Integration of CRISPR/Cas9 with artificial intelligence for improved cancer therapeutics

As our understanding of the underlying genetic and molecular basis of malignancy has rapidly increased through massive tumor genetic profiling, modeling, and characterization, the ever-evolving list of molecular alterations in cells holds great potential for identifying actionable genomic events and treating malignancies. The emergence of gene-editing tools in the last few decades has enabled scientists to manipulate genomic sequences to understand gene function better and develop targeted treatments for inherited and acquired diseases. Although the 1970 discovery of restriction enzymes, the original genome editor, was a breakthrough enabling the recognition of specific nucleotide sequences, the use of restriction enzymes was limited due to their inability to direct targeted DNA cleavage at specific sites [1]. Efforts to improve the accuracy and specificity of restriction endonucleases led to the discovery of mega nucleases such as zinc-finger nucleases (ZFNs) and transcription activator-like effector nucleases (TALENs). These engineered nucleases have facilitated genetic manipulation by inducing targeted double-stranded breaks (DSBs), culminating in the activation of either of the two major cellular DNA repair mechanisms, the non-homologous end joining (NHEJ) or homology-directed repair (HDR) (Fig. 1) [2].

The discovery of the clustered regularly interspaced short palindromic repeats (CRISPR)/CRISPR-associated protein 9 (Cas9) system as a genome-editing toolbox significantly transformed the editing of genomic targets by enabling researchers to manipulate genomic elements efficiently and precisely. Initially reported in the prokaryotic genome as part of an antiviral defense mechanism against bacteriophages, the CRISPR/Cas9 system was later recognized as a revolutionary genome-editing tool enabling insertion, removal, and deletion of existing genes with high precision and specificity [3].

In the last few decades, several genes associated with cancer initiation and progression have been identified using high-throughput screening technologies such as next-generation sequencing (NGS) and whole-genome, -exome, and -transcriptome sequencing [4‐6]. CRISPR/Cas9 has been a tool of choice for studying the function and regulation of specific genes and in high-throughput screening approaches (Fig. 2). However, the data generated from these high-throughput technologies require testing and validation using suitable genetic models to infer drug targets and develop efficient treatments. In this context, CRISPR/Cas9 has been the tool of choice for studying the function and regulation of those genes in valuable genetic models, such as isogenic cells, with the same genetic background. In association with sequencing technology, CRISPR/Cas9 has shown promise not only for testing and validating drug targets but also for identifying functional genes such as tumor suppressors, oncogenes, drug resistance genes, cancer stem cells (CSCs), and cancer metabolism–related genes, thus improving our understanding of cancer initiation and progression, a critical step in developing precision treatments (Fig. 1) [7]. To facilitate the success of clinical therapeutics in the drug discovery process, validation of the drug target is a necessary and crucial step. Early drug target validation enables an increased understanding of the effect of target manipulation on disease biomarkers and disease endpoints and the clinical spectrum of the disease. The identification of functional mutations that confer drug resistance is considered the gold standard for drug target engagement and confirmation. In this regard, the high specificity and the ability of the CRISPR/Cas9 system to efficiently manipulate gene targets allows the selection of ideal therapeutic drug targets, thus simplifying the process of drug target selectivity and validation [8‐10].

As the next challenge, the potential of the CRISPR/Cas9 system to correct these cancer-associated aberrations should be a concerted effort in cancer therapy with improved on-target and reduced off-target effects. This review discusses the applicability of the CRISPR/Cas9 system for genome engineering in cancer research and treatment, along with addressing the various algorithms that have greatly improved the efficacy of the system in clinical therapeutics.

In 1987, Ishino et al. reported homologous sequences of 29 nucleotides arranged as direct repeats with 32 nucleotide spacers in the alkaline phosphatase isozyme conversion gene of Escherichia coli [11]. In 2002, with the advances in DNA sequencing technology, similar repetitive DNA sequences were found in archaea and bacteria by using in silico analysis [12]. These sequences are termed 'clustered regularly interspaced short palindromic repeats (CRISPR) because of their characteristic structural features that include the presence of direct repeats (21–50 bp) interspaced by non-repetitive sequences or spacers beside CRISPR-associated (Cas) genes. CRISPR's unique spacer sequences are homologous with viral or bacteriophage sequences that infect bacteria and archaea, suggesting that they might be part of an adaptive immune system providing immunity against foreign nucleic acids [13‐16].

