key: cord-0800718-o5rt8m41
authors: Kumar, Aditya; Palmer, Nathan; Miyasaki, Katelyn; Finburgh, Emma; Xiang, Yichen; Portell, Andrew; Dailamy, Amir; Suhardjo, Amanda; Chew, Wei Leong; Kwon, Ester J.; Mali, Prashant
title: Extensive in vitro and in vivo protein translation via in situ circularized RNAs
date: 2022-02-11
journal: bioRxiv
DOI: 10.1101/2022.02.11.480072
sha: 0d3d5b68eb2e4f5bf7a5ac040fe19dbe11390f3f
doc_id: 800718
cord_uid: o5rt8m41

RNAs are a powerful therapeutic class. However their inherent transience impacts their activity both as an interacting moiety as well as a template. Circularization of RNA has been demonstrated as a means to improve persistence, however simple and scalable approaches to achieve this are lacking. Utilizing autocatalytic RNA circularization, here we engineer in situ circularized RNAs (icRNAs). This approach enables icRNA delivery as simple linear RNA that is circularized upon delivery into the cell, thus making them compatible with routine synthesis, purification, and delivery formulations. We confirmed extensive protein translation from icRNAs both in vitro and in vivo and explored their utility in three contexts: first, we delivered the SARS-CoV-2 Omicron spike protein in vivo as icRNAs and showed corresponding induction of humoral immune responses; second, we demonstrated robust genome targeting via zinc finger nucleases delivered as icRNAs; and third, to enable compatibility between persistence of expression and immunogenicity, we developed a novel long range multiplexed (LORAX) protein engineering methodology to screen progressively deimmunized Cas9 proteins, and demonstrated efficient genome and epigenome targeting via their delivery as icRNAs. We anticipate this highly simple and scalable icRNA methodology could have broad utility in basic science and therapeutic applications.

RNAs have emerged as a powerful therapeutic class. However their typically short half-life impacts their activity both as an interacting moiety (such as siRNA), as well as a template (such as mRNAs). Towards this, RNA stability has been modulated using a host of approaches, including engineering untranslated regions, incorporation of cap analogs, nucleoside modifications, and codon optimality (1) (2) (3) (4) (5) . More recently, novel circularization strategies, which remove free ends necessary for exonuclease-mediated degradation thereby rendering RNAs resistant to most mechanisms of turnover, have emerged as a particularly promising methodology (6) (7) (8) (9) (10) (11) (12) (13) . However, simple and scalable approaches to achieve efficient in vitro production and purification of circular RNAs are lacking, thus limiting their broader application in research and translational settings.

Utilizing work on autocatalytic RNA circularization by Litke and colleagues (14) , we recently engineered circular guide RNAs for programmable RNA editing (15) . The primary approach for generating these was via delivery of encoding DNA molecules where the guide RNAs were expressed using pol-III promoters, and thereby were both generated and circularized in cells. However, we also made the observation that in vitro transcribed RNAs delivered in linear form could successfully circularize in situ in cells upon entry and were similarly functional as guide RNAs. Motivated by the extreme simplicity of this latter approach, and its compatibility with routine in vitro synthesis and purification processes, we explored if this framework could also be used to generate circular messenger RNAs. Indeed, we show below that thus engineered in situ circularized RNAs (icRNAs) enable extensive protein translation, and we demonstrate their versatility via a range of in vitro and in vivo applications spanning from RNA vaccines to genome and epigenome targeting.

Common to all these applications via icRNA delivery is the critical importance of their immune system interactions. Although for some applications, such as vaccines, robust immune responses to the therapeutic are desirable, for other applications such as genome and epigenome targeting, immune responses can instead inhibit therapeutic effect. Inducing immune responses through RNA delivery has been extensively 2 researched in vaccine development and proven through the success of COVID vaccines based on this technology (16) (17) (18) (19) . However, despite substantial engineering efforts, deimmunization remains a tougher problem to crack. Thus, to facilitate compatibility between persistence of expression and immunogenicity especially when delivering non-human payloads via icRNAs, we also concurrently developed a long-range multiplexed (LORAX) protein engineering methodology based on high-throughput screening of combinatorially deimmunized protein variants. We applied it to identify a Cas9 variant with seven key HLA-restricted epitopes simultaneously immunosilenced after a single round of screening, and showed that icRNA mediated delivery of the same enabled robust genome targeting.

