Multicellular Systems to Translate Somatic Cell Genome Editors to Humans

Abstract

As genome editors move into clinical trials, there is a need to establish ex vivo multicellular systems to rapidly assess and predict toxic effects of genome editors in physiologically relevant human models. Advancements in organoid and organs-on-chip technologies offer the possibility to create multicellular systems that replicate the cellular composition and metabolic function of native tissues. Some multicellular systems have been validated in multiple applications for drug discovery and could be easily adapted to test genome editors; other models, especially those of the adaptive immune system, will require validation before being used as benchmarks for testing genome editors. Likewise, protocols to assess immunogenicity, to detect off-target effects, and to predict ex vivo to in vivo translation will need to be established and validated. This review will discuss key aspects to consider when designing, building, and/or adopting in vitro human multicellular systems for testing genome editors.

Graphical Abstract

1. Introduction

The Clustered Regularly Interspaced Short Palindromic Repeat (CRISPR) system has revolutionized genome editing due to ease of engineering and programmability [1]. For safe human translation, CRISPR-based gene editing must have no off-target effects in the genome [2]. CRISPR itself or its delivery vehicle may have unintended adverse effects within cells or systemically, which warrants detailed elucidation of the toxic effects in appropriate model systems before human administration [3,4]. Significant progress has been made to modify CRISPR-Associated (Cas) nucleases or the guide RNAs (gRNAs) to tackle specificity [5,6]. Likewise, novel viral and non-viral strategies with reduced immunogenicity and increased delivery efficacy for genome editors are underway [7] (Table 1). Precise genome-wide techniques for detection of off-target insertions/deletions (indels) have also been created [8,9]. In contrast, little progress has been made to examine the gene editors in platforms that can yield human relevant readouts to enable us to rapidly predict and assess toxic effects of CRISPR. Most studies have used human cell lines or animal models, which fall short in accurately predicting the response of primary human cells to genome editors. Through integration of stem cells or multiple cell types, physical and biochemical cues in three dimensions (3D), organs-on-chip and organoid technologies seek to establish ex vivo human models that can capture the complexity of their in vivo tissue counterparts [10–12]. These multicellular systems provide a platform to fill the gap between animal studies and human trials, enabling prolonged maintenance and manipulation of primary human cells ex vivo. Therefore, human-based multicellular platforms can facilitate studies of the efficacy, toxicity, and immunogenicity of CRISPR and its delivery vehicles.

Table 1:

Challenges of CRISPR-based genome editors for human translation

Challenge	Possible solutions
Genome Editor Efficacy	Development of Cas proteins with improved activity. Identification of other Cas proteins (e.g, Cas13) or smaller Cas protein for easy virus packaging. Directed evolution of Cas orthologs.
Genome Editor Efficiency	Improved delivery system, lentivirus, adeno-associated virus (AAV), nanoparticles with cell-specific receptors for target cell delivery.
Genome Editor Immunogenic Activity	Improved delivery system with reduced immunogenicity (e.gs., Cas9 epitope modification, humanized Cas9 protein orthologous to build all-human Cas enzyme), immune orthogonal Cas orthologs
Genome Editor in vitro to in vivo Translation	Development of near-physiological multicellular tissue constructs with controllable flow, oxygen, and long-term culture. Possibility of connecting multiple tissue constructs

2. Multicellular Systems for Modeling Genome Editing in Humans

2.1. Organoids

Organoids are self-organizing 3D cellular structures that develop from adult stem cells or from induced pluripotent stem cells (iPSCs) that resemble organ functionality and tissue patterning similar to in vivo tissues [13,14]. The self-organizing capacity of stem cells enables initiation and execution of intricate cellular interactions and developmental processes that eventually result in functional tissue structures that enable decoding design principles of tissue morphogenesis, physiology, and disease states [15,16].

