Regulatory Considerations for Clinical Trial Applications with CRISPR-Based Medicinal Products

Introduction

Since first proposed as a new tool in gene targeting and genome-editing applications, ¹ the CRISPR-Cas system has evolved as a burgeoning area of research, bringing to light a large variety of new developments in many scientific areas. The versatility of the system opened up the possibility of modifying the genome easily in a highly specific and efficient manner.

Initial studies analyzing CRISPR-Cas9-mediated therapeutic approaches for human medicine showed potential for correcting incurable genetic diseases and for enhancing cell-based therapies. Indeed, CRISPR-Cas9 quickly progressed to the clinical stage, ² and the list of clinical trials is growing fast. At the end of 2021, more than 40 clinical trials using CRISPR were listed in the ClinicalTrials.gov database, mainly for inherited (blood) disorders, malignancies, and metabolic disorders (Table 1).

There are several modalities of CRISPR-mediated gene editing that can be harnessed for medical applications. For example, the non-homologous end joining (NHEJ) pathway following a CRISPR-mediated DNA cut has been used for generating deletions that precisely disrupt gene expression. NHEJ in combination with a homologous repair template resulting in homology-directed repair (HDR) can be used to introduce corrected genomic sites replacing mutated sequences. Specific Cas9-mediated deletion of certain genomic regions is also possible. Point mutations could be corrected via single-base editing by combining CRISPR-Cas9 with HDR-mediated single-base editing. ³

The CRISPR toolkit is growing at a fast pace. The discovery of Cas9 orthologues that recognize different protospacer adjacent motif sequences provides additional target sites ⁴ and Cas9 versions reduced in size. ⁵ A further Cas endonuclease variant is Cas12a, which generates a cut with a 5′ overhang. This property could be harnessed to introduce DNA sequences in a predefined orientation. ⁶

It is not only DNA that can be targeted by CRISPR technology: RCas9 ⁷ and Cas13a ⁸ endonucleases target and cleave RNAs—a characteristic that could be exploited specifically to reduce the accumulation of a toxic protein. A further variant of Cas9, dCas9, was obtained by destroying its catalytic activity while retaining the DNA-targeting function. This allows sequence-specific control of gene expression by binding protein effectors to dCas9 to achieve CRISPR-mediated RNA interference, ⁹ CRISPR-mediated gene activation, ¹⁰ or epigenome editing by acetylation and methylation of histones and methylation of DNA. ¹¹

A proof-of-principle study demonstrated the capabilities of CRISPR-Cas9-mediated gene editing and gene therapy in two patients with sickle cell disease and β-thalassemia. High levels of allelic editing, resulting in increased fetal hemoglobin and favorable clinical translation (i.e., absence of vaso-occlusive disease or transfusion requirement, respectively), were achieved following ex vivo electroporation of CD34+ hematopoietic stem and progenitor cells with CRISPR-Cas9 targeting the BCL11A erythroid-specific enhancer.^12,13 In 2021, CRISPR-Cas9 in vivo gene editing for transthyretin amyloidosis was successfully applied in patients to reduce accumulation of misfolded transthyretin (TTR) protein in tissues by targeted knockout of TTR. ¹⁴

The medical applicability of CRISPR has expanded, aiming to treat inherited diseases and delving into areas such as immuno-oncology and transplantation. In the first U.S.-based study, T lymphocytes from cancer patients were modified by CRISPR-Cas9 to disrupt three genes (TRAC, TRBC, and PDCD1) with the goal of improving antitumor immunity. ¹⁵ Another early study used CRISPR-Cas9 editing of immune checkpoint genes with the goal of improving the efficacy of T-cell therapy in patients with refractory non-small-cell lung cancer. ¹⁶

Moreover, CRISPR-mediated chimeric antigen receptor (CAR) T-cell editing could be used to achieve universal T cells evading donor human leukocyte antigen–mediated immune rejection ¹⁷ —an aspect that could also be relevant for induced pluripotent stem cell (iPSC)-based therapies ¹⁸ and opening new prospects even in the field of organ transplantation. ¹⁹ A further safety-relevant example of using CRISPR was proposed in iPSCs when a transcriptional link between the suicide gene herpes simplex virus thymidine kinase (HSV-TK) and a cell-division gene (CDK1) was established; by inserting TK into the 3′ untranslated region of CDK1, a suicide system active in dividing cells was achieved. ²⁰