To engineer icRNAs, we generate in vitro transcribed linear RNAs that bear a twister ribozyme flanked internal ribosome entry site (IRES) (20, 21) coupled to a messenger RNA of interest (Figure 1a) . Once transcribed, the flanking twister ribozymes rapidly self-cleave, enabling hybridization of the complementary ligation stems to one another.

Upon delivery into cells, these linear RNAs are then circularized in situ by the ubiquitous RNA ligase RtcB. To evaluate this approach, we first assayed green fluorescent protein (GFP) translation via flow cytometry in the icRNA format in vitro in HEK293T cells. As a side-by-side comparison, we also engineered linear in situ circularization defective RNAs (icdRNAs) by utilizing catalytically inactive mutants of the twister ribozymes.

Specifically, HEK293Ts were transfected with circular GFP icRNA or linear icdRNA and RNA was isolated at 6 hours, 1 day, 2 days and 3 days after transfection. We observed similar amounts of GFP RNA at 6 hours (Figure 1b, left panel) , confirming that approximately equal quantities of icRNA and icdRNA were delivered to cells. However, GFP RNA with functional circularization was significantly higher at days 1, 2, and 3 than icdRNA, indicating improved RNA persistence via circularization (Figure 1b, middle   panel) . This improved RNA persistence also correlated with increased GFP translation after 3 days (Figure 1b, right panel) . To confirm icRNAs were covalently circularized in cells upon delivery in vitro, we performed via RT-PCR by designing outward facing primers that selectively amplified only the circularized RNA molecules. Indeed, we only 3 observed a PCR product for icRNAs, confirming circularization (Figure 1c) . Next, to extend these results in vivo, lipid nanoparticles (LNPs) (22, 23) containing circular icRNA and linear icdRNA were generated. No difference in LNP size was observed between icRNA and icdRNA (Supplementary Figure 1a) . 10 µg of LNPs were retro-orbitally injected into C57BL/6 mice, livers were isolated 3 and 7 days later, and RNA was extracted. RT-PCR confirmed circularization of icRNA in vivo, with persistence extending to at least 7 days (Supplementary Figure 1b) . Finally, we screened a panel of IRES sequences, ligation stems, and 3' untranslated regions (UTRs) to optimize protein translation (Supplementary Figure 1c ) (24) (25) (26) (27) (28) . These studies demonstrated the ability to tune protein translation from icRNAs over an order of magnitude. Among the examined constructs, the medium yielding UTR version 2 was chosen for all subsequent studies.

We initially explored icRNA application across two distinct therapeutic transgene delivery contexts: one, to enable immunization via proteins delivered in the icRNA format, and two, to enable genome targeting via delivery of proteins.

Towards the former, we assessed the production of IgG binding antibodies against SARS-CoV-2 Omicron variant spike protein in BALB/c mice via ELISA. icRNAs and icdRNAs bearing the Omicron spike (K986P, V987P) protein were generated (29) , encapsulated in LNPs, and delivered via a single intramuscular injection at a dose of 2µg icRNA or icdRNA/mouse. We confirmed robust induction of anti-spike IgG in the sera of animals receiving icRNA at 3 weeks post injection compared to other groups (Figure 1d) .

Towards the latter, we generated zinc finger nuclease (ZFN) icRNAs and icdRNAs targeting the GFP and CCR5 genes (30, 31) . Being a fully protein based genome engineering toolset we anticipated ZFNs would be particularly suited for this mode of delivery, and indeed observed robust genome editing via icRNAs compared to icdRNAs upon their delivery into HEK293T cells (Figure 1e) .

