Integration of CRISPR-engineering and hiPSC-based models of psychiatric genomics

Abstract

Neuropsychiatric disorders are highly heritable polygenic disorders arising from the complex interplay of highly penetrant rare variants and common variants of small effect. There is a large index of comorbidity and shared genetic risk between disorders, reflecting the pleiotropy of individual variants as well as predicted downstream pathway-level convergence. Importantly, the mechanism(s) through which psychiatric disease-associated variants interact to contribute to disease risk remains unknown. Human induced pluripotent stem cell (hiPSC)-based models are increasingly useful for the systematic study of the complex genetics associated with brain diseases, particularly when combined with CRISPR-mediated genomic engineering, which together facilitate isogenic comparisons of defined neuronal cell types. In this review, we discuss the latest CRISPR technologies and consider how they can be successfully applied to the functional characterization of the growing list genetic variants linked to psychiatric disease.

1. Introduction

The complex pathology of psychiatric disorders reflects a highly heritable and polygenic genetic architecture. Nonetheless, diagnosis remains based only on symptoms and behaviors, and available treatments prove inadequate for many patients. We predict that functional evaluation of the link between genetic predisposition, molecular mechanisms and cellular phenotypes should facilitate the development of more precise disease identifiers and treatments.

Major psychiatric disorders have been associated with hundreds of risk loci (Grove et al., 2019; Pardiñas et al., 2018; Stahl et al., 2019), which vary in allele frequency (rare to common) and effect size (high to low penetrance) (Gershon and Grennan, 2015). Consistent with overlapping symptoms, there is also shared genetic risk in neuropsychiatric disease (Cross-Disorder Group of the Psychiatric Genomics Consortium, 2013; Hammerschlag et al., 2019). Much of the genetic liability for psychiatric disease is presumed to arise from common variants of small effect (Purcell et al., 2014), which is enriched in non-coding regions and has a presumed regulatory role in gene expression (Roussos et al., 2014;Fromer et al., 2016). This common variation may underlie disease specificity at the gene (The Bipolar Disorder and Schizophrenia Working Group of Psychiatric Consortium et al., 2018) and transcriptome level (Gandal et al., 2018). Integration of transcriptomic analyses from post-mortem brains with genome-wide association studies (GWAS) can predict cell types enriched for these variants (Skene et al., 2018), informing functional characterization studies to uncover the cell type(s) of origin across psychiatric disorders. Altogether, disease liability reflects the total accumulation, distribution and penetrance of risk factors in each patient.

In summary, there remains a critical need to integrate current clinical, imaging and postmortem analyses with expanding genetic studies of psychiatric disease (Insel and Cuthbert, 2009). In this review, we describe the integration of human induced pluripotent stem cells (hiPSCs) and CRISPR genome engineering technologies towards the functional validation of genetic risk factors associated with psychiatric disorders. We highlight current applications of this technology to characterize variants associated with schizophrenia (SZ) and autism spectrum disorder (ASD). We discuss novel CRISPR applications in other fields that could be translated to advance psychiatric genomics, towards identifying key biological factors underlying risk and uncovering novel therapeutic strategies for the prevention and/or treatment of psychiatric disorders.

1. Introduction to hiPSCs models of disease

hiPSCs have the potential to differentiate into any somatic cell type, offering a great advantage for the study of psychiatric disorders, where live brain tissue is not available for study. Moreover, compared to patient-derived post-mortem tissue, hiPSC-derived neural cells are largely free of confounding factors stemming from donor medical and pharmacological history and environmental exposures. hiPSCs retain the genetic background of the donor, and so recapitulate the complex genetic etiology of disease. Importantly, hiPSC-derived neural cells tend to most resemble the prenatal brain (Brennand et al., 2015; Hoffman et al., 2017), making them particularly well-suited for the study of risk factors thought to act during cortical development. Altogether, patient-specific hiPSC models can query the mechanisms by which genetic variants might predispose individuals to adverse molecular and cellular phenotypes that underlie disease risk.

