE9 astrocytes. (Konttinen et al., 2019) fAD (PSEN1) No Fibroblasts and blood mononuclear cells Astrocytes ----Induction of Amyloid-β42 Production by Fipronil and Other Pyrazole Insecticides. (Cam et al., 2018) fAD (APP) No Fibroblasts Cortical Neurons ----Extracellular Forms of Aβ and Tau from iPSC Models of Alzheimer's Disease Disrupt Synaptic Plasticity (Hu et al., 2018) fAD (APP and PSEN1) and DS No Not stated Cortical Neurons ↑ (PSEN1 [approx.. 2x]); -(DS and APP) --70 -80 J o u r n a l P r e -p r o o f Human fibroblast and stem cell resource from the Dominantly Inherited Alzheimer Network (Karch et al., 2018) fAD (APP, PSEN1, PSEN2) No Fibroblasts iPSCs ----iPSC Modeling of Presenilin1 Mutation in Alzheimer's Disease with Cerebellar Ataxia (Li et al., 2018) fAD (PSEN1) No Blood mononuclear Cells Cortical Neurons ↑↑↑ (approx. 2.5x [extracellular]) ; N.D. Intracellular ↑↑ (approx. 1.5x) ↑↑↑ (approx. 4x) 42 -70 iPSC-derived familial Alzheimer's PSEN2N141I cholinergic neurons exhibit mutation-dependent molecular pathology corrected by insulin signaling (Moreno et al., 2018) fAD (PSEN2) No Fibroblasts Basal Forebrain Cholinergic Neurons Measured but not compared between lines --34 The Impact of APP on Alzheimer-like Pathogenesis and Gene Expression in Down Syndrome iPSC-Derived Neurons (Ovchinnikov et al., 2018) DS No Fibroblasts Cortical Neurons ----Overactive BRCA1 Affects Presenilin 1 in Induced Pluripotent Stem Cell-Derived Neurons in Alzheimer's Disease. (Wezyk et al., 2018) fAD (PSEN1) No Fibroblasts Cortical Neurons ↑ (approx. 2x -5x) --Not specified Phenotypic Screening Identifies Modulators of Amyloid Precursor Protein Processing in Human Stem Cell Models of Alzheimer's Disease (Brownjohn et al., 2017) fAD (APP and PSEN1) and DS No Fibroblasts Cortical Neurons ----J o u r n a l P r e -p r o o f 40 Neurons Derived from Induced Pluripotent Stem Cells of Patients with Down Syndrome Reproduce Early Stages of Alzheimer's Disease Type Pathology in vitro. (Dashinimae v et al., 2017) DS No Amniotic fluid Cortical Neurons AB ratio not measured. ---REST suppression mediates neural conversion of adult human fibroblasts via microRNA-dependent andindependent pathways (Drouin-Ouellet et al., 2017) fAD (APP) No Fibroblasts Various ---80-94 Mitophagy Failure in Fibroblasts and iPSC-Derived Neurons of Alzheimer's Disease-Associated Presenilin 1 Mutation (Martín-Maestro et al., 2017) fAD (PSEN1) No Fibroblasts Cortical Neurons -n.s. -40 Cell-type Dependent Alzheimer's Disease Phenotypes: Probing the Biology of Selective Neuronal Vulnerability (Muratore et al., 2017) fAD (APP) No Fibroblasts Specialized Neurons ↑↑↑ (approx. 3x) ↑ (approx. 1.5x) ↑ (approx. 1.5x) 40 (AB); 100 (Tau) PSEN1 Mutant iPSC-Derived Model Reveals Severe Astrocyte Pathology in Alzheimer's Disease (Oksanen et al., 2017) fAD (PSEN1) No Fibroblasts Astrocytes ↑↑↑ (approx. 3.5x) --7 (following plating of pure cultures) CRISPR/Cas9-Correctable mutationrelated molecular and physiological phenotypes in iPSC-derived Alzheimer's PSEN2N141I neurons. (Ortiz-Virumbrales et al., 2017) fAD (PSEN2) No Fibroblasts Cortical Neurons ↑↑↑ (approx. 2x [healthy control at day 11]); ↑↑ (approx. 2x [isogenic + healthy day 34]) --11, 34 J o u r n a l P r e -p r o o f Early pathogenic event of Alzheimer's disease documented in iPSCs from patients with PSEN1 mutations (Yang et al., 2017) fAD (PSEN1) No Fibroblasts NPCs and Cortical Neurons n.s. (approx. 1.5x [D21]); ↑ (approx. 3x [D25]); ↑↑ (approx. 3.5x [D28]) -↑ (approx. 2 -4x) 21 -28 Extracellular vesicle-associated Aβ mediates trans-neuronal bioenergetic and Ca2+-handling deficits in Alzheimer's disease models (Eitan et al., 2016) fAD (PSEN1) No Fibroblasts Cortical Neurons ↑↑↑ (approx. 3x) --Not specified Single-Cell Detection of Secreted Aβ and sAPPα from Human IPSC-Derived Neurons and Astrocytes (Liao et al., 2016) fAD (APP) No Fibroblasts Multiple (Forebrain Neurons, astrocytes, GABAergic interneurons) ----Self-Organizing 3D Human Neural Tissue Derived from Induced Pluripotent Stem Cells Recapitulate Alzheimer's Disease Phenotypes (Raja et al., 2016) fAD (APP and PSEN1) Yes Fibroblasts Cerebral Organoids n.s. -n.s. (Day 60); ↑↑ (approx. 1.5x [day 90]) 60, 90 N-butylidenephthalide Attenuates Alzheimer's Disease-Like Cytopathy in Down Syndrome Induced Pluripotent Stem Cell-Derived Neurons (Chang et al., 2015) DS No Amniotic fluid Cortical Neurons Ratio not reported ↑↑ (approx. 2x) ↑↑ (Negligible in Control) 43 -45 Direct Conversion of Normal and Alzheimer's Disease Human Fibroblasts into Neuronal Cells by Small Molecules. (Hu et al., 2015a) fAD (APP and PSEN1) No Fibroblasts Cortical Neurons ↑↑ (approx. 2x); n.s. for some lines Increased (via western blot) Increased (via western blot) 14 (ELISA); W.B. unclear Controlling the Regional Identity of hPSC-Derived Neurons to Uncover Neuronal Subtype Specificity of Neurological Disease Phenotypes (Imaizumi et al., 2015) fAD (PSEN1 and PSEN2) No Fibroblasts Cortical Neurons and Neurospheres --Increased (via fluorescent imaging) Unspecifie d J o u r n a l P r e -p r o o f Modeling neurogenesis impairment in Down syndrome with induced pluripotent stem cells from Trisomy 21 amniotic fluid cells. (Lu et al., 2013) DS No Amniotic fluid NPCs and cortical neurons ----APP Processing in Human Pluripotent Stem Cell-Derived Neurons Is Resistant to NSAID-Based γ-Secretase Modulation (Mertens et al., 2013) fAD (PSEN1) No Fibroblasts Cortical Neurons ↑ (approx. 1.2-2x) --Not specified A human stem cell model of early Alzheimer's disease pathology in Down syndrome (Shi et al., 2012) DS No Fibroblasts Cortical Neurons ↑↑ (approx. 1.5x) ↑↑ (approx. 4x) ↑↑ (Negligible in Control) 55 Modeling familial Alzheimer's disease with induced pluripotent stem cells. (Yagi et al., 2011) fAD (PSEN1 and PSEN2) No Fibroblasts Cortical Neurons ↑ (approx. 2x) --14