Spurred by these results, we next explored if icRNAs could be used to deliver the CRISPR-Cas9 systems. We conjectured that the prolonged expression via icRNAs 4 could facilitate genome and especially epigenome targeting. However, this same feature of persistence could also aggravate immune responses in therapeutic settings as CRISPR systems are derived from prokaryotes (32) (33) (34) . Thus, to enable compatibility between persistence of expression and immunogenicity, we sought first to develop a methodology to screen progressively deimmunized SpCas9 proteins by combinatorially mutating particularly immunogenic epitopes (35) .

While variant library screening has proven to be an effective approach to protein engineering, applying it to deimmunization faces three important technical challenges.

One, the need to mutate multiple sites simultaneously across the full length of the protein; two, reading out the associated combinatorial mutations scattered across large (>1kb) regions of the protein via typical short read sequencing platforms; and three, engineering fully degenerate combinatorial libraries which can very quickly balloon to unmanageable numbers of variants. To overcome these challenges we developed several methodological innovations which, taken together, comprise a novel long range multiplexed (LORAX) protein engineering platform capable of screening millions of combinatorial variants simultaneously with mutations spread across the full length of arbitrarily large proteins (Figure 2a) .

Towards library design, in order to narrow down the vast mutational space associated with combinatorial libraries, we utilize an approach guided by evolution and natural variation (36, 37) . As deimmunizing protein engineering seeks to alter the amino acid sequence of a protein without disrupting functionality, it is extremely useful to narrow down mutations to those less likely to result in non-functional variants. To identify these mutants we generated large alignments of Cas9 orthologs from publicly available data to identify low-frequency SNPs that have been observed in natural environments. Such variants are likely to have limited effect on protein function, as highly deleterious alleles would tend to be quickly selected out of natural populations (if Cas9 activity is under purifying selection) and therefore not appear in sequencing data (38) . To further subset these candidate mutations, we evaluated for immunogenicity in silico using the netMHC epitope prediction software (39, 40) , in order to determine to what degree the candidate mutations are likely to result in the deimmunization of the most immunogenic epitopes in 5 which they appear. This is a critical step as many mutations may have little effect on overall immunogenicity. Screening for decreased peptide-MHC class I binding filters out amino acid substitutions which are likely immune-neutral, substantially increasing the likelihood of functional hits with enough epitope variation to evade immune induction (41, 42) .

Next, to enable readout, we applied long-read nanopore sequencing to measure the results of the screens of our combinatorial libraries. This circumvents the limit of short target regions and obviates the need for barcodes altogether by single-molecule sequencing of the entire target gene, enabling library design strategies which can explore any region of the protein in combination with any other region without any complicated cloning procedures required to facilitate barcoding. To date, the adoption of nanopore sequencing has been limited by its high error rate, around 95% accuracy per DNA base (43) , as compared to established short read techniques which are multiple orders of magnitude more accurate. To address this challenge, we designed our libraries such that each variant we engineered would have multiple nucleotide changes for each single target amino acid change, effectively increasing the sensitivity of nanopore based readouts with increasing numbers of nucleotide changes per library member. The large majority of amino acid substitutions are amenable to a library design paradigm in which each substitution is encoded by two, rather than one, nucleotide changes, due to the degeneracy of the genetic code and the highly permissive third "wobble" position of codons.

The scale of engineering which would be required to generate an effectively deimmunized Cas9 is not fully understood, as combinatorial deimmunization efforts at the scale of proteins thousands of amino acids long have not yet been possible. Therefore, to roughly estimate these parameters we developed an immunogenicity scoring metric which takes into account all epitopes across a protein and the known diversity of MHC variants in a species weighted by population frequency to generate a single combined score representing the average immunogenicity of a full-length protein as a function of each of its immunogenic epitopes (44). Formally, this score is calculated as:

where I x = Immunogenicity score of protein x, i = epitopes, j = HLA alleles, = allelê specific standardization coefficient, w j = HLA allele weights, k ij = predicted binding affinity of epitope i to allele j, and y = protein specific scaling factor. We then predicted the overall effect of mutating the top epitopes in several Cas9 orthologs (Supplementary Figure 2a) . As might be expected, this analysis suggests that single-epitope strategies are woefully inadequate to deimmunize a whole protein for multiple HLA types, and also that there are diminishing returns as more and more epitopes are deimmunized. Our analysis suggests that it may require on the order of tens of deimmunized epitopes to make a significant impact on overall, population-wide protein immunogenicity. The scale of engineering demanded by these immunological facts has previously been intractable, but by applying LORAX we conjectured one could now make substantial steps, several mutations at a time, through the mutational landscape of the Cas9 protein.

