Traditionally, cell lines and animal models have been widely used for disease modeling, but they present several challenges. Cell lines and/or primary patient-derived cells have been used to model several diseases, including neurological disorders such as Alzheimer’s and Parkinson’s disease (Ferrari et al, 2020; Cetin et al, 2022), immunological disorders rooted in macrophage dysfunction (Daigneault et al, 2010; Chanput et al, 2014; Mendoza-Coronel and Castañón-Arreola, 2016), cystic fibrosis (McCarron et al, 2021), and others. However, some limitations as disease models exist, as outlined below.

1. Primary cells have limited expansion capacity, and certain populations of cells may be difficult to access.

PSC-derived cells offer advantages in all these aspects. Human embryonic stem cells (hESCs) and iPSCs can be differentiated into all the cell types originating from the three germ layers (Thomson et al, 1998; Reubinoff et al, 2000; Takahashi and Yamanaka, 2006; Takahashi et al, 2007) in addition to trophoblast (Kojima et al, 2017) and yolk-sac derived cells (Atkins et al, 2021).

As noted in point 4 above regarding mutations in cell lines, while PSCs may also manifest with mutations, these can be minimized. Three broad sources lead to accumulation of somatic mutations in iPSCs: pre-existing mutations in the starting somatic cells, reprogramming-induced mutations, and passage-induced mutations, all of which can be avoided. A large proportion of iPSC lines were originally derived from skin fibroblasts, which could explain the higher mutational load in starting somatic cells, as skin cells are exposed to environmental stress and are therefore more prone to acquire mutations. However, iPSCs can also be derived from several other starting cells, such as peripheral blood mononuclear cells (PBMCs) (Loh et al, 2009), cells isolated from urine (Xue et al, 2013), mesenchymal stromal cells derived from wisdom teeth (Oda et al, 2010), or human umbilical vein endothelial cells (Panopoulos et al, 2011). PBMC-derived iPSCs have higher cytogenic stability and lower mutational burden as compared to skin fibroblast-derived iPSCs (Panther et al, 2021). Hence, cell types with lower mutational burden may be an optimal choice to reprogram into iPSCs.

The reprogramming method can affect the number of mutations acquired during the reprogramming process. Studies have shown that the use of non-integrating vectors for iPSC reprogramming minimizes the genomic instabilities compared to the use of integrating vectors such as retroviruses (Kang et al, 2015; Schlaeger et al, 2015; Turinetto et al, 2017). The length of iPSC culture can also affect the acquisition of mutations, which can be limited by reducing culture time (Turinetto et al, 2017) and optimizing culture conditions to reduce oxidative stress (Kuijk et al, 2020).

Recent research comparing mutational burden of iPSCs with isogenic embryonic cells during embryogenesis revealed a similar mutational rate and burden (Hasaart et al, 2022), indicating that iPSCs do not have increased propensity to acquire mutations in-vitro. Furthermore, advanced NGS techniques such as whole exome sequencing and whole genome sequencing should be used to estimate the mutational load and prioritize iPSC lines for use in disease modeling and drug discovery.

iPSC-derived cells also offer advantages over animal models of disease. It has been reported that about 92% of drugs found safe and therapeutically effective in animal models fail in clinical trials either due to toxicity or inefficacy (Arrowsmith, 2011; Mak et al, 2014; Seifirad and Haghpanah, 2019). Additionally, more than 90% of existing drugs only work in 30%–50% of people (Roses, 2000). These failures are attributed to the inability of animal models to fully recapitulate human disease phenotypes. For example, studies have shown that gene expression profiles of human autoimmune disease and murine models of autoimmune disease are dissimilar (Liu et al, 2004). Similarly, murine models have been found to be quite inadequate for metabolic disorders like Fabry disease (Ohshima et al, 1997) and Lesch-Nyhan syndrome (Moro and Hanna-Rose, 2020). In contrast, iPSC-derived disease models, being of human origin, have the potential to better recapitulate human disease phenotypes, as observed in cases of complex psychiatric disorders. iPSC-derived cell models of psychiatric disorders offer advantages over animal models in replicating the genetic and physiological complexity of human disease. In addition, they allow for high-throughput neurophysiological assessment of neural networks and biochemical/epigenetic assessment at the cellular and subcellular level. These advantages have been reviewed by Falk et al (Falk et al, 2016).