Yet, despite these initial successes and promising developments, the road to CRISPR therapies is not without challenges, with the novelty and characteristics of the technology bringing along uncertainties that warrant special attention. It has been shown that significant on-target mutagenesis, such as large deletions and more complex genomic rearrangements at the targeted sites may occur following CRISPR-Cas9 editing in mitotically active cells. ²¹

The occurrence of potential off-target effects was also recognized early, and strategies were developed to detect these accurately. ²² Yet, strategies ²³ to mitigate off-target effects still need to be refined. These strategies could involve both the design of optimal guide RNA sequences and the engineering of Cas enzymes. Editing efficiency is an important aspect for clinical applications and is dependent on the design of the system, the method of delivery, or the ex vivo or in vivo setting of the gene-editing step. ²⁴

On the other hand, editing efficiency may be also hampered by the influence of immunogenicity. The guide RNAs may trigger innate immune responses, ²⁵ and the immune response against Cas9^26,27 could also negatively influence treatment outcomes. Finally, the double-strand breaks (DSBs) induced by Cas9 could be toxic to cells and may induce apoptosis; moreover, as this phenomenon is P53/TP53-dependent, it could hamper CRISPR applications in iPSCs, which are prone to accumulate P53 mutations. ²⁸

While acknowledging the opportunities this technology brings for the development of medicinal products, timely development of regulatory oversight and adaptation of regulatory requirements is necessary to guarantee the quality, safety, and efficacy of products based on CRISPR technology. In the European Union (EU), a clinical trial authorization is the responsibility of the national competent authorities of the member states in which the trial is going to be performed.

In this respect, taking into account the novelty of the technology and the limited experience with clinical evaluation of medicinal products manufactured with the CRISPR-Cas system, the expectations on data required for clinical trial approval might be different in the different EU member states. Nevertheless, the basic scientific approach toward regulation of gene editing investigational medicinal products (IMPs) seems to be comparable between member states and between the EU and the United States. The Paul-Ehrlich-Institut (PEI) is the competent authority in Germany for the approval of clinical trial applications (CTAs) for vaccines and biomedicines, including cell and gene therapies, and is part of the EU regulatory framework for EU-wide scientific advice procedures and marketing authorization assessments for these product classes.

This paper highlights some general aspects relevant for advancing CRISPR-Cas product development toward CTAs. Additionally, it focuses on the current regulatory framework applicable for CRISPR-based developments, including key aspects of manufacture and quality control, as well as nonclinical and clinical development requirements relevant for CTAs in Germany.

General Regulatory Framework

Currently, genome-editing products face some uncertainty regarding regulatory classification. Products fulfilling the advanced therapy medicinal product (ATMP) definition (EMA/319248/2020) are evaluated under the respective ATMP regulation (Council Regulation [EC] 1394/2007 of 13 November 2007 on ATMPs and amending Directive 2001/83/EC and Regulation [EC] 726/2004).

There is an ongoing scientific debate regarding potential CRISPR-based products delivered in forms other than a recombinant nucleic acid (e.g., the CRISPR system could be delivered as a ribonucleoprotein—i.e., a recombinant protein and synthetic guide ribonucleic acid). Such a product would not fit the description of recombination under the European definition of gene therapy medicinal products and thus would not fall under the said regulation. ²⁹

In addition, for some genome-editing products, the genetically modified organisms (GMO) regulation and requirements are considered to apply (Directive 2001/18/EC on the deliberate release into the environment of GMOs; Directive 2009/41/EC on the contained use of genetically modified microorganisms). This is of relevance for CTAs in EU member states, where often AMTP GMO submissions and CTAs are the responsibility of different competent authorities, particularly with the Clinical Trial Regulation coming into force. ³⁰ Changes to the regulatory framework to clarify the definition and classification of genome-editing products as well as international harmonization is one key recommendation of the EU-IN Horizon Scanning Report on genome editing (EMA/319248/2020).

Table 2 contains a detailed list of guidelines, directives, and regulations cited in the text.