Specifically, applying the procedure above, we designed a library of Cas9 variants Sequencing revealed that the library was significantly shifted in the mutation density distribution, suggesting that the majority of the library with large (>4) numbers of mutations resulted in non-functional proteins which were unable to survive the screen.

Meanwhile, wild-type, single, and double mutants were generally enriched as these proteins proved more likely to retain functionality and pass through the screen (Supplementary Figure 2c) . Additionally, the two independent replicates of the screen showed strong correlation (R 2 = 0.925) providing further evidence of robustness ( Figure   2c ). We also analyzed the change in overall frequency of mutations in the pre-and post-screen libraries to see if a pattern of mutation effects could be inferred. Although the wild-type allele was enriched at every site in the post-screen sequences, nearly every site retained a significant fraction of mutated alleles, suggesting that the 8 mutations, at least individually, are fairly well-tolerated and do not disrupt Cas9 functionality (Supplementary Figure 2d) .

In order to select hits for downstream validation and analysis, we devised a method for differentiating high-support hits likely to be real from noise-driven false positive hits. To do this we hypothesized that the fitness landscape of the screen mutants is likely to be smooth, i.e. variants that contain similar mutations are more likely to have similar fitnesses in terms of editing efficiency compared to randomly selected pairs (47) . We confirmed this by computing a predicted screen score for each variant based on a weighted regression of its nearest neighbors in the screen. This metric correlates well with the actual screen scores and approaches the screen scores even more closely as read coverage increases. This provides good evidence that the fitness landscape is indeed somewhat smooth (Supplementary Figure 3a) . Next, we reasoned that because the fitness landscape is smooth, real hits should reside in broad fitness peaks which include many neighbors that also show high screen scores, whereas hits that are less supported by near neighbors are more likely to be spurious as they represent non-smooth fitness peaks. Formalizing this logic, we performed a network analysis to differentiate noise-driven hits from bona fide hits by looking at the degree of connectivity with other hits (Figure 2d ).

Applying these analyses to the screen output led us to select and construct 20 variants Figure 3b) . Variants highly connected to neighbors were capable of editing, whereas those not connected were non-functional, validating the network-based approach we used to select hits as enriching for truly functional

sequences. Among the screen hits was the L614G mutation first identified by Ferdosi and colleagues (14) as a functional Cas9 variant with a critical immunodominant epitope 9 de-immunized (V1). This concordance with previous work provided further confidence in our screening method. Interestingly, we discovered another deimmunizing mutation within the same epitope, L622Q, which similarly retains Cas9 functionality, but appears to be more epistatically permissive, as many of our multi-mutation hits combine this mutation with other deimmunized epitopes. From these multi-mutation hits we chose V4, which demonstrated high editing capability while still bearing simultaneous mutations across seven distinct epitopes, as well as family members V3, a variant bearing two mutations, and V5, a variant bearing the seven changes from V4 plus one additional mutation. We then further evaluated the efficacy of these mutants gene were transfected into HEK293T (51) . Excitingly, we observed both robust genome and epigenome targeting via the icRNA delivery format (Figure 2f,g) .

Utilizing autocatalytic RNA circularization we engineered in situ circularized RNAs (icRNAs). This extremely simple approach enables icRNA delivery as simple linear RNA, thus making them compatible with routine laboratory synthesis, purification, and delivery formulations. We confirmed extensive protein translation and persistence from icRNAs both in vitro and in vivo, and confirmed their versatility and activity in applications spanning from RNA vaccines to genome and epigenome targeting. Notably, the icRNA strategy allowed for generation and delivery of large constructs, such as CRISPRoff, which would be more cumbersome to deploy via lentiviral and adeno-associated virus (AAVs) due to packaging limits (50) .

