Induced pluripotent stem cell models as a tool to investigate and test fluid biomarkers in Alzheimer's disease and frontotemporal dementia

Abstract

Neurodegenerative diseases are increasing in prevalence and comprise a large socioeconomic burden on patients and their caretakers. The need for effective therapies and avenues for disease prevention and monitoring is of paramount importance. Fluid biomarkers for neurodegenerative diseases have gained a variety of uses, including informing participant selection for clinical trials, lending confidence to clinical diagnosis and disease staging, determining prognosis, and monitoring therapeutic response. Their role is expected to grow as disease‐modifying therapies start to be available to a broader range of patients and as prevention strategies become established. Many of the underlying molecular mechanisms of currently used biomarkers are incompletely understood. Animal models and in vitro systems using cell lines have been extensively employed but face important translatability limitations. Induced pluripotent stem cell (iPSC) technology, where a theoretically unlimited range of cell types can be reprogrammed from peripheral cells sampled from patients or healthy individuals, has gained prominence over the last decade. It is a promising avenue to study physiological and pathological biomarker function and response to experimental therapeutics. Such systems are amenable to high‐throughput drug screening or multiomics readouts such as transcriptomics, lipidomics, and proteomics for biomarker discovery, investigation, and validation. The present review describes the current state of biomarkers in the clinical context of neurodegenerative diseases, with a focus on Alzheimer's disease and frontotemporal dementia. We include a discussion of how iPSC models have been used to investigate and test biomarkers such as amyloid‐β, phosphorylated tau, neurofilament light chain or complement proteins, and even nominate novel biomarkers. We discuss the limitations of current iPSC methods, mentioning alternatives such as coculture systems and three‐dimensional organoids which address some of these concerns. Finally, we propose exciting prospects for stem cell transplantation paradigms using animal models as a preclinical tool to study biomarkers in the in vivo context.

In this review article, McInvale et al. describe the current state of biomarkers in the clinical context of Alzheimer's disease and Frontotemporal dementia and they discuss how induced pluripotent stem cell (iPSC) models of these diseases have been used to investigate and test biomarkers in differentiated neurons and glial cells.

1. INTRODUCTION

Fluid biomarkers can be defined as molecules, traditionally proteins or peptides, which are closely related to disease processes [1]. They may be involved in either pathogenesis of disease or the physiological response. Fluid biomarkers have potential uses ranging from diagnostic to prognostic to therapeutic response monitoring [2]. Importantly, biomarkers should have adequate diagnostic sensitivity and specificity such that they can be used to distinguish among clinically overlapping diseases [2]. Biomarkers can be detectable in the cerebrospinal fluid (CSF), and in peripheral fluids including blood, saliva, and urine [2]. Biomarkers may also be used to monitor disease progression and response to disease‐modifying treatments once they become available. While proteins have been canonical biomarkers, there is recent interest in exploring the potential of other biomolecules such as lipids, microRNA, epigenetically modified genomic DNA [3, 4], and cell‐free DNA [5] as biomarkers [1, 6, 7, 8, 9].

Neurodegenerative diseases are a clinically and pathologically diverse group of diseases, which are highly prevalent and carry great social and economic burden [10, 11]. Clinical manifestations vary across diseases but can include dementia, motor dysfunction, and behavioral and emotional dysregulation [12]. The incentive for discovering and validating biomarkers for neurodegenerative diseases is especially high, because once symptoms appear, damage is already severe and perhaps irreversible [13]. Reliable biomarkers which can predict disease risk in presymptomatic or prodromal individuals would be of particular use. Beyond prediction, biomarkers have the potential to become part of routine diagnostic, screening, and monitoring measures. This is already common in the realm of cancer [14, 15] and such practices could be implemented for screening and diagnosis of neurodegenerative diseases as treatments become available. This review will focus on two prevalent neurodegenerative diseases: Alzheimer's disease (AD) and frontotemporal dementia (FTD), also referred to as frontotemporal lobar degeneration. Except for the controversial 2021 approval of aducanumab, a monoclonal antibody directed against amyloid β (Aβ) for AD, there are currently no other approved disease‐modifying therapies for AD or FTD [16].

Given the need for effective treatments for AD, numerous therapeutics are in various stages of development from preclinical characterization to clinical trials [17]. Such studies have taken advantage of known biomarkers to guide participant enrollment and endpoint determination. The Clarity AD clinical trial, which began in 2019, tested lecanemab, an anti‐Aβ monoclonal antibody for people with early‐stage AD [18]. It showed efficacy at reducing amyloid burden on imaging, and moderate but significant effects on slowing cognitive decline with a treatment length of 18 months. Drawbacks included a high level of infusion‐related adverse reactions as well as evidence of cerebral edema in a proportion of patients [18]. A recent case report of cerebral hemorrhage after receiving lecanemab suggests that the long‐term safety needs to be clarified [19]. While the primary endpoint in the lecanemab trial was change in the Clinical Dementia Rating Scale, biomarkers measured in the CSF and plasma were used to enroll patients and throughout the study to monitor treatment effects.

The recent TRAILBLAZER‐ALZ 2 clinical trial highlighted another monoclonal antibody, donanemab, as highly effective at lowering amyloid plaque burden in those with previously demonstrated amyloid and tau deposition on positron emission tomography (PET) imaging [20]. Treatment was associated with lower risk of progression from mild cognitive impairment (MCI) to dementia and from mild dementia to moderate dementia. This trial similarly included plasma biomarker measurement as an exploratory endpoint.

Lecanemab and donanemab, while early stage and imperfect, do show efficacy and represent the initial foray into therapy for Alzheimer's dementia. There is still room for development of less expensive, more accessible disease‐modifying therapies which can benefit patients with more advanced disease. Other neurodegenerative diseases which are unlikely to be affected by anti‐amyloid therapies such as FTD represent a substantial disease burden. This burden is expected to grow as the population ages. One study predicted at least 130 million people to be affected by dementia alone by 2050 [21] which is not even to account for other psychiatric or motor‐predominant neurodegenerative diseases. Therefore, the impetus to solidify biomarkers to accurately detect disease and inform treatment response is of utmost priority.

2. THE ROLE OF IPSC TECHNOLOGY

While studies on biomarkers most often use ex vivo material from human participants, it is also desirable to manipulate and measure biomarkers in more highly controlled and reproducible experimental systems. Such systems allow for mechanistic investigation of biomarker function in a way which is not feasible in human subjects, as well as measurement of biomarker response when testing preclinical therapeutic candidates.

In the past decade, the push for studying diseases using more sophisticated models has driven the development of stem cell technology. Briefly, a sample of somatic cells, such as skin fibroblasts or peripheral blood mononuclear cells (PBMCs) is reprogrammed by directing them into a less differentiated state resembling embryonic stem cells. In 2006, Takahashi and Yamanaka [22] made the landmark discovery that only four transcription factors, Oct4, Sox2, Klf4, and c‐Myc, were required to reprogram mouse fibroblasts into an undifferentiated inducible pluripotent stem cell (iPSC) state. This paradigm was later replicated in human cells [23, 24], ushering in the stem cell era. Subsequent work described alternative combinations of reprogramming factors [25, 26] and reduced the minimum necessary factors to Oct3/4 alone [27]. Studies have used highly efficient lentiviral or retroviral vectors for reprogramming. Since these vectors integrate into the genome and transgenes persist in the cells [28], alternatives have been developed including the nonintegrating RNA Sendai virus [29], episome‐based reprogramming factor expression [30], mRNA‐based transduction systems [31], and transposase‐based systems removable by Cre‐lox [32]. Other groups have elucidated methods for cell reprogramming using small molecules [33]. These advances have positioned iPSC technology to be at the forefront of disease modeling and therapeutic screening. iPSC methods have been applied to a neurodegenerative context, including AD, FTD, diffuse Lewy body disease, amyotrophic lateral sclerosis (ALS), Parkinson's disease (PD), among others [34, 35, 36, 37, 38]. Of most obvious interest to dementia is the creation of iPSC‐derived neurons, astrocytes, and microglia (Figure 1). Such models have been used to study cellular dynamics in both healthy states and diseases. Somatic cells can be obtained from either healthy control individuals, or from patients with genetic or sporadic neurodegenerative disease. Use of CRISPR‐Cas9 and other genome editing technologies allows for correction or introduction of a particular disease‐associated mutation [39]. This allows comparison of the disease states to their isogenic controls, which reduces genetic background‐driven variation, and thus a greater confidence in detecting disease‐specific mechanisms.