Quality Requirements

General quality requirements for ATMPs based on gene editing are comprehensively described in the guideline on gene therapy medicinal products (EMA/CAT/80183/2014) and the guideline on medicinal products containing genetically modified cells (EMA/CAT/GTWP/671639/2008 Rev. 1—corr). This guidance also applies to plasmid DNA or viral vector particles used in CRISPR-Cas approaches based on the HDR pathway. When genome editing is conducted ex vivo, the edited cells are defined as the medicinal product, and the genome-editing tools are considered as starting material. For an in vivo approach, the genome-editing tools are delivered systemically, for example complexed in lipid vesicles.^31,32

The on-target efficiency in the target cells is one of the key quality parameters in a genome-editing approach. Different analytical methods are available to measure on-target efficiency, which is a most relevant parameter for drug product batch release. Common methods employed are next-generation sequencing (NSG), SMRT sequencing, fluorescent gene addition, restriction fragment length polymorphism gel assay, or traffic light reporter. ³³

The chosen method should in any case be qualified and should reliably quantify the on-target frequency in the respective target cell. Ideally, the selected analytical method allows valid comparison throughout the process development, for example when moving from a model cell system to the target cell system such as from a surrogate cell line to autologous patient cells, or when changing the delivery system or other parameters in the manufacturing process of the genome-editing medicinal product. Depending on the editing strategy used, the assay should be able to measure quantifiably small NHEJ or HDR. Some assays enable simultaneous measurement of edited NHEJ and gene targeting by HDR, which may be useful when using both editing strategies in parallel.

The majority of the available NSG methods show a high sensitivity (<2%) and allow high-throughput sequencing, which is often an important criterion for suitability. As this method will be crucial during product development and batch release, method validation for the most critical parameters (limit of detection, limit of quantification, working range, and linearity) would already be expected at a very early time point of clinical development.

Analysis of on-target efficiency is an important readout for product potency. However, additional analyses that directly or indirectly measure the intended biological function and correlate it to genome-editing efficiency are ideally included early in product development as a batch-release criteria and are expected at later time points in product development. Such assay would for example determine the increase or decrease in expression and/or enzymatic activity of the protein in question. Ideally, correlations between the on-target editing frequency and the change of protein expression and/or biological activity can be demonstrated, thereby allowing setting of suitable acceptance limits for on-target efficacy at batch release.

Besides on-targeting, the used CRISPR-Cas-system needs to be analyzed for potential genome-wide off-target editing. The principles of the strategy to address this issue are described in the following sections addressing computational algorithms and nonclinical considerations.

Potential off-targeting will mostly be driven by the single-guide RNA (sgRNA) used. The chemical synthesis of a sgRNA (consisting of the CRISPR and the tracer RNA) can be especially challenging with respect to accuracy and efficiency due to their large size compared to small interfering RNAs. Increasing sgRNA purity and removal of aberrant RNA fragments should be achieved as early as possible during development. For in vivo CRISPR-Cas approaches, high sgRNA purity and adequate control methods are considered indispensable for starting clinical development. This includes sequence confirmation of the sgRNA batch by including a qualified sequencing assay, which should also be part of the sgRNA release tests.

From a quality point of view, it is important that the sgRNA batch used in the off-target analysis is representative for the batch that will be used in the manufacturing of the intended investigational medicinal product. At an early time point of development, when the number of produced sgRNA batches might be low, which is often accompanied by limited amounts of characterization data, or in case the manufacturing of the sgRNA is adapted, it may be reasonable to repeat the confirmatory study of the nominated off-target sites with the newly manufactured sgRNA batch.

Besides the sgRNA, the Cas protein^23,34 and delivery method of the CRISPR-Cas tools might affect the likelihood of off-target effects and should be thoughtfully selected for the manufacturing of the genome-editing medicinal product. Considering that permanent presence of genome-editing tools will make genomes more susceptible to off-target cleavage, ³⁵ the choice of a delivery method should only allow transient presence of the genome-editing tools in the target cells for such products for human use.