A pioneering study by Schwank, et al., in 2013 combined organoid and CRISPR-Cas9 technologies for the successful therapeutic repair of the cystic fibrosis transmembrane conductor receptor (CFTR) gene in intestinal organoids derived from cystic fibrosis patients [17]. This approach has been recently extended to repair the CFTR gene in patient-derived upper-airway organoids using CRISPR-Cas9 [18] and to repair a CFTR point mutation in multiple patient-derived intestinal organoid lines using CRISPR-based adenine base editors [19]. Further, tissue-derived or iPSC-derived organoids can be used to create in vitro models of genetic mutations found in cancer patients using CRISPR-based genome editing [20,21].

Organoid biobanks can provide an easy to use, medium-throughput platform to capture disease heterogeneity, sex-specific disease differences, donor-to-donor (epi)genetic variability, and patient-to-patient variable responses to genome editors [19,22]. As most gRNAs are designed based on the human reference genome, these gRNAs run the risk of having off-target effects due to population genome variability [23]. Therefore, it might be possible to use patient-derived organoid biobanks and in silico screening, to validate the best gRNA for individual patients. Organoids contain a proliferative stem cell population that can be propagated continuously, thus, they might be useful models to follow changes in the genome (on- and off-target) over multiple generations after genome editing. One challenge is that some organoid models undergo necrosis due to the accumulation of apoptotic cells and lack of oxygenation and need to be passaged every 7-10 days; therefore, long-term effects of genome editors are limited without disrupting the original organoid population [12].

While organoids provide an emerging platform for assessment of genome editors using human stem cell populations, the lack of immune cells in some organoids [24] can reduce their utility for investigating immune response that could be harmful in vivo [25]. Additionally, in the context of iPSC-derived organoids, limited maturity can impede their capacity to capture adultlike human phenotypes. However, recent advances in iPSC-derived organoids have improved organoid maturation, vascularization, and tissue function which collectively provide a promising outlook for using organoids to test genome editors [11,26–31].

2.2. Organs-on-chips

Unlike organoids in which the 3D architecture spontaneously develops, organs-on-chips offer the possibility of designing topographical features of the native tissue. Organs-on-chips are complex ex vivo models that are built using microprinting, soft lithography, and polymer molds [32,33]. The integration of cell culture flow for perfusion, mechanical stimulation, and oxygen gradient regulation in organs-on-chips mimic features of native tissues in ways unachievable with organoids [10,12,34]. Various organs-on-chip models have been created for brain, lung, liver, gut, heart, bone marrow, thymus, and kidney tissues [33].

An attractive feature of organs-on-chips is engineering physiochemical properties of tissue such as the flow of culture media to allow controlled nutrient replenishment and continuous measurement of secreted cytokines or metabolites that might be affected by genome editing, which combined with multi-omics (e.gs., proteomics, metabolomics) and deep sequencing (e.g., RNAseq), offers a comprehensive analysis of the effects of genome editors in multicellular tissue constructs [34]. However, organs-on-chips models have been heavily developed using cell lines (e.g., cancer cells) that are easy to manipulate and more resistant to stress [10]. Other organs-on-chip models combine cell types from different sources (e.gs., immortalized cell lines, iPSCs, human and mouse primary cells) which results in tissue-models of mixed cell phenotype population, or tissue-models that represent early states in human development, especially when using iPSC-derived cells that usually do not mature beyond a fetal state [33,35]. The organs-on-chips field has started to use primary cells isolated from patient tissues to faithfully mimic physiological processes [32,33,35]. While the use of primary cells allows for the creation of tissue models for personalized medicine and the study of donor-to-donor variability, the use of primary cells raises additional technical challenges including: a) reproducibility in methods to promote tissue formation due to donor’s age and pathological state, b) cell loss during tissue processing and cell isolation, c) a common media formulation to maintain different cell types, d) a suitable scaffold to support multiple cell types and tissue geometries, and e) the lack of feasibility to mark and enrich primary cells that undergo a genome editing event.