FIGURE 1.

Overview of induced pluripotent stem cell technology for modeling neurodegenerative diseases. Peripheral cells such as fibroblasts or peripheral blood mononuclear cells can be sampled from patients or healthy controls and reprogrammed to express pluripotency markers using integrating or nonintegrating viral vectors. Neural progenitor cells can be differentiated to neurons or astrocytes, and hematopoietic progenitor cells can be differentiated to microglia. Mature differentiated cells perform physiological functions which recapitulate the cell type of interest. This includes, amongst others, astrocyte endfoot physiology and lipid homeostasis, neuronal synaptic formation and electrophysiological activity, and microglial phagocytosis and cytokine signaling. Induced pluripotent stem cell (iPSC)‐derived cells from patients also recapitulate disease‐associated biomarker phenotypes including glial acidic fibrillary protein (GFAP) and apolipoprotein E (APOE) expression; amyloid beta (Aβ), phosphorylated tau (p‐tau), and neurofilament light chain (NfL) accumulation; and complement signaling and synaptic phagocytosis.

iPSC‐derived models provide a novel and attractive platform for biomarker discovery, functional investigation, and high‐throughput therapeutic screening. In vitro systems are a reproducible means to clearly define the identity, magnitude, and directionality of change of known biomarkers in a cell‐specific and disease stage‐specific manner. The disease stage is especially salient, because a reliable biomarker which can be detected before onset of clinical symptoms could inform prevention strategies and early interventions once disease‐modifying therapies become more available. iPSC models can also provide a paradigm for measuring treatment response to experimental therapies. iPSC models can be used for mechanistic studies of known biomarkers whose function remains underinvestigated, with exploratory multi‐omics readouts including proteomics, lipidomics, transcriptomics, and metabolomics (Figure 2). They are also well‐suited to discover novel biomarkers with clinical translatability potential, especially for rare genetic variants where large cohort sizes are impossible, and for diseases with overlapping clinical features where the differential diagnosis is increasingly reliant on molecular profiling and imaging. Additionally, iPSC technology allows for differentiation along different lineages to generate multiple cell types, allowing researchers to generate complex, interconnected models using biological material directly from affected individuals. Given the incredible potential, this field has not reached its peak, with certain diseases remaining relatively understudied, leaving clinicians and researchers with a paucity of reliable biomarkers and much room for discovery.

FIGURE 2.

Use of induced pluripotent stem cell (iPSC) models to conduct in vitro biomarker investigation and discovery. Differentiated iPSC‐derived neurons, astrocytes, microglia, or cocultures can undergo targeted experimental treatment, or high‐throughput drug screening. Targeted readouts include measuring specific, predefined biomarkers already used clinically. Exploratory multiomics readouts include proteomics, lipidomics, and transcriptomics. Aβ, amyloid beta; miRNA, microRNA; NfL, neurofilament light chain; p‐tau, phosphorylated tau; t‐tau, total tau.

The present review will discuss the current state of clinical biomarkers in neurodegenerative disease with a focus on AD and FTD, along with a discussion of how iPSC models are being used as platforms for biomarker research.

3. AD BIOMARKERS

Dementia is defined as a significant reduction in at least one cognitive domain which causes impairment in daily functioning [40]. AD is the most common cause of dementia, though additional pathology such as vascular disease and Lewy body pathology often coexist with AD, especially in those over 65 [40]. AD classically presents as a slowly progressive amnestic disease with marked hippocampal atrophy and eventual global cognitive decline [41, 42]. Less commonly, nonamnestic AD presents with deficits in language, visuospatial abilities, and executive function, but relatively preserved episodic memory [43], though these presentations are typically in younger patients without familial mutations [44]. AD is generally categorized into two forms: familial AD (fAD), and sporadic AD (sAD). In fAD, highly penetrant autosomal dominant mutations in amyloid precursor protein (APP), presenilin 1 (PSEN1), or presenilin 2 (PSEN2) lead to early onset usually before the age of 65 [45]. Despite the potent effect, these mutations are rare and account for only a portion of early‐onset AD cases, which itself accounts for <10% of all AD cases [46, 47]. Late‐onset, sAD accounts for the majority of disease burden and is now understood to be polygenic in nature, with variants in apolipoprotein E (APOE) conferring the most genetic risk [48]. Even with the large effect of APOE variants, sAD risk is not fully explained by APOE. More than 40 other loci have been linked to LOAD risk through genome‐wide association studies [49]. Despite the genetic heterogeneity, the neuropathological hallmarks of early‐onset and late‐onset AD have considerable overlap. Both forms of the disease are characterized by intracellular hyperphosphorylated tau protein (p‐tau) forming neurofibrillary tangles (NFTs), and extracellular plaques containing aggregated Aβ oligomers [42]. Neuroinflammation, synaptic degeneration, and lipid homeostasis shifts have also been posited as central to AD pathology [50, 51]. Animal and immortalized cell line models of AD driven by familial mutations have been extensively employed to understand mechanisms common to both forms of the disease, such as the biology of Aβ formation [52].

3.1. Aβ and Tau

Given their central presence in AD, Aβ, total tau (t‐tau) and p‐tau have been long investigated as AD biomarkers [53]. Aβ, t‐tau, and p‐tau can be measured in the CSF of individuals with suspected or confirmed AD, and changes in Aβ have been detected even in presymptomatic individuals who later developed dementia [54, 55]. Measuring these biomarkers can aid clinical decision‐making, especially in discriminating among dementias which can closely mimic AD, such as FTD, as well as primary psychiatric causes of cognitive impairment. The most widely used CSF biomarkers to distinguish AD from the healthy state or other non‐AD pathology are (1) total Aβ, (2) the Aβ42/ Aβ40 ratio, (3) t‐tau, and (4) p‐tau [41, 56]. The Aβ42/ Aβ40 ratio represents the relative propensity of γ‐secretase, a main APP processing enzyme, to generate aggregation‐prone 42‐amino acid Aβ peptide versus the physiologically most abundant 40 amino acid peptide [57, 58, 59]. Levels of total CSF Aβ as well as the ratio of Aβ42/ Aβ40 correlate inversely with levels in the brain, with lower levels in the CSF thought to indicate either increased deposition of Aβ into neuritic plaques, impaired clearance, or a degree of both [60]. In the healthy brain, Aβ is consistently cleared from the parenchyma, with evidence that much of this clearance happens through a paravascular drainage system termed the “glymphatic system” [61, 62]. This may be an explanation for the inverse relationship between total Aβ measured in the CSF versus the brain. In fact, a lower CSF Aβ42/Aβ40 ratio is correlated with increased amyloid plaque burden on PET scanning [62]. The CSF levels of t‐ and p‐tau; however, correlate directly with degree of NFT burden and cognitive impairment [63, 64] Notably, the progression of p‐tau pathology as measured by the Braak stage [65, 66] and progression of Aβ pathology as measured by the Thal phase system [67] do not correlate neuroanatomically [68]. This discrepancy may support the divergent observations of AD biomarkers, aside from tau being intracellular and Aβ being extracellular. More work is needed to address this topic. Regardless of direction of change, however, one study found that the combination of Aβ, t‐tau, and p‐tau yields a sensitivity and specificity for AD of 80%–93% and 82%–90%, respectively [69].

Given the consistent finding of characteristic Aβ42 and tau levels in the CSF of AD patients, it is important that iPSC neuron models of AD can recapitulate this finding. Indeed, increased Aβ peptides of various amino acid lengths in iPSC‐derived AD neurons is one of the most reproduced findings. One such study examined ratios of various Aβ species in iPSC‐derived neurons from patients with fAD genotypes. Notably, Aβ42: Aβ40 and Aβ42: Aβ38 ratios were elevated across all lines harboring various mutations in either APP or PSEN1 compared with healthy controls [70].The relative amount of Aβ species normalized to Aβ40 was consistent when measured in CSF, iPSC‐derived cortical neuron cultures, and brain tissue for a single patient [70]. Aβ38 is less amyloidogenic than Aβ42, and evidence suggests that its biochemical properties confer some protection against Aβ42 fibril formation and disease progression [71]. A higher Aβ42: Aβ38 ratio was associated with earlier age at death in males with AD diagnosis [71]. Other studies using iPSC neurons and neuron‐astrocyte cocultures have also observed increased Aβ42 and Aβ40 in cell lines with the R278I PSEN1 mutation [72]. This elevated Aβ42: Aβ40 ratio has also been observed in fibroblasts and reprogrammed neural progenitor cells from PSEN1 mutation carriers [73].