For ex vivo gene editing, the target cells are often collected from a donor/patient and thus will be of an autologous or allogeneic source. Importantly, the target cells will then normally consist of a heterogeneous cell population, harboring slightly different qualitative characteristics. The cellular composition of the medicinal product will affect clinical efficacy or safety and should therefore be well characterized. For example, in a medicinal product based on hematopoietic stem cells, the presence of multipotent progenitor cells will be important in view of a successful transplantation and long-term effect. Similarly, the composition of memory and effector gene-edited CAR-T cells will affect the killing dynamic and the long-term effect of the CAR-T cell product.

With respect to genome editing, the distribution of the on-target efficiency in the different cell subpopulations should be analyzed. It needs to be shown whether the on-target editing is distributed equally in the cell population or is primarily present in a certain subpopulation. In case the on-target editing is preferably present in a certain cell subpopulation, the impact on product safety and efficacy needs to be discussed, ideally based on applicants' own nonclinical data.

For some products, the definition of drug product batch release criteria for a certain cell subpopulation may be helpful in order to control product quality or when certain requirements are important in view of the clinical application. In case the manufacturing process for cell harvesting, selection, editing, and cultivation is changed, the chance that this will also affect the composition of the cellular subpopulations in the product is high. Therefore, the analysis of cell subpopulations should be part of a subsequent quality comparability study.

Besides the adaption of the manufacturing process, additional manufacturing sites may be established. In such cases, a comprehensive quality comparability study, analyzing the potential impact of the respective changes on product quality or confirming the successful transfer of the manufacturing process to another site, would always be required. Such comparability studies are often challenging, as the impact of the donor material is extensive, hampering the comparability assessment, for example due to the wide variability for on-targeting frequency routinely seen for different donors.

In order to circumvent the donor effect, the developer may conduct split runs in which the same donor material is used in parallel in the old and the new manufacturing site. Such studies are usually conducted with material derived from healthy donors, which would also allow testing a sufficient number of runs confirming the results.

As for every gene therapy medicinal product, the amount of product- and process-related impurities needs to be characterized. Depending on the amount introduced into the manufacturing process and/or present in the final product as well as on the individual risk of each impurity in regard to the safety of the patient and the efficacy of the product, a batch release specification for this impurity should be set. The aim is to keep impurity levels constantly low whenever possible.

For in vitro edited cells, the sgRNA and the Cas protein (including mRNA coding for the Cas protein) are potential process-related impurities that may be present in the final medicinal product. The transfer of CRISPR-Cas components with ex vivo edited cells into the patient should therefore be avoided whenever possible to reduce the risk of involuntary uptake by other cells or tissues. It is important to generate characterization data describing the respective impurity levels and their time- or manufacturing process-dependent depletion from the final product.

The residual level of editing-tool impurities will depend on the amount of these starting materials used in the manufacturing and the subsequent processing of the gene-edited cells. Here, media changes, washing steps, and the cultivation period may reduce the levels. Additionally, data providing information on the half-life of the sgRNA, Cas mRNA, or Cas protein will support a risk assessment on their presence at the time of the intended administration into humans.

In case of an in vivo approach, the potential on- and off-target editing in nontarget tissues and cells will need to be addressed carefully with respective nonclinical data. These data will have to allow a comprehensive risk assessment for on- and off-target editing in potentially affected nontarget tissues and cells. Depending for example on the residual amounts of process-related impurities, consistency in clearance during manufacturing, and the risks associated with these impurities for the safety of the patients and the efficacy of the product, impurity levels need to addressed at release by characterization data or at least by scientific discussion of their impact.

Considerations on Computational Algorithms for In Silico Target Site Analysis

CRISPR-Cas may edit sites other than the intended site. Potential off-target editing needs to be thoroughly analyzed for genome-editing products prior to their clinical application, as these may have implications related to the safety of the product. The overall strategy on how genome-wide off-target editing should be addressed for clinical applications is described in the nonclinical section.

Researchers validate CRISPR-Cas using targeted sequencing approaches to identify which sites may have undergone off-target editing. It is not feasible to interrogate all 6.4 billion positions of the human genome with high sensitivity, considering that off-targeting may occur only in a small fraction of cells. Thus, researchers have developed experimental and in silico methods to narrow down the number of sites that need validation in the target cells.