Today, it is possible to generate a variety of hiPSC-derived neuronal (Chambers et al., 2009; Pang et al., 2011; Yoo et al., 2011; Zhang et al., 2013) and glial (Shaltouki et al., 2013; Tchieu et al., 2019; Tcw et al., 2017) cell types for the study of the biological underpinnings of brain disorders. The earliest of these uncovered perturbations in synapse morphology and activity in hiPSC-neurons derived from Rett syndrome patients when compared to controls (Marchetto et al., 2010). Similarly, the first study of idiopathic psychiatric disease revealed reduced neuronal connectivity in SZ hiPSC-neurons (Brennand et al., 2011), and later resolved differences in synaptic function (Yu et al., 2014). Overall, hiPSCs models have great potential to study psychiatric disease, but share inherent limitations that must be thoughtfully considered.

Although hiPSC-based models for neuropsychiatric disease are well-suited to explore cell autonomous molecular and cellular mechanisms underlying disease risk, they fail to fully capture the more complex elements of the 3D spatial organization, cell-cell interactions, complex cell type composition, and functional maturity found in the human brain. Recent advancements in assembling spatially patterned hiPSC-derived organoids into more complex “assembloids” portends future improvements in this direction (Amin and Paşca, 2018). Critically, case/control hiPSC cohort studies are limited by the extent of experimental variation observed (Carcamo-Orive et al., 2017; Kilpinen et al., 2017), which arises at all levels, reflecting heterogeneity different donors (inter-individual) and even from those derived from the same individual or clone (intra-individual). We and others have explored the extent to which biological, technical and/or systematic confounders influence results obtained from hiPSC-derived studies. While a large portion of variability between samples indeed reflects the genomic differences between individuals (Carcamo-Orive et al., 2017; Cuomo et al., 2020; Hoffman et al., 2017), adjusting for stochastic differences in the efficiency of neuronal differentiation can improve reproducibility. In summary, inter-individual genetic variability reduces the statistical power of case/control experiments, even in large cohort studies (Hoffman et al., 2017), limiting the ability to correlate cellular phenotypes to disease-associated genetic variations.

To tackle this issue, disease modeling using hiPSCs can be combined with CRISPR-based technologies to functionally characterize disease-associated variants on a constant (isogenic) genetic background. For example, CRISPR-mediated allelic conversion between disease-associated and wildtype genotypes, and vice versa, can causally demonstrate the phenotypic impact of small changes in the genome (Schrode et al., 2019). CRISPR-editing is particularly advantageous for the study of rare genetic variants, for which carriers are difficult to recruit in sufficient numbers in general cohort studies. Beyond traditional reverse genetic approaches (genotype-to-phenotype), new high-throughput CRISPR-based strategies make possible forward genetic screening (phenotype-to-genotype), evaluating the impact of large numbers of genetic variations, particularly when combined with single cell RNA sequencing (scRNA-seq) (Datlinger et al., 2017). Overall, isogenic comparisons reduce the sample variance associated with phenotypic analysis, improving our power to detect the changes produced by disease-associated genetic variations, even those with small predicted effect sizes.

2. Introduction to the CRISPR-Cas system

The bacterial clustered regularly interspaced short palindromic repeats (CRISPR) and its derivative CRISPR-associated proteins (Cas) evolved from bacterial immune systems and have been repurposed for genome editing in E. coli (Jiang et al., 2013), Drosophila (Bassett and Liu, 2014), zebrafish (Hwang et al., 2013), and mammals, including humans (Cong et al., 2013a). Specifically, Cas9, bound to a synthetic single guide RNA (sgRNA or simply gRNA), recognizes a protospacer-adjacent motif (PAM) of DNA and its endonuclease activity cleaves the complementary sequence.

An engineered catalytically null Cas9, termed dCas9, inhibits gene transcription when targeted to the promoter region of a gene, despite its inability to sever DNA (Qi et al., 2013). Fusing dCas9 to diverse effectors rapidly expanded its uses to transcriptional regulation (Gilbert et al., 2013), genomic loci visualization (Chen et al., 2013), epigenetic modulation (Liu et al., 2016), single base editing (Komor et al., 2016), and even RNA targeting (Nelles et al., 2016). These advantages have made the use of CRISPR-Cas technology widespread in a short period of time and have led to the diversification of its applications far beyond genome editing.