It is also important to consider Aβ species which are secreted into the cell culture media, which are arguably more relevant for biomarker investigations in iPSC models, since most biomarkers are measured as free molecules in fluids. A study which generated iPSC‐derived neurons from fAD patients harboring the APP London mutation (APPV717I) was able to track the increase in activity of β‐ and γ‐secretase throughout the differentiation process, and at the same time observe an increase in secreted Aβ42 [74]. Here, the β‐secretase and secreted Aβ42 levels were dampened by treatment with a γ‐secretase inhibitor [74]. Another study using iPSC‐derived neurons from both sAD and fAD patients found similar results: an increase in secreted Aβ42 and Aβ40, but an increased Aβ42: Aβ40 ratio was observed only in fAD neurons [75]. Like what has been observed in other studies [74], the amount of secreted Aβ also gradually increased throughout the differentiation course until Day 70, which the authors described as a maturation‐dependent secretion profile [75].

While tau has been heavily studied as a clinical biomarker, it also plays an important physiological role. Tau is a microtubule‐binding protein highly expressed in neurons [76]. Among its key functions include providing a structural basis for axonal cargo transport between the soma and dendrites and stabilizing neuronal structure and interactions with other neurons and glial cells [76]. The hyperphosphorylation of tau results in loss of tau function, decreased solubility, and aggregation [76].

iPSC models of AD have similarly been scrutinized for their ability to produce the distinctive tau phenotypes seen in vivo. Muratore et al. [74], aside from describing AD‐like Aβ signatures in iPSC‐derived APPV717I neurons, showed increased t‐ and p‐tau in AD neurons. This phenotype was partially reversible with treatment with anti‐Aβ treatment early in differentiation [76]. Other studies have found similar phenotypes of increased p‐tau in iPSC‐derived neurons or organoids from patients with fAD, including mutations in PSEN1 [77] and APP [78], and in organoids homozygous for APOE ε4 [79].

NFT burden in AD, as measured by the Braak staging system, strongly correlates with cognitive symptoms and disease severity [80, 81], making tau a potential therapeutic target. High‐throughput drug screens are large‐scale methods of identifying compounds with anti‐tau properties. Traditionally, such screens used human immortalized cell lines which have minimal overlap with terminally differentiated neurons. iPSC models are an alternative that closely reproduce cellular phenotypes, but multistep differentiation protocols and heterogeneity within and between preparations had hampered high‐throughput screening efforts. Recently, groups have developed more straightforward methods for iPSC differentiation. Wang et al. [82] differentiated a homogenous population of glutamatergic neurons capable of firing action potentials. Taking a pharmacologic approach, a compound screen revealed two adrenergic receptor agonists as having the greatest tau‐lowering effect, which was confirmed via follow‐up experiments applying alternative adrenergic agonists or antagonists. This study highlights the potential for using iPSC models in high‐throughput drug screening applications followed by monitoring the response of known biomarkers.

3.2. Central nervous system‐derived extracellular vesicle proteins

There has been recent interest in extracellular vesicles (EVs) as a potential source of blood and CSF biomarkers for dementias [79, 83, 84]. EVs are a type of membrane‐bound vesicle secreted by cells into their extracellular matrix, for both local and systemic transport. EVs can pass through the blood–brain barrier into the CSF and peripheral circulation [85], which positions them as promising windows into neuropathology. Central nervous system (CNS)‐derived EVs can be detected in the blood and contain specific protein markers which identify the cell type of origin [86], and many studies have taken advantage of this to characterize EV cargo in AD versus non‐AD states. Differences in blood neuronal‐derived EV (NDE) content have been used to distinguish cognitively unimpaired controls from both AD patients and stable MCI patients, stable MCI patients from AD patients, and stable MCI patients from MCI patients who later developed AD dementia [87]. Blood NDEs from AD patients were found to contain increased Aβ42 and p‐tau 181 [88, 89], and decreased levels of growth factors and synaptic proteins such as pro–brain‐derived neurotrophic factor (proBDNF), AMPA receptors, synaptotagmin, synaptopodin, and synaptophysin [89, 90]. Markers of neuroinflammation such as complement‐related proteins were increased in blood AD NDEs compared with controls, while complement system regulatory proteins such as decay‐regulating factor and mannose‐binding lectin were decreased [90]. Less work has been done on non‐neuronal derived EVs, but one study found decreased levels of glial‐derived neurotrophic factor in blood EVs derived from astrocytes in AD patients [91]. The same group later found decreased levels of growth factors such as fibroblast growth factor 2 and type‐1 insulin‐like growth factor in blood EVs derived from chondroitin sulfate proteoglycan 4 type neural cells isolated from AD patients versus controls [92]. Other groups are developing multiprotein panels of total blood EV‐derived biomarkers for maximum diagnostic sensitivity and specificity [79]. In this context, it should be noted that many studies isolating NDEs from blood use immunoaffinity techniques with antibodies against L1CAM [87, 90, 93], a cell adhesion protein thought to be specific to neurons. Others use alternative markers or ultracentrifugation methods based on EV physical properties [79, 88, 89, 94]. Additional comparative studies may be needed to validate the sensitivity and specificity of these EV isolation approaches [94].

iPSC models have given some insight into the differences in synaptic markers observed in peripheral fluids. Zhao et al. [95] generated cerebral organoids from AD patients or healthy individuals, with either APOE ε3/ε3 or ε4/ε4 genotype. Organoids consisted of multiple cortical‐like layers with neurons positive for markers of deep and superficial cortical layers, as well as a ventricle‐like fluid‐filled space, and glial acidic fibrillary protein (GFAP)‐expressing astrocytes integrated into the neuropil. Analysis of organoids at 4 weeks displayed transcriptional upregulation of presynaptic marker synaptophysin and postsynaptic marker PSD95 in AD organoids regardless of genotype. This was observed also at the protein level: synaptophysin increased at 4 weeks in APOE ε4/ε4 AD neurons compared with both APOE ε3/ε3 AD and controls; however, PSD95 did not display an effect of APOE allele. At 12 weeks, however, protein levels of both synaptophysin and PSD95 were decreased. The authors propose an accelerated maturation followed by hastened degeneration of AD neurons, which has been reported elsewhere [96]. The use of an iPSC model here provided a genotype‐specific temporal insight into the clinical observation of decreased EV‐derived synaptic markers in the periphery [95].

3.3. Micro‐RNAs

With the increasing availability of sequencing technologies in both research and the clinic, clinicians and researchers can gain unprecedented insight into the transcriptomic changes present in disease. Micro RNAs (miRNAs) are a class of RNAs which are 21–25 nucleotides in length and regulate gene expression through translational repression by promoting degradation of mRNA transcripts [97]. miRNAs can be detected extracellularly secondary to active secretion such as in EVs or attached to lipoprotein carriers [6]. They can also be translocated outside the cell after cellular damage or necrosis as free‐floating species [98]. miRNAs are enriched and particularly stable in biofluids when contained in EVs, making this system attractive for the discovery of reliable biomarkers [97]. They also represent a relatively straightforward readout in experimental systems and potential therapeutic delivery platforms [99]. Changes in levels of various miRNAs have been detected in the brain tissue of AD patients. In particular, miR‐125‐b, which is associated with promotion of tau phosphorylation [100], has been repeatedly found upregulated in whole brain tissue lysate from temporal lobe neocortex [101], medial frontal gyrus [102], and hippocampus [102]. This has been paralleled by detection of increased miR‐125‐b isolated from total RNA in the CSF of AD patients [103].

Analysis of miRNA species in the blood has yielded opposite associations for miR‐125‐b: One study observed 105 patients aged 65 or greater with probable AD along with 150 age‐matched healthy controls for changes in serum miRNA isolated from total RNA [104]. miR‐125‐b was among one of the miRNAs found downregulated in AD patients compared with controls, with the highest discriminatory sensitivity and specificity (80.8%/68.3%, respectively). Indeed, other studies have also observed this downregulation of total RNA‐derived miR‐125‐b in the serum or plasma of AD and MCI patients compared with controls [105, 106, 107]. Interestingly, Tan et al. [104] found that blood miR‐125‐b was negatively correlated with Mini Mental State Exam (MMSE) score, with higher levels associated with lower exam scores, indicating more severe cognitive decline.