The use of in silico methods for off-target site-editing prediction follows the same three basic stages: (1) nominate potential off-target sites using a mismatch threshold; (2) calculate a score that estimates how likely nominated sites are to undergo editing or directly predict the editing frequencies; and (3) experimentally validate off-target sites that are likely to undergo editing. For many of the early tools, steps 1 and 2 are the same in that the score is the number of mismatches.

In the first stage, potential off-target sites are nominated using a basic mismatch filter that finds sites that have less than a given maximum number of mismatches. The size of the genome implies that a full genome search for off-target sites is computationally expensive. Therefore, tools use computational algorithms and heuristics to improve the search speed for off-target sites. Non-heuristic algorithms tend to work very well for exact matches but are often inflexible when attempting inexact matching. Heuristic algorithms sacrifice accuracy for speed and may miss real off-target editing sites, ³⁶ potentially compromising the safety of a patient.

Using mismatches alone to determine which sites to validate experimentally is potentially problematic, as the number of nominated off-target sites increases exponentially with increasing maximum number of mismatches. In the second stage, many tools attempt to counteract this by implementing scores that attempt to estimate how likely a nominated off-target site is to be edited or even by directly estimate the editing frequency. Thresholds placed on the scores or editing frequencies can then be used to reduce the number of sites to be validated.

Early tools used only the number of mismatches to estimate how likely an off-target site was to be edited. However, the number of mismatches proved to be an incomplete predictor of off-target editing,^37–39 leading to the development of tools that used hypothesis-driven scores that also incorporated the position and identity of the mismatched nucleotides. The latest tools have data-driven scores calculated from large data sets using machine-learning methods.

Once off-target sites have been predicted, targeted sequencing methods can be used to identify which sites have undergone modification. The deeper each site can be sequenced, the more likely it is that rare editing events can be detected. Thus, score-based tools pose a tempting solution for filtering out sites that do not undergo off-target editing, as fewer sites mean more sensitive detection thresholds.

Identifying the correct use of in silico tools for the prediction of off-target editing poses several challenges that need to be considered when developing a medicinal product. One such problem is which tools are safe and reliable to use. There are dozens of tools, each with different underlying algorithms for nominating and scoring off-target sites and each providing different advantages and disadvantages. For example, tools that use heuristic algorithms may miss crucial off-target sites and seem to have a significant disadvantage, but the extent to which they do so is not thoroughly documented.

Another problem is finding the appropriate number of mismatches to allow between the guide RNA and DNA. Studies show that confirmed off-target sites can have up to five mismatches, ⁴⁰ including single-nucleotide bulges. ⁴¹ However, studies also show that sites with more mismatches have lower rates of editing, ⁴⁰ implying that allowing as many as five mismatches may be excessive. Finding the right balance between increasing the number of allowed mismatches and reducing the number of sites to validate experimentally remains a challenge.

Validating the algorithms themselves also remains an open issue, considering that those are used for the development of a medicinal product for human use. To validate whether a sequence alignment algorithm really does find all sites with a given number of mismatches, another similar algorithm needs to be developed that also needs to be validated in a similar manner, potentially making true validation impossible to resolve. The scores of score-based algorithms and predictions of editing frequency can be validated experimentally. However, experimental validation is necessarily cell-type specific and may give different results if the score-based algorithms were trained on different cell types.

Increasingly, machine learning methods are being used to predict the likelihood and frequency of off-target editing. Existing tools use machine learning algorithms ranging from support vector machines to convolutional neural networks. Such machine learning algorithms produce models that are notoriously “black boxes,” meaning that the reasoning behind the predictions they make is not easily interpretable for human applications and may harbor hidden biases that hinder the generalizability of the model.

Natural variation is also a factor that needs to be considered when predicting off-target editing sites. Even the difference of a single-nucleotide polymorphism (SNP) could reduce the number of mismatches of a site that would normally be filtered out to fall within the acceptable threshold. In silico methods are in the unique position of being able to detect such sites, as they can make predictions on hypothetical genomes. Currently, very few tools have been developed that incorporate natural variations into off-target site editing prediction. ⁴²

Nonclinical Aspects

The nonclinical considerations on CRISPR-Cas-mediated genome editing mainly focus on ex vivo NHEJ approaches used to generate a genetic knockout of a gene of interest. For more general nonclinical requirements for genetically modified cells, guidance is provided in the EMA guideline “Guideline on quality, nonclinical and clinical aspects of medicinal products containing genetically modified cells” (EMA/CAT/GTWP/671639/2008 Rev. 1—corr), which also contains a short paragraph on the nonclinical safety evaluation of cell-based products derived from genome editing.

