A kinetic model predicts SpCas9 activity, improves off-target classification, and reveals the physical basis of targeting fidelity

AlPaCas: allele-specific CRISPR gene editing through a protospacer-adjacent-motif (PAM) approach

Gene therapy of dominantly inherited genetic diseases requires either the selective disruption of the mutant allele or the editing of the specific mutation. The CRISPR-Cas system holds great potential for the genetic correction of single nucleotide variants (SNVs), including dominant mutations. However, distinguishing between single-nucleotide variations in a pathogenic genomic context remains challenging. The presence of a PAM in the disease-causing allele can guide its precise targeting, preserving the functionality of the wild-type allele. The AlPaCas (Aligning Patients to Cas) webserver is an automated pipeline for sequence-based identification and structural analysis of SNV-derived PAMs that satisfy this demand. When provided with a gene/SNV input, AlPaCas can: (i) identify SNV-derived PAMs; (ii) provide a list of available Cas enzymes recognizing the SNV (s); (iii) propose mutational Cas-engineering to enhance the selectivity towards the SNV-derived PAM. With its ability to identify allele-specific genetic variants that can be targeted using already available or engineered Cas enzymes, AlPaCas is at the forefront of advancements in genome editing. AlPaCas is open to all users without a login requirement and is freely available at https://schubert.bio.uniroma1.it/alpacas.

Machine learning-inferred and energy landscape-guided analyses reveal kinetic determinants of CRISPR/Cas9 gene editing

The CRISPR/Cas nucleases system is widely considered the most important tool in genome engineering. However, current methods for predicting on/off-target effects and designing guide RNA (gRNA) rely on purely data-driven approaches or focus solely on the system's thermal equilibrium properties. Nonetheless, experimental evidence suggests that the process is kinetically controlled rather than being in equilibrium. In this study, we utilized a vast amount of available data and combined random forest, a supervised ensemble learning algorithm, and free energy landscape analysis to investigate the kinetic pathways of R-loop formation in the CRISPR/Cas9 system and the intricate molecular interactions between DNA and the Cas9 RuvC and HNH domains. The study revealed (a) a novel three-state kinetic mechanism, (b) the unfolding of the activation state of the R-loop being the most crucial kinetic determinant and the key predictor for on- and off-target cleavage efficiencies, and (c) the nucleotides from positions +13 to +16 being the kinetically critical nucleotides. The results provide a biophysical rationale for the design of a kinetic strategy for enhancing CRISPR/Cas9 gene editing accuracy and efficiency.

Anti-CRISPR Proteins and Their Application to Control CRISPR Effectors in Mammalian Systems

CRISPR-Cas effectors are powerful tools for genome and transcriptome targeting and editing. Naturally, these protein-RNA complexes are part of the microbial innate immune system, which emerged from the evolutionary arms race between microbes and phages. This coevolution has also given rise to so-called anti-CRISPR (Acr) proteins that counteract the CRISPR-Cas adaptive immunity. Acrs constitutively block cognate CRISPR-Cas effectors, e.g., by interfering with guide RNA binding, target DNA/RNA recognition, or target cleavage. In addition to their important role in microbiology and evolution, Acrs have recently gained particular attention for being useful tools and switches to regulate or fine-tune the activity of CRISPR-Cas effectors. Due to their commonly small size, high inhibition potency, and structural and mechanistic versatility, Acrs offer a wide range of potential applications for controlling CRISPR effectors in heterologous systems, including mammalian cells.Here, we review the diverse applications of Acrs in mammalian cells and organisms and discuss the underlying engineering strategies. These applications include (i) persistent blockage of CRISPR-Cas function to create write-protected cells, (ii) reduction of CRISPR-Cas off-target editing, (iii) focusing CRISPR-Cas activity to specific cell types and tissues, (iv) spatiotemporal control of CRISPR effectors based on engineered, opto-, or chemogenetic Acrs, and (v) the use of Acrs for selective binding and detection of CRISPR-Cas effectors in complex samples. We will also highlight potential future applications of Acrs in a biomedical context and point out present challenges that need to be overcome on the way.

Interpretable neural architecture search and transfer learning for understanding CRISPR-Cas9 off-target enzymatic reactions

Finely tuned enzymatic pathways control cellular processes, and their dysregulation can lead to disease. Developing predictive and interpretable models for these pathways is challenging because of the complexity of the pathways and of the cellular and genomic contexts. Here we introduce Elektrum, a deep learning framework that addresses these challenges with data-driven and biophysically interpretable models for determining the kinetics of biochemical systems. First, it uses in vitro kinetic assays to rapidly hypothesize an ensemble of high-quality kinetically interpretable neural networks (KINNs) that predict reaction rates. It then employs a transfer learning step, where the KINNs are inserted as intermediary layers into deeper convolutional neural networks, fine-tuning the predictions for reaction-dependent in vivo outcomes. We apply Elektrum to predict CRISPR-Cas9 off-target editing probabilities and demonstrate that Elektrum achieves improved performance, regularizes neural network architectures and maintains physical interpretability.