To this end, a recent study used iPSC‐derived neurons to investigate mechanisms of miR‐125‐b regulation. In particular, the proteins responsible for epigenetically modifying miRNAs are poorly characterized. One of the more studied proteins, NSun2, is an RNA‐methyltransferase known to exert regulatory effects on noncoding RNA species, but its role in regulating miR‐125‐b in brain was unknown. Kim et al. [108] found markedly decreased levels of NSun2 protein in neurons in the hippocampal formation and BA9 prefrontal cortex in AD patients. Downregulating NSun2 in iPSC‐derived neurons resulted in increased p‐tau burden at several phosphorylation sites including pSer 199‐202, pSer 214, pSer 262 and pSer 396‐404. Overexpression of miR‐125‐b also resulted in increased pSer‐214‐tau, which was exacerbated when the miR‐125‐b vector was mutated to remove the methylation site. Overexpression of NSun2 in iPSC neurons was able to attenuate the increase in pSer‐214‐tau and improve cell viability after treatment with Aβ oligomers [108].

The apparent discrepancies among miR‐125‐b abundance in brain, CSF, and blood remain a point in need of further clarification. First, miRNA profiling studies in patient cohorts are usually examining miRNA derived from either blood, CSF, or brain total RNA and not necessarily enriching for miRNAs contained in EVs [102, 103, 105, 107]. It is therefore unclear to what extent studies are capturing the subtle distinction between miRNAs released from dying neurons versus in actively secreted EVs. Additionally, AD progression is a complex and temporally dynamic process which proceeds over years to decades [109]. Human cross‐sectional studies typically sample from patients at one or two points in time. Emerging evidence of cell‐type dependent differences in vulnerability to AD pathology [110], suggests that cross‐sectional studies from living donors, postmortem brain tissue analyses, and iPSC models are all capturing pathology at different disease stages, perhaps dominated by different cell types. Finally, the biological targets of miRNAs and their regulation are still poorly understood. With this in mind, iPSC models are well‐poised to enhance our mechanistic understanding of miRNAs in disease, which should inform their use as potential biomarkers.

miR‐125‐b is not the only miRNA candidate of interest in biomarker studies. Garcia et al. [111] combined a cross‐sectional clinical study with an iPSC model to investigate miRNA perturbations in AD. They measured total RNA‐derived miRNA in the CSF of a small cohort of five patients with mild AD, and five controls, and found an upregulation of miR‐21, the exact role of which is still unclear but thought to be involved in the regulation of astrocyte reactivity [112, 113] and neuronal response to Aβ [114]. Notably, they did not detect any differences in CSF miR‐125‐b, though the small sample size may have hindered statistical power. The upregulated miR‐21 was then investigated in iPSC neurons and astrocytes from AD patients with the Finnish PSEN1∆E9 mutation, a cause of fAD. Compared with control neurons, AD neurons showed an upregulation of intracellular miR‐125‐b, but a downregulation in EVs. iPSC‐derived microglia showed an increase in intracellular miR‐21, while neurons showed an increase both in the intracellular compartment and in EVs [111]. AD unstimulated astrocytes, however, showed upregulated miR‐21 only intracellularly, with a corresponding downregulation in EVs in the resting state [111]. Upon stimulation with complement component 1q (C1q) + interleukin 1‐alpha (IL‐1α) + tumor necrosis factor‐alpha (TNF‐α), AD astrocytes showed an apparent decrease in intracellular miR‐21 and corresponding increase in EV miR‐21 [111]. These results suggest that astrocyte reactivity changes the secretion pattern of miRNA species into their extracellular environment.

3.4. Glial fibrillary acidic protein

While most biomarker research has focused on proteins which are derived from stressed or dying neurons, it has become apparent that non‐neuronal cell types are also key players in the pathogenesis of neurodegenerative diseases [115, 116, 117]. Astrocytes provide structural and metabolic support to neurons and engage in signaling with other CNS cell types including microglia, oligodendrocytes, and other astrocytes [118, 119]. While thought to be protective in early disease stages, astrocytes can eventually acquire a reactive phenotype that propagates neuronal destruction [120]. One hallmark of reactive astrogliosis is increased expression of GFAP [121]. GFAP is an astrocyte cytoskeletal protein, the expression of which increases in areas adjacent to neurotoxic amyloid plaques [122]. One systematic review evaluated GFAP as a potential AD biomarker. They found that plasma GFAP was significantly higher in AD patients compared with healthy controls, higher in patients with MCI than controls, and higher in patients with Aβ pathology [123]. Plasma GFAP levels were not, however, significantly different between AD and MCI patients, which is an important consideration for the clinical utility of this biomarker given that it cannot distinguish between the two.

Measuring GFAP in iPSC‐derived astrocytes from sAD patients has resulted in seemingly disjointed findings. Some have found increased GFAP via Western blotting in astrocytes derived from only female but not male sAD patients [124]. One study created three‐dimensional renderings of cultured fAD and sAD‐derived astrocytes based on GFAP immunolabeling and found decreased cell surface area and volume in both disease forms compared with healthy control cells [125]. Both sporadic and fAD astrocytes showed less arborization and more fibroblast‐like appearance than healthy cells, indicating a potential compromise in the ability of astrocytes to perform physiological functions [125]. Notably, overall GFAP staining intensity was not different among any of the groups [125]. This study raises the interesting point that astrocytes from both fAD and sAD patients were highly similar [125], suggesting that common astrocyte dysfunction may contribute to both sporadic and fAD, at least in some aspects [126].

It should be stressed that the use of GFAP as a biomarker suffers from important caveats, since astrogliosis is not specific to AD but rather is a component of neurodegeneration more generally [118]. For example, plasma GFAP elevation is part of the hyperacute (within 2 h) sequelae in ischemic stroke patients [127], and GFAP upregulation is present during the later stages of differentiation of mixed iPSC‐derived neuron and glial cultures from a patient with prion disease [128]. Using GFAP as a clinical biomarker may then be more appropriate after alternative diagnoses have been ruled out.

3.5. Neurofilament light chain

Neurofilaments, including neurofilament light chain (NfL) are neuronal cytoskeletal proteins known to be released into the extracellular environment following axonal injury [129]. A recent meta‐analysis of 15 studies comprising over 3000 subjects found convincing evidence that plasma NfL is consistently higher in MCI patients versus controls, in AD patients versus controls, and in AD patients compared with MCI patients [130]. The use of NfL as a biomarker to prognosticate has also been studied. An increase in NfL in presymptomatic MCI individuals who later went on to develop disease has been observed in AD, where plasma NfL levels were higher in individuals with MCI with Aβ pathology in CSF or on amyloid imaging (MCI Aβ+) compared with individuals with MCI and no Aβ pathology [131]. Notably, there was no difference in plasma NfL levels in patients with MCI Aβ + versus those with a diagnosis of AD [131], which suggests that NfL can perhaps be used as a proxy for those who will eventually convert to frank dementia. In a cohort of individuals followed in a longitudinal study, those who were cognitively intact at baseline but later developed AD were found to have higher plasma NfL compared with those who remained healthy [132, 133]. A combination of plasma Aβ42, APOE genotype, plasma p‐tau181, and plasma NfL yielded good predictive value for transition to AD (area under the curve [AUC] 0.81) [134].

Despite its promising potential, NfL as a biomarker may not be unique to AD. Similarly, plasma NfL was substantially elevated in patients at risk of genetic Creutzfeldt–Jakob disease who developed symptoms within 2 years of measurement, compared with those at risk who developed disease more than 2 years after measurement and healthy controls. NfL has also been suggested as a biomarker in patients with PD who are already symptomatic [135], as increases in plasma NfL were positively correlated with cognitive decline [136] and advanced motor stage symptoms [137].

Relatively less work has been done using iPSCs to investigate NfL in an AD context. One recent study created a model of the neurovascular unit (NVU) using primary human endothelial cells, smooth muscle cells, and astrocytes, and iPSC‐derived glutamatergic cortical neurons [138]. NfL was detectable in both the tissue chamber and the circulation media, with significantly higher levels in the tissue chamber, indicating effective endothelial tight junctions [138]. Dependence analysis revealed that levels of t‐tau and NfL within the tissue chamber and circulation media were positively correlated [138]. While the iPSC line was derived from a healthy control, expansion of this model to include patient‐derived cells is an exciting opportunity to further study NVU dysfunction in the context of neurodegenerative diseases.