For CRISPR-Cas-based therapeutic developments, questions concerning the efficiency and specificity of the genome-editing approach dominate the nonclinical program. While both parameters are already considered during the selection process of the sgRNA, the efficiency and specificity of a chosen sgRNA needs to be evaluated in detail before its use in a clinical trial. While the specificity of the CRIPSR-Cas system can be further improved by either truncating or modifying the chosen sgRNA or by the use of an improved Cas variant, ²³ the nonclinical on- and off-target genome-editing evaluation must be conducted with the final CRISPR-Cas components that will be used in the clinical trial.

In addition, certain process-related impurities such as truncated forms of sgRNA, which are not characterized for every sgRNA batch, may have an effect on the efficiency and/or specificity of the system. Therefore, the material used for the nonclinical on- and off-target evaluation should be representative of the clinical material also with regard to the purity and impurities of the CRISPR-Cas components.

The efficiency of the CRISPR-Cas system is determined by the percentage of on-target genome editing in the cell type(s) that will be isolated, modified, and subsequently readministered to the patients. Besides determining the on-target genome-editing efficiency in these cells, cell viability and function, as well as differentiation potential (for genome editing in stem and progenitor cells), also need to be investigated in order to exclude negative effects on cellular physiology.

Similarly, the off-target potential of the CRISPR-Cas system also needs to be addressed in these cells. If the on- and off-target evaluation is performed in cells isolated from healthy donors, the underlying disease in the treated patients potentially affecting the on- and off-target potential of the CRISPR-Cas system will have to be carefully considered. For example, gene expression might differ in cells from diseased versus healthy donors, and the concomitant change of chromatin accessibility might translate to a different likelihood of off-target genome editing within these genes.

A particular challenge for the use of ex vivo genome-editing cells is to determine the therapeutically relevant level of on-target genome-editing efficiency. Such nonclinical studies are often hampered by the lack of an animal model that mimics the human disease to be treated or by the lack of a model that allows engraftment of the genome edited human cells. While such an animal model might only be available in rare situations, a surrogate product might be considered. In most cases, such a surrogate product would consist of cells derived from the animal model and an animal-specific sgRNA, as it is unlikely that the target sequence in human and the animal model would be completely homologous.

However, this raises questions about the comparability of the modified cells with regard to their isolation, manufacturing, and on-target genome-editing efficiencies. Thus, translating nonclinical data obtained with a surrogate product may be challenging. It is therefore important to consider any existing knowledge that allows a valid prediction on the genome-editing efficiency needed for a potential clinical benefit. For example, in some cases, naturally occurring genetic variants in the human population may provide assistance in predefining the genome-editing efficiency needed in a given indication.

While increasing the concentration of the CRISPR-Cas system components used to edit the human cells will certainly increase the on-target genome-editing efficiency, it may also increase the frequency of off-target genome editing. Thus, an optimal on-target genome-editing efficiency should be established that is expected to result in a clinical benefit without introducing off-targeting at a critical site.

Off-target genome editing occurring at critical sites might result in transformation of the cells and subsequent tumor formation and needs to be avoided. A thorough evaluation of the off-target potential of the CRISPR-Cas system is therefore required. Usually, this evaluation is divided into a discovery phase, which identifies potential off target sites, and a validation phase, which evaluates for each identified potential off-target site, irrespective of whether off-target genome editing does occur under the conditions used in the clinical trial.

In the discovery phase, several analyses are combined, including (1) in silico analyses based on the target genome sequence and a given number of possible mismatches and gaps to the target sequence, and (2) unbiased genome-wide cell-based and/or biochemical assays, in which the DSBs introduced into the genome in the presence of the CRISPR-Cas components are marked before the concerned sites are being sequenced. In a genome-wide unbiased cell-based assay, such as GUIDE-Seq, ²² the cell type(s) used should be carefully selected and include the cell type to be modified in the clinical trial.