Monotypic differentiation systems discussed above are good for modeling cell autonomous disease mechanisms. However, many diseases have non-autonomous multicellular contributions, and the disease relevance of functional genomics screens improves as the screening model becomes more like the cell type(s) affected by the disease. This represents another opportunity for PSC-derived cell model development, as PSCs can be expanded and differentiated prior to co-culture, making sophisticated multi-culture and assembloid-type platforms possible. Co-culture systems, where two or more PSC-derived cell types are cultured together, and organoid cultures, where PSCs are differentiated to self-organize into 3D structures, have shown great promise in studying cell non-autonomous mechanisms of disease. Several recent studies have shown that co-culture systems were critical for identifying disease phenotypes that arise from cell-to-cell communication. Neuron-astrocyte co-cultures have shown astrocyte contributions to neurological disorders such as ALS (Zhao et al, 2020), Parkinson’s Disease (de Rus Jacquet et al, 2021), Alzheimer’s disease (Wasilewski et al, 2022), and epilepsy (Ahtiainen et al, 2021). Co-cultures of iPSC-derived cardiomyocytes and cardiac fibroblasts reverted gene expression and electrophysiological properties to a tissue like state (Beauchamp et al, 2020). Neuron, astrocyte and microglia co-culture models have revealed key astrocyte and microglia functions such as cytokine production and synaptic pruning with implications in neurological disease (Harschnitz, 2019; Sellgren et al, 2019; Baxter et al, 2021).

3D organoid systems offer additional advantages over co-culture systems, as they offer a 3D tissue environment and closely mimic cellular organization seen in tissues. Stachowiak et al, 2017 observed abnormal patterns of proliferation of neural progenitor cells (NPCs) in the ventricular zone, intermediate zone and cortical zone in brain organoids modeling schizophrenia which could not have been observed in 2D culture systems. An iPSC-derived cortical organoid model for 22q11.2 deletion syndrome showed transcriptional and electrophysiology defects in neurons and could also be used as a platform to test antipsychotic drugs which could reverse these defects (Khan et al, 2020). Organoid differentiation protocols have now been developed for tissues originating from all three germ layers (ectoderm, mesoderm, and endoderm) and have shown functional properties, which enable modeling human organogenesis, homeostasis, injury repair and disease [reviewed in (Lancaster and Huch, 2019)]. While these sophisticated models are valuable, it is important to balance model complexity versus practicality, as co-culture or organoid systems are often able to do a better job of recapitulating complex disease biology but may be intractable as large-scale screening systems (Durens et al, 2020).

There are several important factors to consider when developing a PSC-derived model, the first of which should be selecting an appropriate cell line or cohort of cell lines. Several repositories exist with patient derived iPSCs from different genetic backgrounds which can serve as valuable models with which to study disease. In addition to genetic background, these cell lines include information on the donors’ age, sex, ethnicity and health status, factors which can further increase the relevance of a disease model (Warren and Cowan, 2018). Depending on the scope of the experiment however, a large cohort of disease relevant iPSC lines may be required. For example, an iPSC-derived screen to uncover disease-associated quantitative trait loci (QTLs) or genome wide association studies (GWAS) may require a cohort large enough to study subtle phenotypic variation at specific loci. In such cases, recruitment of tissue donors may be necessary if there are not enough existing iPSC lines to adequately power the study. Due to the considerable cost associated with generating new iPSC lines, several organizations such as the Next-Generation Genetic Association Studies (NextGen) Consortium and Stanford Cardiovascular Institute Biobank are endeavoring to widen the number of disease relevant iPSC lines available to researchers (Musunuru, 2018).

Of equal importance to gathering relevant iPSC lines is developing a robust and scalable differentiation protocol to a cell type of interest. To this end, gaining a good understanding of the range of differentiation efficiencies to expect by first working in model iPSC lines can aid in prediction of phenotypes that may be encountered in a cohort of experimental iPSC lines. It is also important to make sure that the cell culture requirements necessary to achieve the desired cell type are amenable to the overall experimental design, and in the case of larger population modeling cohorts, is scalable. Finally, establishing differentiation benchmarking metrics, for example, by cell marker expression or functional profiling, is required to ensure that the resultant cell type faithfully represents the phenotypes of interest to the study (Quadrato et al, 2016).

Another consideration when designing an iPSC-derived model is the selection of appropriate controls to isolate genotype-to-disease relationships. When modeling a monogenic disease, this is often accomplished by engineering an isogenic control line using CRISPR-Cas9 to knock in or out the causal gene variant within the same iPSC genetic background (Bassett, 2017). This method is straightforward, however, with a larger cohort of iPSCs it becomes difficult to generate isogenic pairs for each unique donor line. In instances like this with large populations of donor lines, there is a degree of built in locus-specific control as genomic regions in each unique iPSC line can act as controls to one another. This form of locus-specific control is appropriate for use in GWAS or QTL based studies, however the gold standard would still be generation of isogenic controls. When attempting to model more complex polygenic, or highly penetrant diseases, the engineering challenge is greater for establishing iPSC line controls. However, if the causal genes associated with the disease being modeled are well known, the possibility of making multiple gene perturbations simultaneously in the same cell has been shown to be effective (González et al, 2014) and could allow for the generation of isogenic controls for more complex diseases.