3.6. Apolipoprotein E

An individual's APOE allele status is recognized as one of the most prevalent risk factors for sAD [134]. APOE is the main CNS lipid binding protein which through interactions with lipids and other lipoproteins, mediates lipid and cholesterol transport between cells [139]. The ε4 allele confers risk, the ε3 is considered neutral, and the ε2 allele is protective [140]. One study reported that the presence of the ε4/ε4 genotype results in a 9‐ to 15‐fold increased risk of late‐onset AD [140].

In the CNS, APOE is most highly expressed in and secreted by astrocytes, though other cell types including neurons can express APOE under pathological circumstances [139]. Given the large effect size of risk or protection conferred by the different APOE alleles combined with the secreted nature of the protein product, several studies have investigated the utility of APOE as a biomarker for AD. One such study assessed plasma and CSF APOE levels in 125 Norwegian patients of varying APOE genotypes who were either cognitively healthy controls, had MCI but no AD diagnosis, or those with an AD diagnosis at baseline and again at 24 months later [141]. They found that plasma APOE levels were affected in a genotype‐dependent manner, with carriers of the ε4/ε4 alleles having the lowest levels, and those with ε2/ε3 having the highest. Validation of APOE against known AD biomarkers revealed that APOE plasma levels were positively correlated with MMSE score, higher CSF Aβ42 (suggesting intact glymphatic clearance), and negatively correlated with CSF t‐tau, p‐tau, and NfL levels [141].

Another study used a multiplex immunoassay panel on 112 patients with AD, 396 patients with MCI, and 58 patients with intact cognition to identify potential biomarkers relative to APOE genotype [133]. APOE plasma levels were decreased in both AD and MCI cohort compared with controls at baseline and at 1‐year follow‐up. Additionally, plasma APOE levels were lower in MCI subjects who progressed to AD dementia after 1 year. In concordance with other studies, carriers of an ε4/ε4 or ε3/ε4 alleles had the lowest levels of plasma APOE regardless of diagnosis, and those with an ε2 allele had the highest. In line with observations that neuroinflammation is highly prevalent in AD and other neurodegenerative diseases [13], plasma inflammatory‐related proteins such as interleukin‐13 (IL‐13) and macrophage inflammatory protein‐1 were associated with APOE genotype, with presence of an ε4 allele associated with higher IL‐13 [133].

Several studies have looked at APOE in iPSC‐derived astrocyte models, both looking at secretion of APOE isoforms and the direct cellular functions of APOE. One study differentiated iPSCs into astrocytes from APOE ε3 or ε4 genetic background. Lower APOE was detected both in the cell lysate and in secreted media of ε4, which is consistent with clinical studies [142]. APOE4 astrocytes showed impaired Aβ42 clearance, upregulated genes known to be involved in cholesterol metabolism including CROT, LPGAT1, and PLPP3, and increased levels of cholesterol both intracellularly and in conditioned media [142]. Neurons differentiated from APOE4 genetic background showed increased Aβ42, an observation that provides additional support for previous findings using iPSC‐derived AD model neurons and strengthens a link between an AD risk factor and AD‐related neuropathology.

While APOE has been associated with AD risk, neuroinflammation [143], and control of metabolism [144, 145], a complete understanding remains elusive. One study used iPSC astrocytes to establish a novel link between mitochondrial dysfunction, APOE secretion, and inflammatory signaling [145]. Using cell lines followed by iPSC‐derived neurons and astrocytes, the authors linked mitochondrial dysfunction to APOE expression. After treatment with sublethal doses of antimycin, an inhibitor of electron transport chain complex III, iPSC‐derived astrocytes, but not neurons, displayed increased APOE expression and secretion. At the transcriptional level, APOE was identified as being concordantly upregulated after both inner mitochondrial membrane knockout of mitochondrial‐membrane‐associated transport proteins SLC25A1 and SLC25A4 in antimycin‐treated astrocytes. Gene ontology analysis of transcripts upregulated after astrocyte antimycin stimulation revealed pathways related to cytokine signaling, among other inflammatory pathways. Notably, these pathways were distinct from those elicited by stimulating iPSC astrocytes with lipopolysaccharide to induce a reactive “A1” phenotype.

While clinical cohort studies are more consistent in terms of observing decreased peripheral APOE in individuals with ε4 genotype, iPSC studies such as those discussed above have found either increased or decreased APOE levels depending on the experimental manipulation [142, 145]. While some of the discordance may be due to inherent limitations of iPSC systems, there is also plausible biological explanation for the observed differences. For example, APOE4 and its lipid cargo are particularly vulnerable to sequestration within cells, which results in lipid accumulation and cellular toxicity [146]. Studies may also capture different stages of disease, missing subtle shifts in APOE localization. Additionally, CNS and peripheral APOE derive from different sources: glial cells and the liver, respectively [147, 148]. Evidence shows that APOE in the blood and APOE in the CSF are metabolized and regulated independently, and the two are not necessarily correlated [149, 150, 151]. Thus, any changes in APOE found using iPSC models may be biologically significant, but not necessarily reflected in patient blood. For this reason, it may be especially important to consider CNS and peripheral APOE as separate entities for purposes of biomarker measurement.

Another consideration regarding APOE as a biomarker is the importance of APOE for lipid metabolism throughout the body, not just in the CNS. Levels of APOE in pathological contexts other than AD may or may not follow the same directionality. For example, higher levels of APOE in the plasma have been associated with risk of ischemic heart disease in men, but not women [152]. Despite this discrepancy, AD and cardiovascular disease seem to have a significant association, with hyperlipidemia being a potential common risk factor [153, 154]. This confounding effect means that interpretation of peripheral APOE levels should be considered in the context of other genetic and lifestyle risk factors. At the same time, with further research it offers unique potential to inform patients and clinicians about an individual's risk of comorbidities which often exist alongside dementia.

4. BIOMARKERS IN FRONTOTEMPORAL DEMENTIA

FTD is a heterogeneous group of disorders with differing clinical, genetic, and neuropathological characteristics. FTD can be either familial or sporadic in nature, with familial forms accounting for 30%–50% of cases [155]. The most common causal genetic alterations are found in genes encoding microtubule‐associated protein tau (MAPT), progranulin (GRN), and chromosome 9 open reading frame 72 (C9orf72) [156]. Other rare mutations each accounting for 1% or less are found in genes including valosin‐containing protein 1 (VCP1), chromatin‐modifying protein 2B (CHMP2B), Tar DNA‐Binding Protein (TARDBP), Fused in Sarcoma (FUS), T‐Cell‐Restricted Intracellular Antigen‐1 (TIA‐1), and others. The neuropathology of FTD is similarly varied, with protein inclusions in neurons and glial cells consisting predominantly of hyperphosphorylated tau (40% of cases), TDP‐43 (50% of cases), FUS (5%–10%), among others accounting for a minority of cases [156]. Clinically, patients can present with a spectrum of neurological and psychiatric deficits, which has led to the classification of several subtypes of FTD based on the predominant clinical characteristics. Behavioral variant FTD (bvFTD) is a common presentation and includes loss of inhibition and empathy, development of compulsive or inappropriate behaviors, and deficits in executive functioning [156]. bvFTD cases caused by C9orf72 hexanucleotide repeat expansions commonly include phenotypes which overlap with motor neuron disorders, especially ALS, as well as cortical‐basal syndrome (CBS) and progressive supranuclear palsy (PSP) [156]. Other subtypes of FTD include various language‐predominant variants such as primary progressive aphasia (PPA) [156]. For a comprehensive review of FTD genetics and clinical characteristics see Antonioni et al. [156].

The discovery and validation of biomarkers for FTD has been challenging given the genetic, clinical, and neuropathological heterogeneity, leading to smaller cohort studies with less statistical power. Biomarkers delineated for one form of genetic FTD cannot be assumed to apply to other genetic forms. Nevertheless, biomarkers would be of high clinical use and are being investigated.

4.1. Tau

The MAPT gene encoding tau protein is located on chromosome 17q21 and contains 16 exons, of which exons 2, 3, and 10 are alternatively spliced to generate six protein isoforms [157]. Exons 2 and 3 code for parts of the N‐terminal domain, and exon 10 codes for the second repeat of the microtubule‐binding domain (MBD), referred to as R2. This is in addition to three other MBD repeats R1, R3, and R4. The presence or absence of exons 2, 3 and 10 produces isoforms containing three or four MBD repeats and 0, 1, or 2 N‐terminal elements, designated as 0N3R, 1N3R, 2N3R, 0N4R, 1N4R, and 2N4R [158, 159]. In the healthy adult brain, the ratio of 3R to 4R isoforms is ~1:1 [160].