In a bacterial CRISPR system, the Cas9 nuclease moderates the invading bacteriophage DNA cleavage, which is incorporated between CRISPR repeats as “spacers” that later act as genomic signatures of the pathogen. Upon subsequent bacteriophage invasion, spacers produce CRISPR RNA (crRNA) containing protospacer regions complementary to the foreign DNA. The crRNA hybridizes with a transactivating crRNA (tracrRNA) [17], encoded by the CRISPR system. The resulting hybrid crRNA-tracrRNA is then associated with the Cas9 nuclease, establishing a CRISPR/Cas9 system. The protospacer of the crRNA recognizes its complementary region on the foreign DNA, which is followed by its cleavage (adjacent to motif sequence “NGG”) by Cas9’s nuclease domain [18]. The double-stranded breaks (DSBs) generated by the RNA-guided Cas9 activate the DNA repair machinery via error-prone NHEJ, leading to random insertion, deletion, or mutation in the genome or via template-dependent HDR [19, 20]. In 2011, Sapranauskas et al. described the successful transfer of the CRISPR/Cas9 system from Streptococcus thermophilus to Escherichia coli [21]. These findings were essential to understanding the mechanism of the naturally occurring CRISPR immunity system and laid the foundation for establishing the CRISPR/Cas9 system as a genome-editing tool.

In 2012, the functional application of the CRISPR/Cas9 system was first carried out in vitro, demonstrating the role of crRNA in target sequence recognition and Cas9 protein-mediated DNA cleavage [19, 20]. In their study, Jinek et al. used a single guide RNA (sgRNA) designed by fusing the crRNA and tracrRNA sequences, an established feature of the CRISPR/Cas9 system. In 2013, three pioneering studies engineered a type II bacterial CRISPR/cas9 system to successfully edit genes using custom-designed RNA-guided nuclease activity in mammalian cells [22‐24]. These studies marked the beginning of a paradigm shift in basic and clinical research as the CRISPR/Cas9 system provided researchers with a potent tool for targeting any desired genomic loci. In 2013, Qi et al. developed the catalytically dead dCas9 protein, deficient in endonuclease activity but able to initiate CRISPR interference (CRISPRi) and repress target genes [25]. Systems developed by Meader et al. have also fused the dCas9 with the VP64 transcriptional activation domain capable of initiating CRISPR activation (CRISPRa) to increase the expression of target genes [26].

Furthermore, multiple naturally occurring Cas proteins and their engineered variants, such as Streptococcus pyogenes Cas9 (SpCas9), SpCas9-VRER variant, SpCas9-NG variant, SpCas9 variant SpRY, SaCas9, CjCas9, xCas9 3.7, Cas12a, Cas13a, Cas13d, and dCas13, have been developed with different novel applications and target sequence recognition specificities, expanding the potential applications of the CRISPR/Cas9 technology (Table 1).

Cancer research needs rapid and practical tumor models that can recapitulate multiple molecular events that drive tumor progression in a cell. In this context, the CRISPR/Cas9 system has been shown to assist in developing accessible and feasible in vitro models of mammalian cells that can help identify genes and signaling pathways underlying cancer development and recurrence. In vitro CRISPR/Cas9 screening involves loss-of-function (CRISPR), CRISPR/Cas interference (CRISPRi), and CRISPR/Cas activation (CRISPRa) screens. For example, loss-of-function screens (CRISPR) targeting nearly 18000 genes in melanoma cell lines have been described as assisting in identifying genes resulting in resistance to RAF inhibitors [27]. Similar studies have identified ENL (eleven nineteen leukemia) as a critical domain for leukemic transformation [28]. CRISPRi and CRISPRa screens initially demonstrated by Gilbert et al., however, involve catalytically dead dCas9 to alter gene expression [29]. The CRISPRi uses dCas9 fused with KRAB repressor to repress targeted gene expression transcriptionally; yet, in CRISPRa, dCAS9 is linked to an activator to overexpress target genes. Further fine-tuning of these methods has led to the development of inducible systems that use doxycycline for dCas9-KRAB expression [30]. Variants such as the Cas12a system have facilitated the use of multiple genome editing technologies [31]. Such combinatorial gene knock-out screens can identify more than one target gene, thus providing valuable information on biological pathways, sensitivity, or resistance to drugs in cancer research.