Compared to previously widely used genome engineering tools, such as zinc-finger nucleases (ZFNs) (Porteus and Baltimore, 2003) and transcription activator-like effector nucleases (TALENs) (Moscou and Bogdanove, 2009), the CRISPR-Cas genome editing system is advantageous in several ways. First, it is easier to design a target sequence under the relaxed requirements of the PAM sequence. Second, simultaneous expression of different gRNAs enables multiplexed genome editing at different genomic locations, or even high throughput genome editing screening when combined with single cell transcriptomic analysis technology. Third, CRISPR-Cas9 editing utilizes a less laborious and more straightforward experimental approach than ZFN and TALEN.

The most important application of CRISPR-Cas9 technologies to functionally characterize GWAS risk variants is gene editing, which can empirically demonstrate causality, unlike in silico predictive models such as transcriptome wide association studies (TWAS) (Gandal et al., 2018; Gusev et al., 2018; Pain et al., 2019; Walker et al., 2019) or to experimental associative approaches (Eckart et al., 2016; Lawrenson et al., 2015; Warren et al., 2017). In addition, CRISPR-mediated genomic manipulation allows for endogenous characterization of gene regulatory motifs, which is not possible in transgenic approaches such as massively parallel reporter assays (MPRA) (Ulirsch et al., 2016) or self-transcribing active regulatory region sequencing (STARRseq) (Arnold et al., 2013; Liu et al., 2017). Furthermore, coupling CRISPR to hiPSC models has the advantage of characterizing patient-specific risk variants on a human-specific genetic background, which is not recapitulated by non-human animal validation systems that often focus on knockout phenotypes (Thyme et al., 2019). CRISPR-based technologies are rapidly evolving to accommodate multiplexed and high-throughput assessment of the increasing number of GWAS associated risk variants. Overall, CRISPR-Cas technology is a flexible and unbiased validation tool with great potential to uncover the molecular etiology of common and rare variants contributing risk to neuropsychiatric disorders. The remainder of this section is dedicated to discussing novel CRISPR-Cas applications that could bring us to a closer understanding of these processes.

CRISPR knock-out.

The simplest of the existing repurposing strategies, CRISPR knock-out (KO) relies on the exonuclease-mediated deletion of a coding DNA sequence essential for a protein’s function. Following DNA cleavage, fragments are joined by abundant events of non-homologous end joining (NHEJ), which frequently cause local indels, yet rarely by seamless, homology-directed repair (HDR). Guided by a specific gRNA, often targeted to the start codon, CRISPR-Cas9 genome engineering is capable of rapidly generating null models for in principle any gene. Compared to traditional KO methods, such as homologous recombination, CRISPR-based KOs can be performed more swiftly and simply.

CRISPR knock-in.

An alternative approach involves replacing a coding sequence essential for protein function with a foreign DNA sequence that often includes reporter or selection marker sequences via homology directed repair (HDR). Although HDR naturally occurs at an extremely low rate, targeting with CRISPR-Cas9 can enhance HDR frequency and increase the success rate of generating knock-in models. For instance, a pioneering application of the knock-in strategy using CRISPR-Cas9 targeted the human EMX1 gene in 293T human cell lines and improved the HR rate from nondetectable levels to 0.7% (Cong et al., 2013b). With further optimization to improve HDR efficiency, CRISPR-based genome editing is being widely adapted to generate knock-in models including drosophila (Lamb et al., 2017), zebrafish (Sung et al., 2014), mouse (Miura et al., 2018) as well as hiPSCs (Andersson-Rolf et al., 2017).

CRISPR gene editing.