Variants of FTD caused by sporadic or familial MAPT mutations are considered primary tauopathies and are, in most parts, characterized by shifts in the 3R:4R tau ratio [35]. A key neuropathological characteristic is neurotoxic aggregation of hyperphosphorylated tau, which can be comprised of predominantly 3R or 4R species depending on the specific disease [161, 162]. Some cases of FTD caused by MAPT mutations as well as AD show aggregation of both 3R and 4R isoforms [157].

Given its centrality to the neuropathology of FTD and related primary tauopathies, tau has been investigated as a biomarker. Increased t‐ and p‐tau has been observed in the plasma of FTD patients compared with healthy controls [163], though other studies have found that only p‐tau is increased in FTD plasma compared with controls [164]. Carriers of MAPT mutations do not have higher CSF tau levels compared with those with non‐tau mutations [165]. Importantly, many studies combine clinical FTD subtypes including bvFTD, PPA, PSP, and CBS into a single FTD group regardless of casual variant. Perhaps due to this heterogeneity, evidence suggests that using a combination of biomarkers, for example, p‐tau and NfL (which will be discussed later on) yields better ability to discriminate between FTD and AD [164].

Models of FTD using iPSC neurons derived from patients with MAPT mutations have been established. Cultures composed of iPSC neurons carrying the MAPT‐N279K mutation, the MAPT‐V337M mutation or the MAPT‐A152T variant contain increased numbers of p‐tau‐positive neurons and show increased fragmentation of tau [166, 167]. iPSC neurons with the MAPT‐A152T variant also display increased total and hyperphosphorylated tau, decreased tau solubility, and enhanced neuronal vulnerability to stressors including ETC complex I inhibitors (rotenone), excitotoxic agonists (glutamate and NMDA), proteasome inhibitors (MG132 and epoxomicin), and cytotoxic Aβ1‐42 peptide [168]. iPSC neurons carrying the MAPT A152T variant from three different patients displayed decreased cell viability in response to these stressors compared with control neurons. This effect was reversed almost to control levels upon targeted knockdown of tau, indicating that cell vulnerability in FTD is partially dependent on the presence of pathological tau species [168].

The MAPT 10 + 16 mutation, which results in supraphysiological levels of aggregation‐prone 4R tau, has also been connected to metabolic stress. iPSC‐derived neurons with this mutation showed increased mitochondrial‐derived reactive oxygen species, resulting in increased AMPA and NMDA receptor expression via disturbed intracellular shuttling, and ultimately calcium‐induced excitotoxicity [169]. Application of 4R tau recapitulated these effects, while administration of mitochondrial antioxidants attenuated them, suggesting that targeting tau‐induced mitochondrial dysfunction can ameliorate neuronal distress [169]. Imamura et al. [170] also implicated calcium excitotoxicity in tau‐mediated pathology. iPSC‐derived neurons harboring intronic MAPT 10 + 14 or exonic R406W mutations showed increased 4R tau both intracellularly and secreted into media. MAPT 10 + 14 neurons were more sensitive to electrical stimulation and vulnerable to cell death afterwards. This effect was partially rescued by calcium signaling inhibition, which also enhanced cell survival after stimulation and reduced intracellular and secreted tau.

4.2. Neurofilament light chain

Like in AD, NfL has been proposed as a biomarker for FTD [7]. It is consistently found at higher concentrations in the CSF of FTD patients, who along with ALS and Huntington's disease patients, have among the highest CSF NfL levels across a range of other neurological diseases [171, 172]. Looking within FTD, CSF NfL correlated with disease severity such that higher NfL was associated with a higher degree of white matter atrophy on structural magnetic resonance imaging [173]. NfL levels in the blood are more variable and usually, though not universally, positively correlated with those in the CSF [174, 175]. Blood NfL was observed to exhibit a sharp increase in the prodromal stage, directly preceding symptom onset [176]. One study with over 300 total participants with various dementias found increased levels of NfL in the plasma of clinically diagnosed FTD patients compared with healthy controls, Lewy body dementia (LBD), and a combined “AD and MCI” group [177]. Additionally, the ratio of CSF Aβ42/Aβ40 was higher in FTD patients than in AD + MCI patients, which could be an important parameter for narrowing a differential diagnosis. GFAP was elevated in the MCI + AD and LBD group compared with controls, but no difference was detected between FTD patients and controls [178]. Due to the low sample size of 28 FTD patients, patients with different genetic mutations (MAPT, C9orf72, and GRN) were combined into one FTD group, indicating that at least NfL is valid across different FTD subtypes.

Most studies characterizing NfL have used iPSC neurons derived from patients with ALS motor neuron disease caused by C9orf72 hexanucleotide repeat expansion. A growing body of evidence suggests that ALS and FTD represent two ends of a phenotypic spectrum of the same disease. A proportion of FTD patients have motor symptoms [156], so identifying and studying motor neuron degeneration biomarkers is a needed research avenue. One study differentiated neurons from iPSCs derived from patients with common ALS‐causing genetic alterations, including C9orf72, SOD1, and TARDBP. In all ALS lines compared with controls, neurons displayed marked accumulation of NfL in the soma, which was accompanied by immunocytochemically evident disrupted neurite structure. This pathology was present even in young motor neurons (i.e., early in the differentiation process), and electrophysiological and axon initial segment structural abnormalities became apparent in more mature ALS neurons [178]. The findings link NfL, a protein put forth as a clinical biomarker, to specific dysfunction. Since NfL measured in the plasma of patients is an ostensibly soluble protein shed from stressed or dying neurons, further studies should characterize the extent to which iPSC‐derived neurons release NfL into the cell culture media as a way to enhance the translatability potential of these systems.

4.3. Innate immunity

Inappropriate activity of the innate immune system has been implicated in a variety of neurodegenerative diseases, including AD and FTD. Overactivation of the complement system, leading to synaptic engulfment via phagocytosis [179], and impaired cognitive processing, has been proposed as a major mechanism of cognitive decline [180]. C1q and C3 have been identified as important for tagging synapses and binding to complement receptors on microglia, activating their phagocytic capabilities [181].

Though familial FTD caused by genetic mutations represent straightforward models and account for a substantial percentage of FTD cases, sporadic cases, where family history is more often absent and multiple genes are likely responsible for disease development, should not be ignored. Biomarkers discovered using cohorts of familial FTD patients may be variant‐specific and not applicable to sporadic cases. One recent study investigated serum biomarkers via nanocapillary liquid chromatography–tandem mass spectrometry in a group of 72 patients with sporadic bvFTD in blood samples taken 12 months apart [182]. The analysis revealed differential abundance of proteins related to both innate immunity and calcium signaling. Focusing specifically on innate immunity, several components of the complement cascade were hits including upregulated C3 and downregulated C1s and C7. Increased C3 in the serum of bvFTD patients from the same brain bank had been observed in another study [183].

Though the complement system as FTD biomarkers is promising, the diagnostic utility is questionable given the nonspecific nature of immune reactivity and overlap with other neurological diseases. Increased levels of both intact and processed C3 have been also found in AD patient brains and CSF [184], and levels correlate positively with CSF tau, though the effect was more pronounced in late‐stage AD compared with early‐stage [185]. Integration of biomarkers with lower specificity into multi‐marker panels may improve diagnostic accuracy. One study combined measurements of a panel of peripheral blood proteins including C3 with basic clinical information such as age, sex, and APOE genotype to yield high diagnostic capability for early‐stage AD versus normal cognition (AUC 0.837) [186]. While the practicality needs to be further delineated, these studies suggest that measuring innate immune markers may be a valid avenue of monitoring neurodegenerative disease progression and treatment response.

iPSC models can be used to study biomarker mechanisms in the context of rare familial or de novo disease‐causing variants, since a single patient can provide a theoretically unlimited amount of biological material. One study differentiated astrocytes from iPSCs from patients with nonsense point mutations in charged multivesicular body protein 2B (CHMP2B), a rare cause of genetic FTD [187]. This protein functions in the endolysosomal pathway and its truncation results in defects in cellular cargo trafficking and degradation, which culminates in a buildup of early endosomes and neurotoxicity [188]. The study found a robust upregulation of C3 at the protein level in both iPSC‐derived astrocytes from two patients harboring CHMP2B mutations as well as in brain tissue from CHMP2B mutant mice. Overall, this suggests that astrocytes may be the source of at least some of the C3 protein detected in peripheral fluids in human patients. Additionally, targeting astrocyte pathology may be a valid avenue for combatting neuroinflammation in neurodegenerative diseases.