In addition to in vitro screening, the CRISPR/Cas9 system is applicable for in vivo screening as well, involving both an indirect and autochthonous screening approach. In an indirect screen, immortalized cancer cell lines (constitutively expressing the Cas9 nuclease) are transduced with guide RNA libraries and then transplanted into animals for induction of tumor growth and metastasis [32]. Different delivery methods are used for direct in vivo mutagenesis, including lentiviruses and adeno-associated viruses (AAV9). A study by Chen et al. utilized the mouse genome-scale CRISPR/Cas9 knock-out library (mGeCKOa) containing 67,405 single-guide RNAs (sgRNAs) to mutagenize a non-metastatic murine non-small cell lung cancer (NSCLC) cell line and the resulting mutant cell pool was found to generate metastasis when transplanted into the flanks of immunocompromised mice [32]. It was observed that specific loss-of-function mutation accelerates tumor growth and metastasis, and CRISPR/Cas9 serves as an efficient tool for the assessment of phenotypic loss-of-function mutations in vivo [32]. On the other hand, in an autochthonous screen, guide RNA libraries and Cas9 are directly delivered to animal models using adeno-associated viruses (AAVs) to generate tissue-specific cancer models, such as those in the liver, lung, and brain [33, 34]. A study utilized CRISPR/Cas9 genome editing to map functional cancer genome atlas of tumor suppressor gene (TSG) variants mutated in human cancers in an autochthonous mouse cancer model [34]. In the study, Cre-inducible CRISPR/ Cas9 mice livers were pool-mutagenized using AAVs carrying a sgRNA library targeting mutated tumor suppressor genes (mTSGs) [34]. It was shown that immunocompetent mice that received the AAV-mTSG library developed complex autochthonous liver tumors and died within 4 months [34]. To reveal the mutational landscape of the tumors, molecular inversion probe sequencing was performed to generate a direct readout of the Cas9 generated variants [34]. Thus, the study demonstrated the significance of AAV-mediated autochthonous CRISPR screens for mapping a provisional functional cancer genome atlas of tumor suppressors, oncogenes, or any other genetic events associated with tumor evolution in vivo.

Our understanding of the molecular regulatory network in cancer has been further advanced by combining next-generation sequencing platforms with CRISPR/Cas9-based screens. In a standard CRISPR/Cas9 screen, the guide RNA, transcribed by the RNA polymerase III at a U6 promoter, lacks a poly-A tail, so it cannot be funneled into RNAseq analysis, limiting the understanding of molecular mechanisms as a result of genetic perturbation. Novel techniques such as Perturb-seq [35], CRISP-seq [36], and CROP-seq [37] were developed by merging single-cell RNAseq analysis with the CRISPR/Cas9 screen. In CRISPR-seq, a polyadenylated unique guide index and a fluorescent selection marker are included with the gRNA vector module [36]; in CROP-seq, a second gRNA module with a poly-A signal is used for RNAseq [37], and in Perturb-seq, a guide barcode with a poly-A tail is included in the construct.

Furthermore, cancer cells evading immune treatment poses a major therapeutic challenge. Recently, studies by Frankiegh et al. have combined Perturb-seq and epitope-seq (CITE-seq) [38] to investigate the genes involved in immune checkpoint resistance (Table 1). Patient-derived melanoma cells were targeted with 248 intrinsic immune checkpoint inhibitors, and resistance signature genes, single-cell transcriptomics, and 20 cell surface proteins were profiled in more than 200,000 cells. Among several targets, CD58 was identified as an essential factor conferring immune evasion ability [39]. The development of these CRISPR-assisted technologies to study complex cellular circuitry has provided an efficient framework for an in-depth investigation of oncogenic drivers in cancer cells.