Otherwise, differences in chromatin accessibility might misrepresent the off-target potential for the target cell type(s). Biochemical assays, such as SITE-Seq, ⁴³ have the advantage that the DSBs are not influenced by the chromatin structure. Consequently, a larger number of potential off-target sites are usually being identified by a biochemical assay compared to a cell-based assay. However, with an increased number of potential off-target sites identified by the biochemical assay, the number of potential false-positive sites may also be increased.

Thus, the ideal combination of assays used in the discovery phase may vary, depending on the genome-editing approach, as suggested in a recent publication by Chaudhari et al. ⁴⁴ From a regulatory point of view, it is necessary to justify the assays chosen, as well as the key parameters set in the analyses, such as the number of mismatches considered in the in silico analysis or the cell types used in the cell-based assay.

In the validation phase, all previously identified potential off-target sites within coding and noncoding sequences should be included in the analysis independently of the assay that led to their identification. Again, the cell type(s) used as well as the concentration of the CRISPR-Cas components are crucial for the validation of the potential off-target sites. In addition, the sequencing depth and the sensitivity of the assay used in the validation phase are similarly important. In case one or more off-target sites should be confirmed in the validation phase, any risks associated with off-target genome editing at these sites needs to be carefully evaluated. This may also include a nonclinical evaluation of the consequences of genome editing at the confirmed off-target site(s).

A particular challenge for addressing the off-target potential is certainly the SNPs present in humans. While SNPs within the targeted DNA sequence of the sgRNA can be avoided, the large number of SNPs present in human genomes makes it impossible to validate all of them experimentally when addressing the off-target potential of the CRISPR-Cas components. However, SNPs with a high frequency within a certain population could be included in the in silico analyses. In addition, increasing the number of donors included in the cell-based and biochemical assays used for the discovery phase as well as in the assay used for the validation phase might help to reduce the likelihood of critical off-target sites being missed due to the presence of SNPs.

Finally, DNA structural variants, including larger insertions and deletions at the on-target site, as well as translocations and chromosomal rearrangements between DSBs introduced at the on- and off-target sites, need to be addressed. Again, the cell type(s) and the methodology used, including the sensitivity limits of the assays, should be justified. While DNA structural variants resulting from the natural DSB repair mechanisms of the cell cannot be completely avoided, the percentage of DNA structural variants introduced by the genome-editing approach should remain low. Moreover, none of the translocation or chromosomal rearrangements identified should be associated with a known oncogenic risk.

For in vivo NHEJ applications, where the CRISPR-Cas components are administered directly to the patients, evaluation of the specificity of the genome editing is conducted in a similar way as described for ex vivo approaches. However, identifying the appropriate cell types to be included in the analyses is more challenging.

In addition to the target cell type(s), other cells with a certain off-target genome-editing potential should be included in both discovery and validation phases when evaluating off-target sites. Cell types with a certain off-target genome-editing potential are best identified by investigating on-target genome editing in different tissues and organs in a relevant animal model. Ideally, the CRISPR-Cas components mostly reach the target organ when administered in vivo, but cells in nontargeted organs and tissues may become genome edited as well.

Depending on the chosen route of administration, analysis of the on-target genome-editing efficiency in a relevant animal model will provide information on the other cells and tissues reached and modified by CRISPR-Cas components. Cells and tissues with a certain on-target genome-editing efficiency may subsequently be included in the analysis of off-target genome editing and the analysis of DNA structural variants. It is obvious that both the dose of the CRISPR-Cas components administered and the in vivo clearance rate of the CRISPR-Cas components are key safety parameters, as a high dose level and/or a low clearance rate of the CRISPR-Cas components will potentially increase the off-target genome editing of a CRISPR-Cas-based in vivo investigational product.

Clinical Considerations

General considerations on the clinical development (efficacy and safety of the product, including assessment of pharmacokinetics/biodistribution, pharmacodynamics, and identification of a recommended dose) apply, if not precluded by product-specific characteristics.