New organ-on-chip applications combine patient-derived iPSCs with microfluidic flow to create organoids or iPSC-derived organs-on-chips that can also improve cell maturation and tissue organization [12,32]. Altogether, organs-on-chip models offer possibilities as platforms to test genome editors at different stages of technical development; however, it will be critical to engineer organs-on-chip models that enable the viability and assessment of multiple cell types, particularly from cells derived from the same donor. Recent ongoing efforts have also focused on studying human physiology by combining up to ten organ systems modularly [10,36–38]. These efforts together with organoids will establish a human-based multicellular system that enables assessment and improvement of genome editors in a physiologically relevant model and can fill the gap between mouse studies and human trials.

3. Delivery Options for Genome Editors in Multicellular systems

Wild-type CRISPR-associated protein 9 from Streptococcus pyogenes (WT-SpCas9) or some other CRISPR orthologs such as Staphylococcus aureus (WT-SaCas9) induces double stranded breaks (DSBs) in the genome [39–41]. DSBs are repaired either through error-prone non-homologous end joining (NHEJ) or through homology directed repair (HDR) [42]. The improper repair of on-target DSBs can lead to small indels, large deletions, or inversions [43]. Engineered Cas9 variants or other CRISPR orthologs have been introduced to increase the specificity of gene editing, single base pair replacement, targeted epigenome engineering, modification of RNA molecules, and decreased immunogenicity, to name a few [1,5,42,44–46]. CRISPR-Cas genome editors can be delivered using physical methods (e.gs. electroporation or microinjection), viral transduction (e.gs. lentivirus (LV) or adeno-associated virus (AAV)), or nanoparticles (e.g. lipid nanoparticles) [47]. Collectively, all of these methods have demonstrated utility to edit cells grown as monolayers, dispersed single cells in suspension, and in animal models [47,48].

Although the utility of organoids, organs-on-chips, and CRISPR-based genome editing is well documented, there are limited reports using organoids or organs-on-chips as models for CRISPR-based genome editing. The most extensive use of CRISPR-based genome editors with organoids has been to correct a specific mutated gene or to generate organoid cell lines with mutations found in cancer patients, using a parental organoid cell line. For most in vitro editing of tissue-derived organoids, electroporation is the preferred approach for CRISPR-Cas delivery, followed by positive antibiotic selection and expansion of transduced stem cell population [17–19]. Once the edited population is expanded, they can be enriched to detect on- and off-targets genomic effects. This approach works well for organoids that contain a stem cell population but might not be applicable for terminally differentiated cells present in multicellular ex vivo models that cannot be expanded after editing. Further, as 3D multicellular systems recapitulate some aspect of tissue architecture found in vivo, the delivery of genome editors in these ex vivo models encounters more challenges compared to dispersed single cells in suspension. For instance, cells located at the periphery of a multilayer 3D cellular system will be modified whereas cells in the center might not. Further, unless there is a fluorescence reporter or a selection marker (i.e., antibiotic resistance gene) added into the delivery vector, the identification and isolation of edited cells will be challenging, which will impact the quantitative analysis of on- and off-targets effects.

An additional challenge for efficient delivery of genome editors in multicellular systems is the selective editing of a cell population within the ex vivo model. This can be accomplished by engineering viral vectors with cellular tropism or by engineering nanoparticles with cell-specific epitopes for receptor-mediated delivery. Although both approaches work, in the case of engineered nanoparticles, there might be additional optimization for endosomal escape and effective genome editing [49]. Ultimately, the identification of an efficient delivery method of genome editors to multicellular systems will require the empirical test of multiple delivery methods in parallel, followed by quantitative analyses of delivery efficiency and off-target effects.

4. Capturing CRISPR-based DNA Cleavage and its Off-target Toxicity in Multicellular Systems

CRISPR-Cas uses a gRNA to direct the Cas nuclease to the target site via Watson-Crick base pairing (on-target) to create DSBs; however, the Cas nuclease can often create DSBs elsewhere in the genome (off-target) too [1]. The host DNA repair mechanism in response to induced DSBs can sometimes result in unintended genome alterations such as indels, chromosome rearrangements, and translocations that could affect cellular function, morphology, or the proliferative potential of the edited cells negatively (toxic effect, leading to apoptosis) or positively (turning on oncogenes, leading to over proliferation) [39,40,50].