Cancer is a highly complex disease where the "one size fits all" module of molecular characterization is insufficient for successful therapeutic intervention, necessitating personalized approaches to treatment for better prognostic outcomes for patients with cancer. The CRISPR/Cas9 system can be tailored to investigate gene functions via genome-wide screens and perform rational drug designing for targeted cancer therapy (Fig. 3). This section emphasizes the role of CRISPR gene-editing technology in characterizing cancer heterogeneity to enable precision medicine. CRISPR technology has been used to identify and validate novel drug targets and biomarkers, understand drug resistance mechanisms, and construct cancer models, which are critical areas for developing targeted therapeutic agents needed for precision cancer treatment.

As the field of precision medicine is rapidly evolving, advancements are being made to develop personalized treatments based on a patient’s risk of disease or predicted response. Novel and emerging personalized medicine approaches, such as genome editing technologies using CRISPR/Cas9, can allow genetic modifications tailored to treat various Mendelian and complex diseases. Most CRISPR/Cas9 studies are performed in pre-clinical settings (in vitro and ex vivo studies). The system’s implementation in clinical settings is challenging owing to editing efficiency and off-target effects [86]. The off-target effects occur when Cas9 binds to and cleaves unintended genomic binding sites, resulting in abnormal gene function [87]. However, some of the latest approaches/strategies, such as cytosine or adenine base editors, ribonucleoprotein delivery, truncated g-RNAs, prime editing, and selection of different Cas variants, can minimize the off-target effects of the CRISPR system [87].

Recently, human induced pluripotent stem cells (hiPSCs) have been extensively used to represent human disease models as they exhibit phenotypes that closely mimic human pathologies [88]. The CRISPR/Cas9 system and hiPSCs could be helpful in drug development and screening, gene therapy/gene editing, and as a therapeutic immune response strategy against viral infections [89, 90]. Patient-specific induced pluripotent stem cells, with high replicative capacity and pathogenic genetic alterations, can be utilized to study the underlying molecular mechanisms in complex genetic disorders. However, this strategy is the genetic manipulation of the pluripotent human cells that prevents the generation of efficient genetically defined disease models [91]. Furthermore, in contrast to ZFNs, which require protein recoding for each new target site, CRISPR/Cas9 can be designed to target any genomic site by simply altering the protospacer sequence of gRNA [92]. Moreover, the CRISPR/Cas9 system can facilitate genome editing at multiple locations by encoding multiple RNA guide sequences into a single CRISPR array [24].

In addition, curative therapies, which are one-time treatments curtailing the symptoms of a disease permanently or semi-permanently, are gaining prominence in precision medicine to treat rare inherited disorders. One example is Maino et al.'s recent study that used the CRISPR/Cas9 system to correct Duchenne muscular dystrophy (DMD) tandem duplication mutation in mouse models [93]. The study used an sg-RNA and a Cas9 system to remove the DMD duplication mutation and restore full-length dystrophin expression in the mice [93].

Recently, the CRISPR system has also been used for editing mouse zygotes [94, 95]. The CRISPR/Cas9 components were shown to be delivered into the zygotes via microinjection or electroporation to produce genetically modified offspring [94]. Although the future use of the system in zygote editing seems promising, ethical concerns regarding the use of CRISPR in human germline editing remain. In this context, few recent studies have revealed unintended genome editing outcomes in CRISPR/Cas9–targeted human preimplantation embryos [96]. The study observed loss of heterozygosity and segmental loss and gain of chromosome 6 in the edited human embryo cells [96]. Yet another recent study evaluated the repair outcomes of a Cas9-induced double-stranded break on the paternal chromosome carrying a frameshift mutation that causes blindness. The most common repair mechanism was microhomology-mediated end joining, which successfully restores the reading frame in the embryos [97]. Although the Cas9 successfully repaired DSBs to some extent, more than half of the DSBs that lead to indels and chromosomal loss remained unrepaired [97].