In addition to generating gene deletions, CRISPR-mediated HDR is useful for the alteration of disease-relevant genetic variations, including point mutations, deletions and duplications of different sizes. This has provided a better opportunity to study the molecular mechanisms of various genetic mutations associated with disease, especially in human cells, when combined with hiPSC technology. CRISPR-Cas9 genome editing in hiPSCs, corrected a point mutation in SOD1, a causal mutation to amyotrophic lateral sclerosis (ALS) (Bhinge et al., 2017). Isogenic comparisons between the SOD1+/mut and SOD1+/+ hiPSC-derived motor neurons (MNs) confirmed that the genetic modification rescued motor neuronal death and poor neuro-morphological development. Furthermore, targeting CAG repeats expanded in hiPSCs derived from a patient with Huntington’s disease, Xu et al. reduced CAG repeat numbers to a normal level (180 to 18 repeats) (Xu et al., 2017). Compared with (CAG)180 hiPSC-neural cells, (CAG)18 cells showed restored neural rosette formation and mitochondrial respiration. These CRISPR-based correction approaches extend our ability to study the phenotypic consequences of genetic mutation through isogenic comparisons.

CRISPR interference (CRISPRi).

In order to achieve efficient transcriptional repression in mammalian genomes, dCas9 can be fused to a repressive chromatin modifier such as the KRAB (Krüppel-associated box) domain of Kox1. Our group employed dCas9-KRAB to establish the feseability of CRISPR-mediatiated functional repression of SZ-genes across hiPSC derived neuroprogenitor cells, NGN2-neurons and astrocytes (Ho et al., 2017). Recently, neurodegenerative disease-linked genes were also targeted using dCas9-KRAB, specifically targting different TSS sites, demonstrating that CRISPRi can specifically perturb endogenous expression of a single isoform, which is particularly notable given that many brain-expressed genes are alternatively spliced (Heman-Ackah et al., 2016).

CRISPR activation (CRISPRa).

A variety of transcriptional activators have been fused to dCas9 (Chavez et al., 2015; Gilbert et al., 2014; Hilton et al., 2015; Konermann et al., 2015; Maeder et al., 2013; Tanenbaum et al., 2014) Each has its own unique advantages, and a systematic comparison of the various dCas9-effectors has reviewed their efficiency in different cell types across multiple species (Chavez et al., 2016). CRISPRa can serve as a tool for disease modeling when genetic conditions result from excessive expression of genes. For example, activation via dCas9-VPR fusion increased both mRNA and protein levels of α-SYNUCLEIN in hiPSC-neurons (HemanAckah et al., 2016) and significantly upregulated the transcription of psychiatric associated genes in hiPSC-derived brain cells (Ho et al., 2017). The efficiency of CRISPRa has been further verified in an in vivo setting, a system more relevant to translational research, to simultaneously activate multiple genes and long non-coding RNAs (using up to 10 different gRNAs) in the mouse brain (Zhou et al., 2018). This proof-of-principle multiplexed regulation of gene expression in vivo demonstrates the potential of CRISPRa to ameliorate deficient gene expression in the in vivo brain, even those that underlie polygenic disorders such as SZ and ASD.

CRISPR splicing.

At the RNA-targeting level, a new CRISPR-Cas system (type VI-D CRISPR-Cas13d or CasRx), shows RNase activity (Konermann et al., 2018). Compared to classical RNAi methods, gRNA-guided RNA-targeting CRISPR systems show little to no off-target effects, and has already been applied for multiplex RNA-knockdown in hiPSC-derived neurons of three splice isoforms of the MAPT gene, which encodes the frontotemporal dementia-associated tau protein (Konermann et al., 2018).

CRISPR for combinatorial gene perturbations.

CRISPR-Cas9 and its derivatives have been repurposed for multiplex pertubations (Cong et al., 2013b; Zheng et al., 2018). By delivering a pooled library of gRNAs vectors, random combinations of multiple gRNAs were transduced into individual cells; a median of 28 perturbations were studied within single cells (Gasperini et al., 2019). Various attempts have been made to discover CRISPR-associated enzymes that prove to be more versitile for combinatorial perturbations, focusing on smaller Cas enzymes (Zetsche et al., 2017) and multiplexed expression of gRNAs (Fonfara et al., 2016). These advances have made possible single cell multiplex gene editing (Zetsche et al., 2017) and isogenic mutiplexed up- or down- regulation of more than one gene at the same time (Tak et al., 2017). While this is a first step, eQTL analysis of disease-associated SNPs showed varied directionality of differential gene expression (some genes are up-, others down-regulated) (Fromer et al., 2016). Although our ability to achieve large-scale bidirectional network perturbations remain very limited, early attempts have upregulated one gene while knocking out the other (Boettcher et al., 2018) and achieved switches between gene up-regulation and KO (Breinig et al., 2019).