Concurrently, to enable compatibility between persistence of expression and immunogenicity, we also developed the LORAX protein engineering platform that can be applied iteratively to tackle particularly challenging multiplexed protein engineering tasks by exploring huge swaths of combinatorial mutation space unapproachable using previous techniques. We demonstrated the power of this technique by creating a Cas9 variant with seven simultaneously deimmunized epitopes which still retains robust functionality in a single round of screening. This opens up a critical door in applying gene editing to long-persistence therapeutic modalities such as AAV or icRNA delivery.

Furthermore, while this methodology is particularly suited to the unique challenges of protein deimmunization, it is also applicable to any potential protein engineering goal, so long as there exists an appropriate screening procedure to select for the desired functionality.

While icRNAs are a versatile platform with broad application, we anticipate three major avenues for further engineering their efficacy: one, comprehensive screens of nucleobase modifications, IRESs and UTRs to systematically boost protein translation (52) (53) (54) (55) (56) ; two, further ligation stem engineering could enable greater in situ cyclized fractions of the icRNAs, which in turn would positively impact the yields and durability of protein translation; and three, the impact of icRNA delivery on the innate immune response and approaches to modulate the same will need investigation (9, 57, 58) . 11 Similarly, the versatility of the LORAX platform comes with a set of limitations and tradeoffs that must be managed to leverage its utility. Naturally, library design is of critical importance. Here we have leveraged several features such as Cas9 evolutionary diversity, MHC-binding predictions, HLA allele frequencies, and calculated immunogenicity scores to generate a useful library of variants to test. Other approaches may bring in more sources of information from places like protein structure (59) Looking ahead, in addition to its core utility in applications entailing transgene delivery, we anticipate that icRNAs will be particularly useful in scenarios where a longer duration pulse of protein production is required. These include, for instance, epigenome engineering and cellular reprogramming, as well as transient healing and rejuvenation applications. Taken together, we anticipate the highly simple and scalable icRNA methodology could have broad utility in basic science and therapeutic applications. 12 data generated at the UC San Diego IGM Genomics Center utilizing an Illumina

NovaSeq 6000 that was purchased with funding from a National Institutes of Health SIG grant (S10 OD026929). Some schematics were created using BioRender.

Authors have filed patents based on this work. P.M. is a scientific co-founder of Shape Therapeutics, Navega Therapeutics, Diagon Therapeutics, Boundless Biosciences, and Engine Biosciences. The terms of these arrangements have been reviewed and approved by the University of California, San Diego in accordance with its conflict of interest policies. The remaining authors declare no competing interests.

All key reagents used in the study will be made available via Addgene. In vitro transcription: DNA templates for generating desired RNA products were created by PCR amplification from plasmids or gBlock gene fragments (IDT) and purified using a PCR purification kit (Qiagen). Plasmids were then generated with these templates containing a T7 promoter followed by 5' ribozyme sequence, a 5' ligation sequence, an IRES sequence linked to the product of interest, a 3' UTR sequence, a 3' ligation sequence, a 3' ribozyme sequence, and lastly a poly-T tail to terminate transcription.

Linearized plasmids were used as templates and RNA products were then produced using the HiScribe T7 RNA Synthesis Kit (NEB) per manufacturer's protocol.

In vitro persistence experiments: To assess persistence of circular icRNA, HEK293T cells were transfected with circular icRNA GFP or linear icdRNA and RNA was isolated 6 hours, one day, two days, and three days after transfection. qPCR was performed to assess the amount of GFP RNA and RT-PCR was performed to confirm circular RNA persistence in cells receiving icRNA.

Flow cytometry experiments: GFP intensity, defined as the median intensity of the cell population, was quantified after transfection using a BD LSRFortessa Cell Analyzer. 25 Lipid nanoparticle formulations:

(6Z,9Z,28Z,31Z)-heptatriaconta-6,9,28,31-tetraen-19-yl-4-(dimethylamino) butanoate (DLin-MC3-DMA)  was  purchased  from  BioFine  International  Inc. 1,2-distearoyl-sn-glycero-3-phosphocholine (DSPC) and Cas9 alignment and mutation selection: Naturally occurring variation in Cas9 sequence space was explored by aligning BLAST hits of the SpCas9 amino acid sequence. This set was then pruned by removing truncated, duplicated, or engineered sequences, and those sequences whose origin could not be determined. At specified immunogenic epitopes and key anchor residues, top alternative amino acids were obtained using frequency in the alignment weighted by overall sequence identity to the wild type SpCas9 sequence, such that commonly occurring amino acid substitutions appearing in sequences highly similar to the wild-type were prioritized for further analysis and potential inclusion in the LORAX library.