[27] 53w J o u r n a l P r e -p r o o f 44 Table 2: Studies that have utilized patient-derived induced pluripotent stem cell models of sporadic Alzheimer's disease. Reported quantitative results for AB 1-42/1-40 ratio, total tau levels, phosphorylated tau levels, and days in vitro when these were recorded.

[28] 44w Result refers to comparison with healthy controls or isogenic lines (↑ = <0.05, ↑↑ = <0.01, ↑↑↑ = <0.001, n.s. = not significant). If multiple results were given for the same mutational category (APP dup , PSEN1, etc.) the most significant result is reported.

[29] 165w Title Reference 3D Origin Cell Type Differentiated Cell Type AB42/40 tTau pTau Days cultured REST and Neural Gene Network Dysregulation in iPSC Models of Alzheimer's Disease (Meyer et al., 2019) Yes Fibroblasts Multiple (NPs, Neurons, cerebral organoids) n.s. -↑ (approx. 1.5x) 42 Neuronal apolipoprotein E4 increases cell death and phosphorylated tau release in alzheimer disease. (Wadhwani et al., 2019) No Fibroblasts Cortical Neurons n.s. n.s. ↑↑ (approx. 1.2 -1.5 [E4/3 compared to isogenic E3/3]) 38 Oxidative stress and altered mitochondrial protein expression in the absence of amyloid-β and tau pathology in iPSC-derived neurons from sporadic Alzheimer's disease patients. (Birnbaum et al., 2018) No Fibroblasts Cortical Neurons ↓ (approx. 1.2 [1 of 5]) n.s. n.s. 21 -23 Modeling Late-Onset Sporadic Alzheimer's Disease through BMI1 Deficiency. (Flamier et al., 2018) Yes Fibroblasts Cortical Neurons ----Leptomeninges-Derived Induced Pluripotent Stem Cells and Directly Converted Neurons From Autopsy Cases With Varying Neuropathologic Backgrounds (Rose et al., 2018) No Leptomeningeal cells iPSC focus and cortical neurons n.s. -n.s. 7 (re-plated neurons)

[30] 70w J o u r n a l P r e -p r o o f Table 3: Studies that have utilized patient-derived induced pluripotent stem cell models of both sporadic and familial Alzheimer's disease, or studies have not specified whether the originating Alzheimer's disease category was familial or sporadic. Reported quantitative results for AB 1-42/1-40 ratio, total tau levels, phosphorylated tau levels, and days in vitro when these were recorded.

[31] 44w Result refers to comparison with healthy controls or isogenic lines (↑ = <0.05, ↑↑ = <0.01, ↑↑↑ = <0.001, n.s. = not significant). If multiple results were given for the same mutational category (APP dup , PSEN1, etc.) the most significant result is reported.