CRISPR-based genomic screenings.

The genetic complexity of psychiatric disorders and the growing list of association studies, that up to this date pinpoint to nearly two hundred loci, call for high-throughput functional characterization of common and rare genetic variants. Cas9-expressing cancer cells were probed with a lentiCRISPR gRNA library of almost 18,000 genes, identifying those genes whose loss confers resistance to an anti-cancer drug (Shalem et al., 2014) or drives primary tumor growth and lung metastasis (Chen et al., 2015). The development of cellular barcoding technologies, at the DNA (Adamson et al., 2016; Rubin et al., 2019), RNA (Wang et al., 2019a), and protein level (Wroblewska et al., 2018), in combination with next generation sequencing, have enabled phenotypic CRISPR-screening beyond cell replication/death designs. Moreover, barcoding to link gRNA to single cell sequencing profiles facilitates genotype to phenotype associations at the transcriptomic (Adamson et al., 2016) or epigenetic (Rubin et al., 2019) level. New approaches have reduced the financial burden associated with single cell sequencing for pooled screens. For example, RNA-level barcoding system of multiple molecular targets can be observed by microscopy using fluorescence in situ hybridization (Wang et al., 2019b). A similar approach was used to screen for genes by devising a barcode system at the protein level; compared to DNA-screens, protein-based reporters can be combined with single cell analysis, such as flow cytometry and assaying for an array of phenotypic markers (Wroblewska et al., 2018).

3. Application of CRISPR-Cas9 technologies to hiPSC-based models of psychiatric disease

The functional roles of synaptic genes linked to SZ have been tested using gene knockouts. The Südhof group, for instance, created hESC lines with a conditional mutation of the presynaptic gene NRXN1, large CNV deletions of which have been associated with SZ and ASD (Pak et al., 2015). NGN2-neurons from the NRXN1+/− mutant hESCs exhibited impaired synaptic activity and neurotransmission without a significant effect on neuronal morphology or synaptic number and size. On the other side of the synapse, conditional knockout of SHANK3, a SZ-associated postsynaptic protein, in NGN2-neurons deteriorated neurite growth and neurite arborization (Yi et al., 2016). These SHANK3 knockout neurons displayed reduced synaptic activity, especially linked to I_h channelopathy, a shared electrophysiological deficit between SZ and autism.

Various SZ genetic studies using hiPSC-neurons have advanced to manipulating SZ-SNPs in non-coding regions. One study prioritized functional SZ-SNPs via global chromatin profiling of hiPSC-neurons (Forrest et al., 2017). Alleleic conversion of rs1198588, positioned upstream of pro-neuronal miR-137, altered miR-137 expression, and decreased neurite development and dendritic protrusion formation, suggesting a causal function for this SZ-SNP in neuromorphological development. A recent elegant high throughput functional characterization of over a hundred SZ-GWAS genes were independently knocked out in zebrafish using CRISPR-Cas9 technology, creating a library of brain imaging and behavioral phenotypic signatures for each perturbed gene (Thyme et al., 2019). Recently, our laboratory demonstrated the downstream effects of the allelic conversion of rs4702, causally linking this SZ-GWAS SNP to expression of its predicted cis-target (FURIN), and showing reduced neurite length, and neuronal activity. We recently uncovered an unexpected synergistic effect between SZ-GWAS genes that was not predicted from single gene perturbations, one that converged on synaptic function and linked the rare and common variant genes implicated in SZ risk (Schrode et al., 2019). Combinatorial but not single-gene perturbation resulted in transcriptional profiles with significant positive correlations to post-mortem schizophrenia brain signatures. Overall, although disease risk is widely held to be additive at the population level, our results suggest that within individual patients, risk variants can sum in different ways, and that a synergistic model may be more biologically plausible than an additive model.