Apart from the functional advantages that PSC-derived model systems offer over other disease models, a unique potential that makes PSCs and PSC-derived cells stand out is the ability to model human genetic variation. Studies profiling transcriptomes of hundreds of iPSCs have revealed that genetic background exerts a larger effect on the variation in resultant iPSC lines than any other non-genetic factor such as culture conditions, passage, gender, etc. (Burrows et al, 2016; Carcamo-Orive et al, 2017; DeBoever et al, 2017; Kilpinen et al, 2017). Since iPSCs can theoretically be differentiated into any cell type of choice, researchers can investigate the effect of genetic variation in the cell type of interest based on the disease. DeBoever et al profiled 215 human iPSC lines and showed that it is possible to examine rare inherited variants (CNVs and SNVs) with moderate effect sizes in iPSCs. This is not possible in animal cell models or patient derived cell models where large numbers of rare cell types may be required. As such, genetically diverse iPSCs can be usefully applied to the three cases outlined below.

First, to examine the effect of a particular pharmacological perturbation on multiple genetic backgrounds. Some ethnicities are underrepresented in clinical trials and conducting pharmacological screens on genetically diverse populations of iPSCs and/or iPSC-derived cells might give insights about drug efficacy and toxicity for such populations before a drug is released into the market. A recent study performed a population-based toxicity screen in iPSC-derived cardiomyocytes and neurons generated from an iPSC bank containing 13 homozygous HLA haplotypes, representing 16% of the population in Taiwan and estimated to represent at least 477,611,135 people in the world (Huang et al, 2022). This group reported inter-individual differences in cardiotoxicity and neurotoxicity for the tested compounds, highlighting the utility of such population-based screens for testing drug toxicity and efficacy. Second, the study of protective or pathogenic variants should be done on multiple genetic backgrounds as this can facilitate assessment of modifier alleles which influence phenotypes leading to novel insights into disease biology. Third, large genetically diverse iPSC cohorts can be used to identify cell type specific eQTL (expression quantitative trait loci) phenotypes. Two different studies identified tissue-specific eQTLs in iPSC-derived hepatocyte-like cells (Pashos et al, 2017; Warren et al, 2017). Pashos et al also performed mechanistic validations to identify causal variants, highlighting the power of this approach to accurately identify causal variants of a phenotype. Such population modeling approaches utilizing large iPSC cohorts can be particularly useful in cases where in vivo tissues/cell types are hard to access. Practical considerations for when and how to design experiments with large iPSC cohorts were reviewed recently (Warren and Cowan, 2018). Population based approaches also require the creation of large iPSC banks which contain iPSCs representing different ethnic backgrounds (Ghosh et al, 2022). While Caucasian populations have the most representation in almost all the major iPSC collections, efforts are now being made to improve the representation from other ethnicities. Table 1 lists publicly available iPSC banks along with their ethnic representations. Such iPSC-derived disease models can further be used in unbiased chemical and/or genetic screens to identify disease mechanisms and new therapeutic targets.

Due to their disease relevance and diverse human origin, hiPSC-based disease models have increasingly been used in the study of disease mechanisms and the development of effective disease-modifying therapeutic targets. The first hiPSC large-scale drug screening was conducted in 2009 by Lee et al, 2009 in familial dysautonomia (FD) patient-specific hiPSCs and led to discovery of the role kinetin plant hormone plays in reducing the pathological phenotype of FD (Lee et al, 2009). Since then, several hiPSC based drug screens have been successfully conducted in various iPSC-derived cell types such as cardiomyocytes, hepatocytes, different disease relevant neurons and neuron progenitors [list summarized in (De Masi et al, 2020)]. Some of the reasons to use iPSC-derived disease models for developing novel drugs include:

1. Lack of appropriate disease relevant cell or animal models that can recapitulate all the disease-specific symptoms

In 2017, Kondo et al, 2017 screened a 1258-pharmaceutical compound library in 13 iPSC-derived neurons from Alzheimer (AD) patients with outputs including amyloid β peptide-Aβ40 and Aβ42 secretion and the Aβ42/40 ratio. After two rounds of screening, 27 Aβ-lowering screen hits were shortlisted, of which 6 lead compounds were further prioritized due to their capacity to reduce Aβ40 and Aβ42 levels in most of the 13 sets of AD neurons. Finally, a combination of three compounds, bromocriptine, cromolyn and topiramate were identified to reduce the Aβ42/40 ratio in iPSC-derived neurons from patients with familial AD but not with sporadic AD. In recent years, targeting Aβ aggregates as the major therapeutic target of AD has been re-evaluated and new therapeutic targets are being investigated.