Genome-wide CRISPR off-target prediction and optimization using RNA-DNA interaction fingerprints

The powerful CRISPR genome editing system is hindered by its off-target effects, and existing computational tools achieved limited performance in genome-wide off-target prediction due to the lack of deep understanding of the CRISPR molecular mechanism. In this study, we propose to incorporate molecular dynamics (MD) simulations in the computational analysis of CRISPR system, and present CRISOT, an integrated tool suite containing four related modules, i.e., CRISOT-FP, CRISOT-Score, CRISOT-Spec, CRISORT-Opti for RNA-DNA molecular interaction fingerprint generation, genome-wide CRISPR off-target prediction, sgRNA specificity evaluation and sgRNA optimization of Cas9 system respectively. Our comprehensive computational and experimental tests reveal that CRISOT outperforms existing tools with extensive in silico validations and proof-of-concept experimental validations. In addition, CRISOT shows potential in accurately predicting off-target effects of the base editors and prime editors, indicating that the derived RNA-DNA molecular interaction fingerprint captures the underlying mechanisms of RNA-DNA interaction among distinct CRISPR systems. Collectively, CRISOT provides an efficient and generalizable framework for genome-wide CRISPR off-target prediction, evaluation and sgRNA optimization for improved targeting specificity in CRISPR genome editing.

Predicting CRISPR-Cas12a guide efficiency for targeting using machine learning

Genome editing through the development of CRISPR (Clustered Regularly Interspaced Short Palindromic Repeat)-Cas technology has revolutionized many fields in biology. Beyond Cas9 nucleases, Cas12a (formerly Cpf1) has emerged as a promising alternative to Cas9 for editing AT-rich genomes. Despite the promises, guide RNA efficiency prediction through computational tools search still lacks accuracy. Through a computational meta-analysis, here we report that Cas12a target and off-target cleavage behavior are a factor of nucleotide bias combined with nucleotide mismatches relative to the protospacer adjacent motif (PAM) site. These features helped to train a Random Forest machine learning model to improve the accuracy by at least 15% over existing algorithms to predict guide RNA efficiency for the Cas12a enzyme. Despite the progresses, our report underscores the need for more representative datasets and further benchmarking to reliably and accurately predict guide RNA efficiency and off-target effects for Cas12a enzymes.

Guide-specific loss of efficiency and off-target reduction with Cas9 variants

High-fidelity clustered regularly interspaced palindromic repeats (CRISPR)-associated protein 9 (Cas9) variants have been developed to reduce the off-target effects of CRISPR systems at a cost of efficiency loss. To systematically evaluate the efficiency and off-target tolerance of Cas9 variants in complex with different single guide RNAs (sgRNAs), we applied high-throughput viability screens and a synthetic paired sgRNA-target system to assess thousands of sgRNAs in combination with two high-fidelity Cas9 variants HiFi and LZ3. Comparing these variants against wild-type SpCas9, we found that ∼20% of sgRNAs are associated with a significant loss of efficiency when complexed with either HiFi or LZ3. The loss of efficiency is dependent on the sequence context in the seed region of sgRNAs, as well as at positions 15-18 in the non-seed region that interacts with the REC3 domain of Cas9, suggesting that the variant-specific mutations in the REC3 domain account for the loss of efficiency. We also observed various degrees of sequence-dependent off-target reduction when different sgRNAs are used in combination with the variants. Given these observations, we developed GuideVar, a transfer learning-based computational framework for the prediction of on-target efficiency and off-target effects with high-fidelity variants. GuideVar facilitates the prioritization of sgRNAs in the applications with HiFi and LZ3, as demonstrated by the improvement of signal-to-noise ratios in high-throughput viability screens using these high-fidelity variants.

Negative DNA supercoiling induces genome-wide Cas9 off-target activity

CRISPR-Cas9 is a powerful gene-editing technology; however, off-target activity remains an important consideration for therapeutic applications. We have previously shown that force-stretching DNA induces off-target activity and hypothesized that distortions of the DNA topology in vivo, such as negative DNA supercoiling, could reduce Cas9 specificity. Using single-molecule optical-tweezers, we demonstrate that negative supercoiling λ-DNA induces sequence-specific Cas9 off-target binding at multiple sites, even at low forces. Using an adapted CIRCLE-seq approach, we detect over 10,000 negative-supercoiling-induced Cas9 off-target double-strand breaks genome-wide caused by increased mismatch tolerance. We further demonstrate in vivo that directed local DNA distortion increases off-target activity in cells and that induced off-target events can be detected during Cas9 genome editing. These data demonstrate that Cas9 off-target activity is regulated by DNA topology in vitro and in vivo, suggesting that cellular processes, such as transcription and replication, could induce off-target activity at previously overlooked sites.