4.4. Cleaved caspase‐3

Caspases are involved in the initiation and execution of apoptosis, a type of programmed cell death [189]. Caspase‐3 is one of the downstream effector molecules of apoptotic signaling [189]. Aggregates of both caspase‐3 cleaved APP and aggregates of caspase‐3 cleaved TDP43 have been neuropathologically described [190, 191]. Caspases in the CSF or blood may be an indirect readout of proteinopathy in FTD.

One study took the opportunity to validate a previously known biomarker and nominate a novel biomarker using iPSC‐derived neurons. Kim et al. [192] differentiated iPSCs derived from PBMC samples from two patients with sporadic bvFTD into choline acetyltransferase expressing cortical neurons. Patients underwent whole exome sequencing to exclude the presence of causal variants of a variety of neurodegenerative diseases, including PGRN, MAPT, or C9orf72. An iPSC line from a 74‐year‐old healthy control as well as an iPSC line carrying the P301L mutation in MAPT were differentiated in parallel to provide a comparison to healthy state and familial FTD, respectively. Stemming from the knowledge that neuronal death leading to cortical atrophy is a hallmark of FTD, the study investigated markers of cell death in their model. Neurons from the two bvFTD patients displayed higher levels of cleaved caspase‐3 after stimulation with staurosporine to induce cell stress. Notably, healthy control cells and MAPT P301L cells showed no change in cleaved caspase‐3 upon stimulation with staurosporine [192]. Taken together these data suggest that sporadic bvFTD neurons are especially sensitive to cellular stress. The mechanism of this enhanced sensitivity was not described, though future studies using iPSC models are well‐poised to investigate differences in vulnerabilities across sporadic and disease models.

Additionally, this study looked at hyperphosphorylated tau, an established biomarker of FTD. The neurons derived from MAPT P301L carriers displayed both increased total AT8 and increased AT8 localized to the dendrites compared with healthy control neurons [192]. This finding was not observed in sporadic bvFTD neurons, which displayed line‐specific increased FUS or TDP‐43 expression. This set of experiments indicates that iPSC models can recapitulate biomarkers specific to certain disease‐causing mutations and their associated neuropathology.

The nomination of cleaved caspase‐3 as a biomarker for FTD‐induced neurodegeneration was supported by another study which differentiated iPSC neurons from a patient with homozygous GRN −/− mutation. The heterozygous GRN mutation is responsible for FTD, while the homozygous mutation is associated with neuronal ceroidolipofuscinosis [193], a lysosomal storage disorder, though there is evidence of shared pathology between the two disorders. Increased levels of cleaved caspase‐3 were detected at 100 days of differentiation in GRN −/− neurons compared with healthy control with wild‐type GRN [194]. Taken together, these studies provide evidence that cleaved caspase‐3 as a marker of apoptosis and neurodegeneration warrants further investigation.

Like biomarkers associated with innate immunity, similar concerns regarding the specificity of cleaved caspase‐3 apply. Apoptosis is involved in a variety of neurological insults [195, 196]. Increased levels of caspases have been described in the blood of stroke and traumatic brain injury patients [197, 198], even up to 6 months after cerebral infarct [199]. Activated caspase‐3 closely associated with Aβ plaques and NFTs has been described in postmorterm AD brains [200], and has been recently tested as part of a biomarker panel to distinguish early‐stage AD from MCI and cognitively normal individuals [201]. Such multibiomarker panels are a potential solution to the issue of specificity and are worth exploring going forward.

5. PROSPECTIVE DIRECTIONS AND LIMITATIONS

iPSC systems are not without limitations. First, traditional reprogramming methods which convert fibroblasts or PBMCs into iPSCs involve erasing the epigenetic signatures present in the cells [202]. Yet, mounting evidence has described changes in epigenetic modifications such as chromatin accessibility and transcription factor activity in both normal and pathological aging [203, 204, 205]. Use of transdifferentiation protocols, where somatic cells are directly reprogrammed into desired cell types without a pluripotent intermediary state, have been used to generate both astrocytes [206] and neurons [207]. These protocols generate cells which retain epigenetic signatures similar to those of donor somatic cells [208]. These models arguably better recapitulate an “aged” phenotype seen in the human brain in age‐related neurodegenerative diseases.

Another limitation of this technology is the difficulty of modeling sporadic disease, which accounts for the majority of disease cases in both AD and FTD. iPSC‐derived cells from patients with sAD less reliably reproduce the full spectrum of disease pathology, though studies have nevertheless found significant features of sporadic diseases such as oxidative stress and metabolic defects in addition to inconsistent Aβ and tau pathology [75, 209, 210]. Modeling sporadic 4R tauopathies using iPSCs is challenging, because in the absence of a defined MAPT mutation, the epigenetic reset which is necessary to induce pluripotency causes a fetal tau expression profile consisting overwhelmingly of 3R tau with little 4R tau [211]. Transdifferentiation protocols have had success with producing neurons which closely resemble the adult brain 3R:4R tau ratio [212]. iPSC lines from sporadic PSP and CBD patients have been developed [213], but the level of fidelity to displaying tau pathology remains to be characterized.

A particular strength of using patient cohorts for biomarker studies is that putative biomarkers can be validated for robustness across populations and disease stages including sporadic cases, at least for the more common diseases. It is relatively cost‐ and labor‐intensive to generate iPSC lines from multiple genetic backgrounds, let alone the hundreds to thousands of individuals used in human studies. Yet, translatability is a critical component to validating biomarkers discovered using any experimental model, so future work aimed at increasing the scalability of iPSC systems is necessary.

Most iPSC studies until recently have used two‐dimensional cultures of neurons, astrocytes, oligodendrocytes, or other CNS cell types either in isolation as monocultures, or as multiple cell types together in coculture. Two‐dimensional coculture and three‐dimensional organoid systems are an attractive avenue to investigate non‐cell‐autonomous mechanisms of biomarker function and response. For example, iPSC‐derived astrocytes harboring the FTD causal MAPT N279K mutation cocultured with healthy neurons render the neurons more susceptible to oxidative stress and induce neuronal upregulation of genes associated with stress response and apoptosis [214]. Future work could determine whether this mechanism is contact independent; that is, transmitted through some secreted factor which itself could be a novel biomarker. iPSC‐derived microglia models are relatively less mature than neurons and astrocytes, attributed to challenges in differentiation protocols, however great progress has been made recently [38]. Microglia in triculture AD models have been found to interact reciprocally with astrocytes to overproduce complement proteins [215]. Future work can continue to incorporate microglia into iPSC models to enhance potential for biomarker investigation.

Aside from being technically difficult to reconcile multiple differentiation protocols, two‐dimensional models cannot reproduce the three‐dimensional cytoarchitecture and complex functional integration present in vivo. Protocols generating three‐dimensional self‐organized organoid systems with multiple cell layers and fluid‐filled lumens can provide a partial solution [216]. Organoids generated using the same protocol reproducibly form diverse cell types with transcriptional profiles resembling the fetal human brain [217], though the same drawback about resetting the epigenetic clock seems to apply to organoids [218]. Organoid models of AD show Aβ and tau pathology which is responsive to γ‐ and β‐secretase inhibitors [220, 221]. Organoid models of FTD caused by MAPT V337M mutations show p‐tau accumulation, autophagy‐lysosome disruption, and loss of glutamatergic neurons [221]. Overall, organoids represent a promising and more sophisticated use of iPSC technology to model neurodegenerative diseases for biomarker research. Even so, biomarkers discovered using two‐dimensional or organoid iPSC models can eventually be investigated in animal models which provide a fully intact blood–brain barrier and periphery as an immediate precursor to clinical applications.

As iPSC technology matures, there is an interest in observing the effect of disease‐associated human differentiated cells on their microenvironment in physiological context. Transgenic mouse models which overexpress familial‐AD causal proteins including APP or PSEN1 have been widely used for research. They recapitulate Aβ plaque deposition, but there are numerous biochemical differences between the plaques seen in rodents and those seen in humans [222]. Additionally, NFT's containing hyperphosphorylated tau are only observed when mutant tau overexpression is forced through an additional transgene [223], even though in humans tau is considered a hallmark pathology of AD [40].