Another example of an iPSC-derived neuronal drug screen is a 3000-compound screen conducted in human neural stem cells and iPSC-derived neurons from children affected with Lesch-Nyhan disease (LND) (Ruillier et al, 2020). LND is a rare monogenic disease caused by deficiency of hypoxanthine-guanine phosphoribosyl transferase (HGPRT) enzyme and characterized by severe neuropsychiatric symptoms that currently cannot be recapitulated in HGPRT-deficient animal models. This screen identified six pharmacological compounds, all possessing an adenosine moiety, that corrected HGPRT deficiency associated neuronal phenotypes of LND.

Amyotrophic lateral sclerosis (ALS) is a neurodegenerative disease characterized by loss of upper and/or lower motor neurons. Ropinirole, Retigabine, and Bosutinib are three candidate anti-ALS drugs that have been identified in three separate iPSC-derived motor neuron-based drug screens (Wainger et al, 2014; Imamura et al, 2017; Fujimori et al, 2018) and are currently being investigated in clinical trials for safety and effectiveness (Okano et al, 2020).

Modern functional genomics screening strategies generally involve CRISPR/Cas9-based tools including CRISPR nuclease, CRISPRi, and CRISPRa (Gilbert et al, 2014; Shalem et al, 2014), modalities that involve different considerations around cell engineering and delivery and readout strategies. CRISPRi and CRISPRa use a catalytically inactive form of Cas9 (dCas9) fused to transcriptionally repressive or activating domains, respectively. Both CRISPR nuclease and CRISPRi allow for testing of the effects of loss of gene expression, but they differ in important ways. CRISPR nuclease cleaves the gene of interest so is irreversible and does not require sustained expression of Cas9 machinery, but can result in truncated or hypomorphic proteins depending on the cut site (Michlits et al, 2017). In addition, it can be delivered recombinantly conjugated with a gRNA via electroporation in the form of a ribonucleoprotein (RNP) complex (Liu et al, 2015), allowing for greater flexibility in delivery mechanism. CRISPRi is transiently active and reversible so most applications require establishment of a cell line with stable expression of the dCas9 fusion protein to maintain repression over time. The repression induced by CRISPRi is generally more uniform across a population versus with CRISPR nuclease but may not be a complete loss of expression depending on the cell type and locus. This lack of complete repression can also be a benefit when the goal of the screen is to mimic the effects of small molecule therapeutics to identify candidate gene targets. CRISPRa screens can provide complementary information to CRISPRi screens, where genes are down- or upregulated at endogenous loci and hits can be compared to identify common hits that are enriched in opposing populations with the two screening modalities.

Other domains that have been fused to elicit different effects on gene regulation include CRISPRoff/on (Nuñez et al, 2021) and other epigenetic editors that have been reviewed elsewhere (Nakamura et al, 2021). Other forms of Cas proteins provide other advantages such as the ability to perturb multiple genes simultaneously with Cas12a, and compact size for efficient packaging and delivery with Cas12f (Wu et al, 2021; Xu et al, 2021).

CRISPR screens require expression of Cas9 machinery as well as the guide RNA targeting the gene of interest, which can be expressed on the same or different constructs. Most screens using iPSC-derived cells with stable expression of CRISPR/Cas9 machinery involve generation of iPSCs with genetically integrated Cas9 machinery and lentivirally expressed guide RNAs (Figure 3A). Alternatively, RNP-based delivery can be more approachable for many cell types but is not suitable for pooled screening unless guides are expressed stably. Likewise, a transposon-based system such as piggyBac can be used for more stable expression of individual and multiplexed sgRNAs, though they have been shown to cause insertional mutagenesis (Li et al, 2017). For stable expression of machinery, options include expression from a safe harbor locus or lentiviral expression. Expression in a targeted safe harbor locus can help prevent silencing of expression machinery over time or upon differentiation (Sadelain et al, 2012), but requires more sophisticated cell engineering techniques and longer timelines for cell line generation as well as clonal selection in the case of PSCs. Lentiviral-mediated expression is simple to implement, and if the construct contains a resistance or fluorescent selection marker then the resulting population can often be used as a pool, thereby limiting concerns about artifacts based on genetic integration site. However, it is crucial to ensure that the CRISPR machinery is expressed and functional on a population-wide level in the differentiated cell population that will be used in the screen, which can be tested using flow cytometry with a guide targeting a cell surface protein. In this assay, a guide targeting a well-expressed but non-essential cell surface protein is stably introduced and cells are analyzed by flow cytometry to verify that every guide-expressing cell has active machinery (Figure 3B).