For clinical trials in the pediatric population, both the ICH E11 (R1) guideline on clinical investigation of medicinal products in the pediatric population (EMA/CPMP/ICH/2711/1999) as well as potential national peculiarities, for example as laid down in the German Medicinal Products Act, need to be adhered to. In addition to disease-specific guidance, including choice of endpoints, the respective guidance for clinical trials in small populations and the points to consider on marketing authorizations based on a single pivotal trial should be considered early on, if applicable (CHMP/EWP/83561/2005, CPMP/EWP/2330/99).

The particularities of the clinical development program will be impacted equally by the targeted disease, for example rarity of the condition, the mode of delivery of the genome-editing tool, and the novelty of the technology itself, resulting in uncertainties for long-term safety and efficacy (EMA/319248/2020). CRISPR-based genome editing has the potential to be utilized in a broad variety of diseases (EMA/319248/2020), including rare ones ⁴⁵ such as hereditary transthyretin amyloidosis (NCT04601051).

In orphan diseases, the number of patients for enrolment in clinical trials is limited. This may result in a clinical development program where distinct Phase I and Phase II trials as well as randomized controlled confirmatory Phase III trials are challenging. As these circumstances would be evident prior to initiation of the first clinical trial(s), the clinical development program could be adapted and designed accordingly.

First-in-human clinical trials should adhere to the guideline on strategies to identify and mitigate risks for first-in-human and early clinical trials with investigational medicinal products (EMEA/CHMP/SWP/28367/07 Rev. 1). For therapeutic confirmatory trials, the randomized controlled trial design should also be preferred in rare disease settings. In such cases, comparison to best standard of care should be considered as an option. When justified, a single-arm trial design may be acceptable if the treatment effect is attributable to the product and if additional (external) data describing the natural course of disease are available to contextualize the results.

The disease to be treated and the treatment approach also influence the clinical considerations of the minimum extent of genome editing required to achieve the expected clinical outcome. For CD34+ genome-editing approaches, this includes, but is not limited to, considerations on the required minimal number of edited CD34+ cells and on whether monoallelic editing would suffice to achieve the expected outcome. If the latter is not the case, the lower limit of cells with biallelic modification defined per release criteria needs to be adequately justified. For in vivo CRISPR approaches, the adequacy of the proposed dose to achieve the extent of genome editing needs to be satisfactorily substantiated (e.g., by nonclinical data).

The benefit/risk of the proposed clinical trial is also influenced by the target cells and by the mode of delivery. Ex vivo genome editing may allow for tighter control of the actual genome editing and thus lessen the safety concerns ⁴⁶ but is technically restricted to certain cell types ⁴⁷ and conditions. Depending on the target tissues, in vivo approaches may include systemic or local delivery, facing additional challenges including degradation, opsonization, or clearance. ⁴⁷ In addition, adverse effects of on-target editing in nontarget cells and tissues may occur. These will extend the uncertainties in addition to the identified off-target potential.

Limitations to extrapolate the results from nonclinical studies to humans leads to uncertainties in the judgment of short- and long-term adverse drug reactions and toxicities. Sponsors need to address the identified and potential risks in the benefit/risk assessment and to implement respective risk-mitigation measures. This refers to, but is not limited to, the potential of CRISPR-based IMPs for on- and off-target effects, immunogenicity, tumorigenicity, and germline transmission. Given the limited clinical experience available to date, close (long-term) monitoring seems warranted for both ex vivo and in vivo approaches. A minimum of 15 years' follow-up of trial participants exposed to CRISPR-based genome editing with close monitoring for (hematologic) malignancies is mandatory.

The permanent alteration of the target genome by CRISPR is associated not only with concerns regarding long-term safety but also with the expectation of long-lasting or even curative effects (EMA/319248/2020). Independent of the respective genome-editing candidate, whether classified as ATMP or not, the underlying principles of generating long-term efficacy and safety data as laid down in the guideline on safety and efficacy follow-up and risk management of ATMPs apply (EMEA/149995/2008 rev.1 draft). Thus, the clinical development program should also be able to confirm long-term durability of the intended genome editing and, preferably, efficacy as well.

Concluding Remarks

As exemplified in this paper, advancing CRISPR-based medicinal products toward clinical trials presents several specific challenges. These can be of general nature related to all ATMPs or specific to CRISPR characteristics. Therefore, early regulatory interaction to align scientific and regulatory considerations and expectations is highly encouraged.