Stem cell transplantation paradigms, where neural, astroglial, or hematopoietic progenitor cells are injected into neonatal or adult mice and allowed to differentiate in vivo are a promising avenue to pursue (Figure 3). Multiomics readouts could serve as targeted biomarker discovery and analysis. Histology can characterize the effect of the transplanted cells on their host environment. The profiles of the grafted cells can also be investigated for a clearer understanding of differences in maturation and function caused by disease states. One study transplanted human neural progenitor cells derived from a healthy individual into either wild type or a mouse model overexpressing mutant human APP and PSEN1 (“AD mice”) [224]. Human neurons formed synapses with mouse host neurons and closely associated with Aβ plaques. The critical finding, however, was that plaque‐associated human neurons showed exacerbated pathology to the AD state that was not fully reproduced by the mouse neurons in the AD mice. For example, human neurons showed the presence of dystrophic neurites, consisting of accumulated mitochondria and proteins such as neurofilaments, synaptophysin, and vesicular glutamate transporter 1, to a greater degree than transplanted mouse neurons. Additionally, human transplanted neurons showed a reduction in cell density and an emergence of 4R tau splicing form by 6 months after transplantation. Hyperphosphorylated tau accumulation was observed adjacent to Aβ plaques in human transplanted neurons. The same group later showed tangle formation in human neurons 18 months after cell injections in an alternative AD mouse model which was paralleled by secretion of the biomarkers p‐tau 181 and p‐tau 231 into the bloodstream [225].

FIGURE 3.

Prospective directions for induced pluripotent stem cells (iPSC) in vivo. Precursor cells or differentiated iPSC‐derived cells can be xenografted into animal models. After further maturation in vivo, experimental drug treatments can be applied followed by biomarker measurement in blood, cerebrospinal fluid (CSF), and tissue using proteomics, lipidomics, transcriptomics, and histology. Promising therapeutics can undergo preclinical safety and toxicity tests with or without concurrent biomarker measurement.

Another study transplanted human astroglial progenitor cells (APC) into the forebrain of APP/PS1‐21 genetic background mice [226]. The APC were derived from patients carrying either APOE ε4/ε4 or gene‐corrected ε3/ε3 genotype. The resultant astrocytes displayed human‐like morphology, had functioning gap junctions, and maintained a human‐like resting membrane potential. They also reacted to the presence of Aβ plaques by acquiring a distinct hypertrophic morphology. Taken together, these studies indicate that transplanted human cells retain distinct functional characteristics after maturation in vivo, and respond uniquely to disease‐associated pathology. Complex diseases such as neurodegenerative diseases can be most effectively studied when the cellular biology of the models most closely mimics what occurs in the disease. Transplantation models are a promising way to combine the mechanistic potential of rodent studies with the translatability of human tissue. Studies using transplantation models to probe the in vivo mechanisms of disease‐associated pathways identified by in vitro models are needed across the spectrum of neurodegenerative diseases.

Other work in this direction has been with the intention of using stem cells as therapeutic options [227, 228, 229, 230]. There is untapped potential for use of this system to investigate biomarkers across the spectrum of neurodegenerative diseases. With the grafted cells in place, peripheral biomarkers can be measured with the benefit of an intact blood–brain barrier and peripheral circulatory system. Further experimental manipulations or therapeutic screening can be applied and biomarkers measured as an index of response to perturbation or treatment (Table 1).

TABLE 1.

Summary of selected biomarker clinical changes and observations in iPSC‐derived CNS cell types.

Biomarker	Disease	Clinical finding	iPSC cell type	Finding	iPSC study reference
Aβ	AD	CSF Aβ42 compared with other dementia patients [60, 69] or patients with subjective memory complaints [69]	Various fAD APP V717I or PSEN1 (int4del, Y115H, M139V, M146I, R278I) neurons	Aβ42: Aβ40 in CM of all fAD lines Aβ42: Aβ38 in all fAD lines, more marked increase in PSEN1 lines vs. APP V717I	[70]
			fAD PSEN1 L286V and R278I neurons	Increased Aβ42 and Aβ40	[72]
			Various fAD PSEN1 and sAD neurons	Increased Aβ42 and Aβ40 in all sAD and fAD lines	[75]
			fAD APP V717I neurons	Increased Aβ42 in CM	[74]
Extracellular vesicle proteins	AD	Aβ42, tau in blood NDEs AD > MCI > Controls [89, 90] proBDNF and synaptophysin in NDEs from AD vs. Controls [89, 90]	APOE ε4/ε4 sAD neurons	Transcriptionally upregulated PSD95 and synaptophysin at 4 weeks Decreased PSD95, synaptophysin protein at 12 weeks	[95]
miRNA	AD	miR‐125‐b in temporal lobe neocortex [101], medial frontal gyrus [102], and hippocampus [102] miR‐125‐b in CSF [103] miR‐125‐b in blood [105, 106, 107]	Healthy control neurons (IMR90 background)	miR‐125‐b‐overexpression caused increased p‐tau Overexpression of miR‐125‐b negative regulator rescued cell viability and decreased p‐tau	[108]
NfL	AD	Blood NfL in MCI patients who went on to develop AD than those who remained with MCI [131] Blood NfL in HCs who went on to develop AD versus HCs who remained cognitively intact [132]	iPSC‐derived glutamatergic neurons from HC+ primary human astrocytes+ primary human endothelial cells + primary human smooth muscle cells	NfL detectable both in the tissue and in the circulation media NfL in the tissue indicating intact neurovascular unit NfL positively correlated with tau in both tissue and circulation media	[138]
NfL	FTD	in CSF [171, 172] in blood [174, 175, 177] Levels become distinguishable from healthy controls immediately preceding symptom onset [176] CSF levels correlate with degree of white matter atrophy [173]	C9orf72 hexanucleotide repeat expansion motor neurons	Accumulated NfL associated with disturbed neurite structure and electrical overexcitability	[178]
Complement proteins	FTD	blood C3 [182] C1s and C7 [182] blood C3 [183]	CHMP2B G/C 31449 astrocytes	C3 protein, IL‐6, IL‐8 leading to reactive gliosis phenotype Metabolic and autophagy deficits	[187]
Tau	FTD	blood p‐tau [164] blood t‐ and p‐tau [163] Plasma t‐ and p‐tau not higher in MAPT mutation carriers vs. non‐MAPT driven FTD [165]	MAPT N279K neurons MAPT V337M neurons	p‐tau positive neurons tau fragmentation oxidative stress	[167]
			MAPT A152T neurons	t‐ and p‐tau, increased neuronal vulnerability to stressors tau fragmentation Axonal degeneration	[166, 168]
			MAPT 10+ 16 intronic mutation neurons	4R tau, ROS release, AMPA+ NMDA receptor expression leading to excitotoxicity	[169]
			MAPT 10+ 14 intronic and exonic R406W neurons	Increased 4R tau, Ca2+ signaling‐mediated vulnerability to electrical stimulation	[170]

Abbreviations: Aβ, amyloid beta; APOE, apolipoprotein E; APP, amyloid precursor protein; CHMP2B, Charged Multivesicular Body Protein 2B; CM, conditioned medium; CNS, central nervous system; CSF, cerebral spinal fluid; fAD, familial Alzheimer's disease; HC, healthy control; IL‐6, interleukin 6; IL‐8, interleukin 8; iPSC, induced pluripotent stem cells; MAPT, microtubule‐associated protein tau; MCI, mild cognitive impairment; miRNA, micro RNA; NDE, neuronal‐derived extracellular vesicles; NfL, neurofilament light chain; PSD95, post‐synaptic density 95; PSEN1, presenilin 1; p‐tau, phosphorylated tau; ROS, reactive oxygen species; sAD, sporadic Alzheimer's disease; t‐tau, total tau.

AUTHOR CONTRIBUTIONS

J.M., G.H., P.C.: conceptualization, writing, and editing the manuscript. G.H. and P.C.: supervision.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflict of interest.

McInvale JJ, Canoll P, Hargus G. Induced pluripotent stem cell models as a tool to investigate and test fluid biomarkers in Alzheimer's disease and frontotemporal dementia. Brain Pathology. 2024;34(4):e13231. 10.1111/bpa.13231

DATA AVAILABILITY STATEMENT

Data sharing is not applicable to this article as no new data were created or analyzed in this study.