HLA frequency estimation and binding predictions: HLA-binding predictions were carried out using netMHC4.1 or netMHCpan3.1. Global HLA allele frequencies were estimated from data at allelefrequencies.net as follows. Data was divided into 11 geographical regions. Allele frequencies for each of those regions were estimated from all available data from populations therein. These regional frequencies were then averaged weighted by global population contribution. Alleles with greater than 0.001% frequency in the global population, or those with greater than 0.01% in any region, were included for further analysis and predictions. 29 Immunogenicity scores: The vector of predicted nM affinities output by netMHC were first normalized across alleles to account for the fact that some alleles have higher affinity across all peptides, and to allow for the relatively equivalent contribution of all alleles. These values were then transformed using the 1-log(affinity) transformation also borrowed from netMHC such that lower nM affinities will result in larger resulting values.

These transformed, normalized affinities are then weighted by population allele frequency and summed across all alleles and epitopes. Finally, the scores are standardized across proteins to facilitate comparison.

Cluster analysis: Network analysis was performed by first thresholding genotypes to include only those identified as hits from the screen. These were genotypes appearing

in the pre-screen plasmid library, both post-screen replicates, and having a fold-change enrichment larger than the wild-type sequence (4.5 fold enrichment). These hits were used to create a graph with nodes corresponding to genotypes and node sizes corresponding to fold change enrichment. Edges were placed between nodes at most 4 mutations distant from each other, and edge weights were defined by 1/d where d is distance between genotypes. Network analysis was done using python bindings of igraph. Plots were generated using the Fruchterman-Reingold force-directed layout algorithm. Ultra-15 centrifugal filter unit with a 100,000 NMWL cutoff (Millipore), and frozen at -80C. 31 RT-qPCR: cDNA was synthesized from RNA using the Protoscript II First Strand cDNA Synthesis Kit (NEB). qPCR was performed using a CFX Connect Real Time PCR Detection System (Bio-Rad). All samples were run in triplicates and results were normalized against GAPDH expression. Primers for qPCR are listed in Table 1 below. 

Generating the optimal mRNA for therapy: HPLC purification eliminates immune activation and improves translation of nucleoside-modified, protein-encoding mRNA

Codon optimality is a major determinant of mRNA stability

Phosphorothioate cap analogs increase stability and translational efficiency of RNA vaccines in immature dendritic cells and induce superior immune responses in vivo

Modification of antigen-encoding RNA increases stability, translational efficacy, and T-cell stimulatory capacity of dendritic cells

Improving mRNA-Based Therapeutic Gene Delivery by Expression-Augmenting 3′ UTRs Identified by Cellular Library Screening

Engineering circular RNA for potent and stable translation in eukaryotic cells

RNA circularization strategies in vivo and in vitro

RNA Circularization Diminishes Immunogenicity and Can Extend Translation Duration In Vivo

Rolling Circle Translation of Circular RNA in Living Human Cells

Pervasive translation of circular RNAs driven by short IRES-like elements

Natural RNA circles function as efficient microRNA sponges

Detecting and characterizing circular RNAs

Highly efficient expression of circular RNA aptamers in cells using autocatalytic transcripts

Efficient in vitro and in vivo RNA editing via recruitment of endogenous ADARs using circular guide RNAs

mRNA vaccines for infectious diseases: principles, delivery and clinical translation

SARS-CoV-2 mRNA vaccine design enabled by prototype pathogen preparedness

Neutralizing antibody vaccine for pandemic and pre-emergent coronaviruses

Efficacy and safety of the BNT162b2 mRNA COVID-19 vaccine in participants with a history of cancer: subgroup analysis of a global phase 3 randomized clinical trial