Interpretable neural architecture search and transfer learning for understanding CRISPR/Cas9 off-target enzymatic reactions

Finely-tuned enzymatic pathways control cellular processes, and their dysregulation can lead to disease. Creating predictive and interpretable models for these pathways is challenging because of the complexity of the pathways and of the cellular and genomic contexts. Here we introduce Elektrum, a deep learning framework which addresses these challenges with data-driven and biophysically interpretable models for determining the kinetics of biochemical systems. First, it uses in vitro kinetic assays to rapidly hypothesize an ensemble of high-quality Kinetically Interpretable Neural Networks (KINNs) that predict reaction rates. It then employs a novel transfer learning step, where the KINNs are inserted as intermediary layers into deeper convolutional neural networks, fine-tuning the predictions for reaction-dependent in vivo outcomes. Elektrum makes effective use of the limited, but clean in vitro data and the complex, yet plentiful in vivo data that captures cellular context. We apply Elektrum to predict CRISPR-Cas9 off-target editing probabilities and demonstrate that Elektrum achieves state-of-the-art performance, regularizes neural network architectures, and maintains physical interpretability.

Titrating CD47 by mismatch CRISPR-interference reveals incomplete repression can eliminate IgG-opsonized tumors but limits induction of antitumor IgG

Phagocytic elimination of solid tumors by innate immune cells seems attractive for immunotherapy, particularly because of the possibilities for acquired immunity. However, the approach remains challenging, with blockade of the macrophage checkpoint CD47 working in immunodeficient mice and against highly immunogenic tumors but not in the clinic where tumors are poorly immunogenic. Even when mouse tumors of poorly immunogenic B16F10 melanoma are opsonized to drive engulfment with a suitable monoclonal antibody (mAb), anti-CD47 blockade remains insufficient. Using both in vitro immuno-tumoroids and in vivo mouse models, we show with CRISPR interference (CRISPRi) that a relatively uniform minimum repression of CD47 by 80% is needed for phagocytosis to dominate net growth when combined with an otherwise ineffective mAb (anti-Tyrp1). Heterogeneity enriches for CD47-high cells, but mice that eliminate tumors generate prophagocytic IgGs that increase in titer with CD47 repression and with tumor accumulation of macrophages, although deeper repression does not improve survival. Given well-known limitations of antibody permeation into solid tumors, our studies clarify benchmarks for CD47 disruption that should be more clinically feasible and safer but just as effective as complete ablation. Additionally, safe but ineffective opsonization in human melanoma trials suggests that combinations with deep repression of CD47 could prove effective and initiate durable immunity.

Engineered Staphylococcus auricularis Cas9 with high-fidelity

CRISPR-Cas9 is a versatile gene editing tool with a broad application of basic research and clinical therapeutics. However, the potential impact caused by off-target effects remains a critical bottleneck. The small Cas9 ortholog from Staphylococcus auricularis (SauriCas9) was identified, which recognizes a 5'-NNGG-3' protospacer adjacent motif (PAM), exhibiting high activity for genome editing. Recently, we also reported enhanced-fidelity Staphylococcus aureus Cas9 (efSaCas9), which harbors a single mutation N260D. Protein sequence alignment revealed that SauriCas9 has 62.4% sequence identity with SaCas9. Because SauriCas9 is more flexible in recognizing the target sequence with PAM of 5'-NNGG-3' than SaCas9 of 5'-NNGRRT-3' PAM, we sought to test whether key mutation(N260D) or adjacent residue mutation in efSaCas9 can be appliable to SauriCas9. With this concept, two engineered SauriCas9 variants (SauriCas9-HF1, harboring the N269D mutation; SauriCas9-HF2, harboring the D270N mutation) dramatically improved targeting specificity by targeted deep sequencing and GUIDE-seq. At certain sites, reduced off-target effects (approximately 61.6- and 111.9-fold improvements) of SauriCas9-HF2 compared with wild-type SauriCas9 were observed. Overall, two identified SauriCas9 variants (SauriCas9-HF1 and SauriCas9-HF2) expand the utility of the CRISPR toolkit for research and therapeutic applications.