One of the first approaches to predict and reduce off-target effects is to use in silico screening (extensively review elsewhere in ref [51]). Multiple tools (e.gs., E-CRISP, Cas-OFFinder, COSMID, Breaking-Cas, CRISTA) exist that computationally predict off-target mutations based on the gRNA and the gene editor used [52]. However, computational models often fail to predict rare off-target events and large chromosomal rearrangements and should be paired with additional in vitro, ex vivo, and in vivo testing.

Quantitative in vitro approaches for on- and off-target detection using genomic DNA isolated from edited cells include mismatch cleavage assays, such as T7E1, TIDE, and IDAA, Sanger Sequencing, targeted sequencing, and exome sequencing [50,53]. Circularization for in vitro reporting of cleavage effects by sequencing (CIRCLE-seq) has been proposed as a superior in vitro technique, compared to mismatch cleavage assays, to detect genome-wide off target mutations by CRISPR [54,55]. CIRCLE-seq has two main advantages over previous assays; it is highly sensitive despite possible cell-type-specific single-nucleotide polymorphisms (SNPs), and is able to detect off-target effects without using a reference genome [54]. Furthermore, there are methods analogous to in vitro mismatch cleavage assays, such as genome-wide unbiased identification of DSB enabled by sequencing (GUIDE-seq) and iGUIDE which employ double strand oligodeoxynucleotide tags integrated in sites of DNA breaks in the genome to detect on- and off-target cleaved sites [8,55]. Most recently, DISCOVER-Seq, a method for in situ discovery of DSBs in a single step that preserves the integrity of chromatin state and modifications, could allow better detection of translocations and large deletions [9].

Together, the techniques described above are enhancing our ability to detect CRISPR off-target events with greater precision in purified single-cell type populations, but these techniques require further refinement in sensitivity and detection capabilities to be able to detect large chromosomal rearrangements, especially when using primary human cells in which genetic variability could influence genome editing in unknown ways. There is also the need to perform a direct comparison of these methods using model cell lines to establish best practices across research groups to detect off-target effects. Adapting these techniques to detect off-target events in multicellular systems might require iteration of experimental conditions that mimic in vivo conditions. A first challenge is to pair a highly sensitive method to detect rare mutation events with the appropriate time scale for tissue collection after genome editing. Further, in multicellular systems with primary human cells, it is unlikely that edited cells will have a selection marker (i.e. antibiotic selection), thus their unsorted harvest can result in the dilution of off-target genome changes, especially for rare off-target events. Waiting too long for cell harvesting could lead to the unintended selection of cells that gained a growth advantage after gene editing, thus masking early off-target toxic events that lead to cell death. The time-dependency of testing for off-target toxicity is exemplified in an experiment that delivered Cas9/Pcsk9-gP in mice followed by on- and off-target effects detection for four days. Off-target mutations were highest after 24 hours and steadily declined for the next few days [9]. Hence, interpreting and extrapolating off-target toxic effects after genome editing using ex vivo multicellular systems will require careful experimental design. Likewise, linking a specific mutation to a cellular biological response in complex multicellular systems would require the establishment of novel experimental approaches to analyze genomic, transcriptomic, proteomic, or metabolomic implications after unintended genome editing events (Table 2).

Table 2:

Proposed metrics to evaluate the effect of genome editors on multicellular systems