The first-ever Chinese clinical trial of the CRISPR/Cas9 system involved injecting genetically modified cells into a patient with lung cancer. In this case, the CRISPR/Cas9 system was used to delete the PD-1 gene that downregulates the immune system response [98]. After the first clinical trial, many clinical trials (in vivo and ex vivo) utilizing the CRISPR/Cas9 system to treat human pathologies emerged (reviewed in [86]).

Apart from gene editing, the CRISPR/Cas9 system has been used in antiviral applications. During the ongoing COVID-19 pandemic, rapid and simple point-of-care testing is urgently needed. COVID-19 fits in the field of precision medicine owing to the diverse nature of disease symptoms across individuals. The CRISPR/Cas complex can be used for nucleic acid detection of SARS-CoV-2 as the complex can bind to the target region and cleave the nearby reporter nucleic acid constructs, thus indicating the presence of a viral nucleic acid [99]. Moreover, many studies have used CRISPR-based diagnostic tools to detect SARS-CoV-2 via CRISPR-Cas effector Cas13, a protein that can target RNA and, therefore, can detect SARS-CoV-2, a single-stranded RNA virus [100, 101].

Although CRISPR/Cas9 is an efficient tool for high-throughput screening, large-scale genome-wide screening using the CRISPR system needs much improvement in specificity, off-target effects, and analytical methods. In addition, more efficient delivery systems must be developed to increase the CRISPR/Cas9-system screening efficiency of the immune system genes in vivo. Thus, the future of CRISPR/Cas9 in precision medicine looks promising so long as the ethical concerns arising from the adverse effects of the system are considered and adequately addressed.

Numerous gRNA features or modulators have been identified that affect the cleavage efficiency of gRNA resulting in off-target effects. These include protospacer adjacent motifs (PAMs), gRNA sequence motifs, nucleotide usage in gRNA, position-specific nucleotide composition, secondary structure, and epigenetic features [102]. With the in silico gRNA design accounting for a crucial parameter in a successful gene edit by the CRISPR-Cas9 system, current efforts are now more focused on refining the gRNA design with improved on-target efficacy and minimum off-target effects [103]. As such, recent studies have described various algorithms that may help predict the various on-target and off-target effects of the CRISPR gRNAs, thus improving the respective activity and specificity of the CRISPR-Cas9 system [104]. These prediction tools have immensely assisted in improving the applications and success rates of the CRISPR system. However, the efficacy of gRNA is dependent on the interactions among various factors, including cellular environment, experimental conditions, gRNA, and target sequence, which can be overcome by machine learning (ML) based algorithms. The ML models are trained using existing datasets and can be used to predict the on/off-target effects of testing datasets. The present ML models are based on three methods: (a) regression-based methods [105‐109], (b) classification-based methods [109‐113], and (c) ensemble-based methods [114, 115]. The main differences in different categories of current ML tools are in the features they use and the presentation of the target site in model [102]. Advanced ML has also enabled the deployment of deep learning (DL) methods, such as artificial neural networks (ANNs) for the CRISPR-Cas9 system allowing high-precision target predictions. DL models in the CRISPR/Cas9 system comprise multiple layers of interconnected compute units. The algorithm takes the encoded gRNA-DNA sequence of length 23 in the matrix format as input. The convolutional layer applies various filers of different sizes to the input matrix. The next layer performs batch normalization to the resultant data from the previous layer to boost learning and avert over-fitting. The third pooling layer applies further filtration to the normalized data. The output of the pooling layer passes through different dense layers, and the neurons in these dense layers are fully interconnected. The last dense layer passes the result to the stop layer and predicts whether the input is off-target or on-target. The model architecture of a DL model is represented in (Fig. 5). Artificial Intelligence–based Machine and Deep Learning (MDL) methods are now gaining immense importance in gRNA design and CRISPR applications. The method uses algorithms based on the ever-increasing gene editing datasets reported globally and assists in predicting CRISPR gRNA activity and specificity scores [103]. Moreover, compared to experimental detection tools such as GUIDE-seq [116], HTGTS [117], or IDLV [118], the MDL-based methods are more efficient and cost-effective. A few examples of MDL-based algorithms that have been developed over the past few years to predict CRISPR on-target efficacy include CRISPRater [107], CRISPRScan [119], Azimuth 2.0 [120], TUSCAN [121], DeepCRISPR [122], DeepCas9 [123], WU-CRISPR [124], SgRNAScorer [113], CRISPRpred [125], DeepHF [126], CNN-SVR [127] and C-RNNCrispr [128].