Doxorubicin is a common chemotherapy compound used to treat various cancers but results in severe side effects, including heart failure. A genome wide pooled CRISPR KO screen in iPSC-derived cardiomyocytes was performed and identified two new targets, SLCO1A2 and SLCO1B3, which are both human-specific transporters whose loss of function protects cardiomyocytes against doxorubicin-cardiotoxicity but does not affect cell death in cancer cells (Sapp et al, 2021).

Stem cell-derived islets generated by directed differentiation from hPSCs are a great source for pancreatic β cell replacement therapy to treat insulin-dependent diabetes. In recent years there has been promising clinical trial data using progenitor cells (Ramzy et al, 2021) or fully differentiated and functional stem cell (SC)-islets (Vertex, 2021). However, the major challenge of protecting SC-islets from an immune response remains. To understand the underlying pathways that drive immunogenicity of SC-islets in inflammatory environments, Sintov et al performed single-cell RNA sequencing and whole-genome CRISPR screen of SC-islets under immune interaction with allogeneic peripheral blood mononuclear cells (Sintov, 2022). The screen results indicated that targeting the JAK/STAT type II interferon pathway by depleting chemokine ligand 10 (CXCL10) will provide reduction of SC-islet immunogenicity.

Examples of screens performed in the context of neuronal disease include a genome-wide pooled CRISPR-Cas9 knockout (KO) screen in human neural progenitors to identify molecular therapeutic targets that disrupt the host-dependent mechanism of Zika virus infection (Li et al, 2019). Another recent example is a kinome-wide (sgRNAs against 736 kinases) survival-based KO screen conducted in hiPSC-derived cortical neurons to identify modifiers of poly (PR) dipeptide repeat protein toxicity, the most common genetic cause of frontotemporal dementia (FTD) and amyotrophic lateral sclerosis (ALS) neurodegenerative diseases (Guo W. et al, 2022). The screen identified NEK6 as a novel therapeutic target for C9orf72-related FTD or ALS, which regulates poly (PR)-mediated p53-related DNA damage.

Some of the challenges with traditional neuronal differentiation methods include multi-step protocols that take several weeks, can be difficult to scale up, and usually yield heterogeneous populations of differentiated cells. One promising alternate method for hiPSC-derived differentiation, however, is transcription factor-directed differentiation of hiPSCs. For example, overexpression of neurogenin 2 (NGN2) transcription factor from AAVS1 safe-harbor locus in hiPSCs directs the differentiation into glutamatergic neurons with high efficiency and homogeneity (Zhang et al, 2013; Wang et al, 2017; Fernandopulle et al, 2018). Tian et al, 2019 utilized Ngn2-driven generation of glutamatergic neurons and performed two CRISPRi/a-based screens with survival or FACS-based phenotypes. In the first study, they revealed neuron-specific essential genes and genes that improved neuronal survival upon knockdown, while in another study the authors identified a novel link between lysosomal failure to ferroptosis in human neurons by knockdown of the lysosomal protein prosaposin (Tian et al, 2021).

Another example of transcription factor-directed differentiation of hiPSCs is generation of microglia-like cells based on the inducible expression of six transcription factors in hiPSCs following an 8-day efficient protocol (Dräger et al, 2022). In this study, inducible CRISPRi and inducible CRISPRa screens were performed against the druggable genome in transcription factor-directed hiPSC-derived microglia, uncovering insights into microglial biology including genes controlling survival, activation, and phagocytosis. The screens identified PFN1 and INPP5D as novel modulators of phagocytosis in microglia, and variants in these genes can possibly be associated with neurogenerative diseases.

Leng et al performed a pooled CRISPRi screen in hiPSC-derived cells, identifying cellular pathways that control cytokine-induced inflammatory astrocyte reactivity (Leng et al, 2022). To generate hiPSC-derived astrocytes in a scalable manner, Tcw et al, 2017 used a modified protocol by overexpressing gliogenic transcription factors NFIA and SOX9 during the differentiation process, as described previously (Li X. et al, 2018). These results and the scalable hiPSC-derived astrocytes platform have the potential to guide the development of therapeutics to selectively modulate different aspects of inflammatory astrocyte reactivity.