	Possible readouts	Considerations
Cell Metabolism	Cell Growth Cell Death	Continuous analysis, easy to set up, colorimetric, fluorometric, quantitative, non-invasive, high throughput, non-cell specific
	Cell-secreted metabolites	Continuous analysis, easy to set up, quantitative, non-invasive, non-cell specific, high throughput (e.g., Luminex)
	Enzymatic Metabolic Activity	Continuous analysis, easy to set up, quantitative, non-invasive, high throughput, cell-type specific
Cell Identity	Cell surface markers	End point analysis, requires optimization, quantitative, high throughput, cell-type specific, (e.gs., Flow cytometry, CyTOF)
Imaging analysis	Bright Field and epifluorescence microscopy	Continuous analysis, easy to set up, can be quantitative by downstream image analysis, medium throughput, non-invasive, fluorescence based on reporter (e.gs., GFP, Tdtomato, etc.). Low resolution
	Second harmonic generation microscopy	Cumbersome to set up, can be quantitative, limited to Collagen I fibers, microtubules, and myosin. Low throughput, high resolution
	Confocal microscopy	Requires some optimization, can be quantitative, limited to commercially available antibodies. Low throughout, high resolution
	Light sheet microscopy	Requires extensive optimization, can be quantitative, limited to commercially available antibodies. Low throughout, high resolution of large tissue structures
Omics	Genomics Epigenomics Transcriptomics Proteomics Glycomics Metabolomics Lipidomics	End point analysis, requires some optimization, generates terabytes of data for downstream analysis, very low throughput, quantitative.

5. Challenges to Investigate Immunogenicity of CRISPR in Humans

While there have been significant advances in engineered genome editors for improved specificity, other areas, such as the activation of the human immunological responses to these novel genome editors, are less developed [56]. The delivery method of choice, Cas protein used, or the secondary structure of the gRNA, all independently can be recognized by the immune system and trigger a cascade of immune reactions with the release of cytokines or chemokines that activate antigen presenting cells to recruit the adaptive immune system [4]. Further, as humans are exposed to the same bacteria from which Cas9 is often derived (S. pyogenes and S. aureus), it is possible to develop a pre-existing immunity against Cas9 [7,57,58]. This preexisting immunity could inactivate Cas9 and negate the therapeutic effects of genome editors in vivo [7,57]. To reduce the immunogenic potential of bacterial-derived WT-Cas9 proteins, it is possible to synthetically alter Cas9 by mutating the precise epitopes that are recognized by T-cells while maintaining its nuclease activity [59]. An alternative approach is to use orthogonal Cas9 proteins sequentially [7] or create a Cas-like protein using human homologous parts, as demonstrated by engineering an epitranscriptomic regulator with comparable results to current CRISPR RNA-targeting systems [60].

The SpCas9 protein can induce a humoral immune reaction in animal models [61], in some cases leading to enlarged lymph nodes and activation of CD4⁺ T cells [62]. However, animal models do not accurately recapitulate safety profiles in humans due to differences in the immune system of human versus mouse, particularly in the major histocompatibility complex antigens [63]. The differences between animals and humans reinforce the need to engineer human microphysiological immune tissue platforms that can be used to investigate the safety of CRISPR-based editors.

An idealized human immunocompetent model for testing genome editors should have multicellular architecture and should be reproducible with controllable experimental parameters to accurately represent the innate and adaptive immune system. Primary human immune organs or tissues, such as, lymph node, lymphatic vessels, spleen, thymus, tonsils, or leukocyte-producing bone marrow platforms could facilitate genome editor testing in a physiologically relevant way [64]. However, ex vivo models of the human adaptive immune system are challenging to create because of their complexity in immune cell maturation and the need of specialized niches (recently reviewed in refs [64,65]). For instance, lymphoid T cells, involved in pathogen detection via antigen-presentation dependency, mature in the thymus, whereas B cells, involved in antibody production for long-term immunity, first mature in the bone marrow then differentiate in secondary lymphoid nodes [64]. As part of the adaptive immune system, both B and T cells are derived from hematopoietic stem cells that reside in the bone marrow. Multicellular systems that recapitulate some aspects of the innate immune system (e.gs., dendritic cells, macrophages) have been a major focus in the engineering of tissue chips. For instance, gut and liver systems have been developed, as these two organs contribute considerably to the overall immunological state of the body and are important in drug discovery [66–69].