The energy landscape for R-loop formation by the CRISPR-Cas Cascade complex

Clustered regularly interspaced short palindromic repeats (CRISPR) sequences and CRISPR-associated (Cas) genes comprise CIRSPR-Cas effector complexes, which have revolutionized gene editing with their ability to target specific genomic loci using CRISPR RNA (crRNA) complementarity. Recognition of double-stranded DNA targets proceeds via DNA unwinding and base pairing between crRNA and the DNA target strand, forming an R-loop structure. Full R-loop extension is a prerequisite for subsequent DNA cleavage. However, the recognition of unintended sequences with multiple mismatches has limited therapeutic applications and is still poorly understood on a mechanistic level. Here we set up ultrafast DNA unwinding experiments on the basis of plasmonic DNA origami nanorotors to study R-loop formation by the Cascade effector complex in real time, close to base-pair resolution. We resolve a weak global downhill bias of the forming R-loop, followed by a steep uphill bias for the final base pairs. We also show that the energy landscape is modulated by base flips and mismatches. These findings suggest that Cascade-mediated R-loop formation occurs on short timescales in submillisecond single base-pair steps, but on longer timescales in six base-pair intermediate steps, in agreement with the structural periodicity of the crRNA-DNA hybrid.

Guide-specific loss of efficiency and off-target reduction with Cas9 variants

High-fidelity Cas9 variants have been developed to reduce the off-target effects of CRISPR systems at a cost of efficiency loss. To systematically evaluate the efficiency and off-target tolerance of Cas9 variants in complex with different single guide RNAs (sgRNAs), we applied high-throughput viability screens and a synthetic paired sgRNA-target system to assess thousands of sgRNAs in combination with two high-fidelity Cas9 variants HiFi and LZ3. Comparing these variants against WT SpCas9, we found that ~20% of sgRNAs are associated with a significant loss of efficiency when complexed with either HiFi or LZ3. The loss of efficiency is dependent on the sequence context in the seed region of sgRNAs, as well as at positions 15-18 in the non-seed region that interacts with the REC3 domain of Cas9, suggesting that the variant-specific mutations in REC3 domain account for the loss of efficiency. We also observed various degrees of sequencedependent off-target reduction when different sgRNAs are used in combination with the variants. Given these observations, we developed GuideVar, a transfer-learning-based computational framework for the prediction of on-target efficiency and off-target effect with high-fidelity variants. GuideVar facilitates the prioritization of sgRNAs in the applications with HiFi and LZ3, as demonstrated by the improvement of signal-to-noise ratios in high-throughput viability screens using these high-fidelity variants.

High-throughput biochemistry in RNA sequence space: predicting structure and function

RNAs are central to fundamental biological processes in all known organisms. The set of possible intramolecular interactions of RNA nucleotides defines the range of alternative structural conformations of a specific RNA that can coexist, and these structures enable functional catalytic properties of RNAs and/or their productive intermolecular interactions with other RNAs or proteins. However, the immense combinatorial space of potential RNA sequences has precluded predictive mapping between RNA sequence and molecular structure and function. Recent advances in high-throughput approaches in vitro have enabled quantitative thermodynamic and kinetic measurements of RNA-RNA and RNA-protein interactions, across hundreds of thousands of sequence variations. In this Review, we explore these techniques, how they can be used to understand RNA function and how they might form the foundations of an accurate model to predict the structure and function of an RNA directly from its nucleotide sequence. The experimental techniques and modelling frameworks discussed here are also highly relevant for the sampling of sequence-structure-function space of DNAs and proteins.

A quantitative model for the dynamics of target recognition and off-target rejection by the CRISPR-Cas Cascade complex

CRISPR-Cas effector complexes recognise nucleic acid targets by base pairing with their crRNA which enables easy re-programming of the target specificity in rapidly emerging genome engineering applications. However, undesired recognition of off-targets, that are only partially complementary to the crRNA, occurs frequently and represents a severe limitation of the technique. Off-targeting lacks comprehensive quantitative understanding and prediction. Here, we present a detailed analysis of the target recognition dynamics by the Cascade surveillance complex on a set of mismatched DNA targets using single-molecule supercoiling experiments. We demonstrate that the observed dynamics can be quantitatively modelled as a random walk over the length of the crRNA-DNA hybrid using a minimal set of parameters. The model accurately describes the recognition of targets with single and double mutations providing an important basis for quantitative off-target predictions. Importantly the model intrinsically accounts for observed bias regarding the position and the proximity between mutations and reveals that the seed length for the initiation of target recognition is controlled by DNA supercoiling rather than the Cascade structure.

CRISPR/Cas9 On- and Off-Target Activity Using Correlative Force and Fluorescence Single-Molecule Microscopy

The discovery of CRISPR/Cas9 as an easily programmable endonuclease heralds a new era of genetic manipulation. With this comes the prospect of novel gene therapy approaches, and the potential to cure previously untreatable genetic diseases. However, reports of spurious off-target editing by CRISPR/Cas9 pose a significant hurdle to realizing this potential. A deeper understanding of the factors that affect Cas9 specificity is vital for development of safe and efficient therapeutics. Here, we describe methods for the use of optical tweezers combined with confocal fluorescence microscopy and microfluidics for the analysis of on- and off-target activity of Cas9 activity.