It is pertinent to mention that the efficacy of MDL-based prediction tools is still not fully understood in different cell types and species [103]. Based on variability among various species, studies have now been focused on developing numerous species-specific software, such as fryCRISPR for Drosophila [129] and CRISPRscan for zebrafish [119]. Apart from predicting the on-target efficacy, algorithms have also been described that help predict the crucial off-target effects of the CRISPR-Cas9 system. These include the FlashFry [130], dsNickFury [131], CRISPR [132], DeepCRISPR [122], CNN_std [133], and Elevation [131] algorithms. The results of these and other studies have indicated that the off-target effects of CRISPR are not random and may be avoided by proper design of gRNA sequences [134]. Moreover, studies have shown that truncating the gRNA, especially the 5ʹ-end reduces the possible off-target effects [135].

The gRNA design algorithms, therefore, serve as a crucial milestone that may account for the proper application and efficient development of the CRISPR system shortly. Nevertheless, the existing models and algorithms suffer from drawbacks, including data imbalance and heterogeneity, lack of training datasets, and inefficiency across species, necessitating refinement before the gene editing system can be fully integrated into therapeutics. With the ever-evolving list of newer algorithms, the CRISPR/Cas9 system is expected to improve its increased on-target activity with corresponding minimum off-target effects, a crucial prerequisite in its clinical and therapeutic applications.

The CRISPR/Cas9 system represents a revolutionary approach to gene editing and a potent tool in precision cancer medicine. The CRISPR/Cas9 cell screens and animal models have shown immense potential in discovering drug targets and cancer biomarkers, enabling more precise cancer treatments. The technology has enabled researchers to determine the frequency with which a tumor metastasizes, its origin, and its spread, making it possible to uncover changes in cancer that were otherwise undetected. CRISPR/Cas9 could also be a game-changer in the next generation of immuno-oncology cancer therapy. The precision and efficiency of multiplexed editing with CRISPR/Cas9 may overcome the primary challenges with the current generation of CAR-T therapies. The system has enabled the generation of off-the-shelf (allogeneic) CAR-T cells, providing significant gains over autologous (patient-derived) products. The system can be used to eliminate or insert genes to create new classes of CAR-T products with improved applicability to solid tumors. The integration of CRISPR-based assays with single-cell multi-omics approaches provides an array of applications that can be used to explore gene alterations and tumor heterogeneity. In addition, incorporating spatial transcriptomics with pooled CRISPR libraries can help understand the impact of genetic alterations on tumor microenvironment interactions [136]. Moreover, the application of CRISPR/Cas9 system in larger animal models will allow an efficient simulation of various human diseases, thus enriching the disease model resource bank.

Moreover, MDL-based algorithms have dramatically improved the system's efficacy concerning the reduced off-target effects, a crucial factor in broadening their application in clinical therapeutics. Using species-specific CRISPR algorithms has further enhanced the system's effectiveness across species. Newer CRISPR systems may continue to astound us once further work is carried out, especially in the context of the well-designed clinical trials needed before the technology can be used for cancer treatment. We expect that in the coming decade, advancements in the next-generation gene editing technologies will expand the versatility of the CRISPR/Cas9 system and enhance its applicability in diverse biological systems, thus becoming an indispensable tool to uncover the complexity of different human diseases.