Although current multicellular immune competent models were not designed to test CRISPR-Cas genome editors, some of them can be useful to study innate and potentially adaptive immune activation. Liver chip models can mimic the innate immune system as they consist of hepatocytes, endothelial cells, Kupffer cells (resident liver macrophages) and stellate cells, all combined in a microfluidic device to provide oxygen to maintain functional hepatocytes [70,71]. The addition of an inflammatory cue (e.g., lipopolysaccharides (LPS)) causes the release of inflammatory cytokines IL-6 and TNFa [72]. Likewise, advanced models of gut-liver interaction with innate and adaptive immune cells can be achieved by combining dendritic cells, macrophages, Treg, and Th7 cells, in a gut chip model and hepatocytes and Kupffer cells, in the liver chip. These chips allow the study of inflammatory gut-liver crosstalk in healthy and ulcerative colitis gut models [73]. Collectively, these gut and liver immune competent models respond to exogenously added cues (e.g., LPS) and therefore can be adapted to study genome editor effects in tissues during homeostasis and inflammatory processes.

In summary, some multicellular innate immune competent models (e.gs. gut and liver) are well characterized and their reproducibility is well documented, thus they can be easily adapted to model genome editor toxicity. Multicellular models of adaptive immunity are in their infancy and will need to be validated before being combined with genome editor testing. Ideally, interconnected tissue chips of the adaptive and innate immune system, especially from cells derived from the same donor, could provide a powerful platform to test genome editors in a physiologically relevant way.

6. Other Limitations of Current Multicellular Systems

Although current multicellular systems can be adapted to test genome editors, there are still some technical challenges that need to be addressed. For instance, as the number and diversity of cells that constitute a given multicellular system increases, it will be necessary to identify appropriate media formulations that support a diverse population of cells. This is critical to allow autocrine and paracrine signaling loops that occur in vivo. Likewise, as the cellular complexity of the ex vivo model increases, there will be a need to design the multicellular systems with the appropriate ratio of cells that can capture in vivo physiology (scaling), especially when the magnitude of cell-cell crosstalk is important to reach metabolic concentrations of cell-secreted factors for correct paracrine signaling or in vivo-like tissue injury response [32,33].

One of the most pressing challenges will be to build multicellular systems using cells from a single donor because using cells from different donors could result in unexpected cell crosstalk or unintended tissue function that does not accurately represent human physiology. This is vital in multicellular systems that will include cells from the adaptive immune system to mimic local or systemic inflammation states. A possible solution is to derive all the cell types from a single iPSC donor, either by the addition of tissue morphogens or via synthetic biology approaches [11,74]. These cells, however, often only mature to a fetal-like state [33,35], thus do not represent adult tissue, providing an important area of future studies in which synthetic biology and genome engineering approaches could turn iPSC-derived immature cells into adultlike tissue.

In addition to the technical and biological challenges described above, the increase in cellular diversity in multicellular systems decreases the ability to correlate an observed response to a given cell type, secreted metabolite, or autocrine and paracrine signaling, so there will be a need to establish metrics for validation and identification of biological outcomes (Table 2). An approach to gain insights into the biology of these multicellular systems is to perform multi-omics (e.gs. RNAseq, metabolomics) analyses. Although possible with a single multicellular system from a single donor, comparing several donors and experimental conditions often results in enormous amounts of data that needs to be interpreted using systems biology approaches and will need further validation to be used as predictor for in-vitro-to-in-vivo-extrapolation [75]. Nevertheless, the establishment of the Somatic Cell Genome Editing working group by the NIH to bring together experts in different fields to share reagents, protocols, unpublished data, and established guidelines for genome editing are moving the field in the right direction [76].

Figure 1. Overall strategy to evaluate genome editors using ex vivo multicellular systems.

Top panel, CRISPR-Cas genome editors and delivery vector engineering (1), can be validated using in silico approaches (2), that further move into the pipeline to be tested in ex vivo multicellular system that recapitulate some aspect of human physiology (3) to predict off-target effects in vivo and establish models for clinical translation (4-5). Bottom panel describes the opportunities and challenges when developing genome editors at each step in the pipeline. Note to editor. Image should be in color.