Initiation of protein synthesis by the eukaryotic translational apparatus on circular RNAs

A segment of the 5' nontranslated region of encephalomyocarditis virus RNA directs internal entry of ribosomes during in vitro translation

Maximizing the potency of siRNA lipid nanoparticles for hepatic gene silencing in vivo

A Novel Amino Lipid Series for mRNA Delivery: Improved Endosomal Escape and Sustained Pharmacology and Safety in Non-human Primates

A mechanistic overview of translation initiation in eukaryotes

Secondary structure of two regions in expansion segments ES3 and ES6 with the potential of forming a tertiary interaction in eukaryotic 40S ribosomal subunits

The beta hairpin structure within ribosomal protein S5 mediates interplay between domains II and IV and regulates HCV IRES function

Pervasive functional translation of noncanonical human open reading frames

Principles of translational control: an overview

Immunogenicity and structures of a rationally designed prefusion MERS-CoV spike antigen

Gene editing in human stem cells using zinc finger nucleases and integrase-defective lentiviral vector delivery

Gene targeting of a disease-related gene in human induced pluripotent stem and embryonic stem cells

Author Correction: Immune-orthogonal orthologues of AAV capsids and of Cas9 circumvent the immune response to the administration of gene therapy

A multifunctional AAV-CRISPR-Cas9 and its host response

Cell Dependent Immunogenicity of Protein Therapeutics Pre-clinical Assessment and Mitigation-Updated Consensus and Review 2020

Multifunctional CRISPR-Cas9 with engineered immunosilenced human T cell epitopes

How single mutations affect viral escape from 37 broad and narrow antibodies to H1 influenza hemagglutinin

A catalogue of biochemically diverse CRISPR-Cas9 orthologs

Nature and intensity of selection pressure on CRISPR-associated genes

Gapped sequence alignment using artificial neural networks: application to the MHC class I system

Reliable prediction of T-cell epitopes using neural networks with novel sequence representations

Removing T-cell epitopes with computational protein design

Elimination of murine and human T-cell epitopes in recombinant immunotoxin eliminates neutralizing and anti-drug antibodies in vivo

From squiggle to basepair: computational approaches for improving nanopore sequencing read accuracy

Population-specific design of de-immunized protein biotherapeutics

Enriching CRISPR-Cas9 targeted cells by co-targeting the HPRT gene

HPRT1 activity loss is associated with resistance to thiopurine in ALL

Quantitative Description of a Protein Fitness Landscape Based on Molecular Features

RNA-guided human genome engineering via Cas9

Situ Gene Therapy via AAV-CRISPR-Cas9-Mediated Targeted Gene Regulation

Genome-wide programmable transcriptional memory by CRISPR-based epigenome editing

Inheritable Silencing of Endogenous Genes by Hit-and-Run Targeted Epigenetic Editing

Structured elements drive extensive circular RNA translation

Extensive translation of circular RNAs driven by N6-methyladenosine

5′ UTR m6A Promotes Cap-Independent Translation

Systematic discovery of cap-independent translation sequences in human and viral genomes

Human 5′ UTR design and variant effect prediction from a massively parallel translation assay

Sensing Self and Foreign Circular RNAs by Intron Identity

N6-Methyladenosine Modification Controls Circular RNA Immunity

Large-scale discovery of protein interactions at residue resolution using 40 co-evolution calculated from genomic sequences

Designs of Antigen Structure and Composition for Improved Protein-Based Vaccine Efficacy

T cell antigen discovery via signaling and antigen-presenting bifunctional receptors

Rapid discovery of potent siRNA-containing lipid nanoparticles enabled by controlled microfluidic formulation

Microfluidic Synthesis of Highly Potent Limit-size Lipid Nanoparticles for In Vivo Delivery of siRNA

We thank members of the Mali lab for discussions, advice and help with experiments.This work was generously supported by UCSD Institutional Funds, NIH grants (R01HG009285, R01CA222826, R01GM123313), Department of Defense Grant (DOD PR210085), and a Longevity Impetus Grant from Norn Group. This publication includes