In contrast, functional characterization of risk variants in the ASD field has mainly been focused on the effect of rare variants with large effect sizes. For example, neurons derived from hiPSCs carrying a CRISPR mediated knockout of CDH8 have goblal gene methylation patterns in relevant gene pathways similar to those of blood samples from carryers of CDH8+/− (Siu et al., 2019). A heterozygous deletion of this gene was sufficient to impact the downstream expression of other genes significantly associated with ASD (Wang et al., 2017). CRISPR KO of the ASD gene KCTD13 in hiPSC-induced neurons revealed effects on neurite number and electrical activity of the neural network (Kizner et al., 2019). However, small differences in KO strategies and experimental design among studies introduces variability and limits cross-study reproducibility and comparisons of relative gene effects, which raises the need for a more systematic approach for modeling psychiatric diseases in a dish. Towards this, a CRISPR mediated gene truncation via stop codon insertion system allowed systematically evaluation of independent ten ASD genes (AFF2/FMR2, ANOS1, ASTN2, ATRX, CACNA1C, CHD8, DLGAP2, KCNQ2, SCN2A and TENM1) in isogenic cultures of hiPSC-neurons (Deneault et al., 2018). Moreover, CRISPR-based isogenic comparisons were validated in parellel with case/control samples in a follow-up study of CNTN5+/− (Contactin 5) and EHMT2+/− (Euchromatic Histone Lysine Methyltransferase 2) deletions in hiPSCderived NGN2-neurons, linked neuronal hyperactivity to ASD (Deneault et al., 2019). Overall, a growing number of rare ASD genes have been linked to aberrant neuronal activity in hiPSC models.

A major challenge for the field is to translate highly complex genetic insights into medically actionable information. Although genetic studies have identified a growing list of loci associated with risk for psychiatric disease, the precise variants, genes and cellular processes impacted remain unclear. Moreover, given long-history of evidence supporting epistatic interactions between genes (Moore, 2003), there is an overwhelming need to investigate multi-variant and multi-gene interactions in psychiatric genomics. Moving forward, the integration of hiPSC-based models and CRISPR engineering make it possible to evaluate a larger number of genes, alone and together, in a cell-type-specific manner, in order to identify points of convergence between these genes and to detect synergistic effects on phenotypes.

Summary

The discovery of CRISPR genome engineering revolutionized the field of biological studies. When combined with hiPSC-based models, CRISPR technology facilitates functional validation of disease-associated variants through isogenic comparisons across a variety of genetic backgrounds and disease-relevant cell types. Collectively, the coupling of hiPSC-derived models with the CRISPR toolbox makes possible the causal investigation of the molecular mechanism(s) by which neuropsychiatric genetic variants contribute to the etiology of disease, alone and in combination. The genetic complexity of psychiatric disorders not only lies in the quantity of the associated loci, but also in their heterogeneous distribution among the population; meaning that no two patients carry exactly the same genetic risk. Our hope is that ongoing functional genomic studies will lead to improved diagnostics, prediction of clinical trajectories and the development of novel precision medicine therapeutic interventions.

Figure 1. INTEGRATION of CRISPR technology and hiPSC-based models for functional genomic studies of psychiatric disorders.

GWAS identify common variants associated to psychiatric disorders, which are then correlated to differential gene expression patterns via eQTL analysis. CRISPR derived technologies can be used to create isogenic hiPSC models of any brain cell type to functionally validate common and rare variants associated with psychiatric disease. These models, retaining the genetic brackground of the donor, have the potential to reveal molecular and cellular phenotypes even for variants with small effect sizes. In addition, top-down analysis can be perfomed with these CRISPR models to screen for variants directly associated with canonical psychiatric cellular phenotypes, such as neuronal connectivity and neurotransmission. In summary, these approaches allow for causation analysis of loci identified through psychiatric genomics.

Highlights.

• An introduction to the complex genetic architectures of neuropsychiatric disease.

• A review of state-of-the-art CRISPR mediated perturbations.

• Applications of hiPSC- and CRISPR-based methodologies to streamline functional validation of neuropsychiatric disease risk loci.