Main
The ability to safely and efficiently deliver macromolecules into relevant cell populations in culture (in vitro) and in the body (in vivo) is a requirement for many emerging therapeutic modalities. Current gene editing technologies 1,2 , for example, are often constrained by the challenge of delivering gene editing agents into relevant cell types in vitro and in vivo 3,4,5 . Viral vectors such as adeno-associated virus (AAV) have been used to deliver gene editing agents into several tissues in vivo, including in clinical trials 3,6,7,8,9,10,11,12 . However, AAV delivery is limited by cargo size restrictions 13 , the possibility of unwanted DNA cargo integration into the genomes of transduced cells 14,15 and the prolonged expression of gene editing agents in transduced cells, which increases risks of off-target editing 3,9,16 . Some nonviral delivery methods, including lipid nanoparticles (LNPs), offer reduced off-target editing by transiently delivering editor-encoding mRNA instead of DNA; however, using LNP delivery to achieve therapeutic gene editing in extrahepatic tissues remains challenging 3,17,18 despite recent encouraging progress 19,20,21,22 . Thus, the development of additional delivery strategies is needed to overcome the limitations of these existing methods.
Recently, we and others have explored the use of virus-like particles (VLPs) as vehicles for delivering gene editing agents into cells in vitro or in vivo 3,16,23,24,25,26,27,28,29,30,31,32,33,34,35 . VLPs consist of viral scaffolds that package and deliver cargo proteins, ribonucleoproteins (RNPs) or mRNAs instead of cargo-encoding viral genomes. Thus, VLP delivery offers the efficient transduction and tissue tropisms of viral delivery methods with the transient cargo expression and reduced off-target editing of nonviral delivery methods 3,16 , an ideal combination for gene editing applications.
Several VLP-based strategies for delivering gene editing agents into mammalian cells have been previously described 3,16,23,24,25,26,27,28,29,30,31,32,33,34,35 . We recently developed engineered VLPs (eVLPs) that enable efficient protein delivery and gene editing in cell culture and in the mouse liver and retina 16 . In eVLPs, desired cargo proteins are fused to retroviral Gag (capsid) proteins, which directs localization of the cargo into viral particles as they form in producer cells. The Gag–cargo linker contains a sequence engineered to be cleaved at a carefully tuned rate by the copackaged retroviral protease following particle formation, which releases the cargo inside the particles and subsequently into the transduced cells. Additionally, the cell-type specificity of eVLPs is determined by the envelope glycoprotein used to pseudotype the particles. By iteratively engineering eVLPs to improve cargo loading, cargo release and component stoichiometry, we developed an optimized fourth-generation (v4) eVLP architecture that was critical for enabling efficient in vivo base editing with minimal off-target editing compared to AAV delivery. We also recently reported prime editor (PE)-eVLPs, which enable transient in vivo delivery of therapeutic PE RNPs with minimal off-target editing and no risk of oncogenic transgene insertion 36 . These favorable characteristics of eVLPs suggest that eVLP delivery has the potential to serve as a useful modality for the in vitro and in vivo delivery of gene editing RNPs or other therapeutic proteins.
Additional improvements to the properties of eVLPs are needed to maximize their potential for research and therapeutic applications. In particular, increasing eVLP packaging efficiency or per-particle transduction efficiency would enable more efficient gene editing with lower eVLP doses. Directed laboratory evolution is a promising approach for improving delivery vehicles and has been used extensively to develop viral delivery vectors with desired properties 3,37,38,39,40,41,42,43,44,45,46 . Existing evolution schemes require each viral variant to package a viral genome that encodes that particular variant’s identity, which allows sequencing the viral genomes that survive a selection to identify variants that possess the desired properties. Because eVLPs do not package any viral genetic material, however, applying directed evolution to improve eVLPs requires developing an alternative strategy to encode an eVLP variant’s identity.
To achieve this goal, we developed a directed evolution system in which each eVLP variant packages RNPs loaded with guide RNAs containing a barcode sequence that uniquely identifies that particular eVLP variant. After applying selections for specific properties, desired eVLP variants are identified by sequencing the surviving barcoded guide RNAs. Using this system, we generated a library of eVLP capsid mutants and performed selections to identify capsid mutants that support improved eVLP production from producer cells or improved eVLP transduction of target cells. By combining the beneficial capsid mutations, we developed v5 eVLPs, which exhibit increased RNP packaging, improved cargo release, distinct capsid structural compositions and a 2–4-fold increase in in vitro delivery potency compared to previous-best v4 eVLPs. A key capsid mutation in v5 eVLPs abolishes an interaction that is critical for packaging viral genomes in wild-type viruses but is not required in RNP-packaging eVLPs that lack viral genomes, highlighting the benefits of mutating and explicitly selecting eVLP capsids to package non-native RNP cargos instead of viral genomes. Our results lay a foundation for evolving eVLPs with improved properties.
Results Barcoded guide RNAs identify eVLPs with distinct properties
All directed evolution systems require a way to identify desired variants following a selection for specific properties 47,48 . To overcome the challenge created by the lack of viral genetic material within eVLPs, we envisioned a strategy to encode the identity of each eVLP variant using barcoded single-guide RNAs (sgRNAs) within eVLP-packaged RNP cargos (Fig. 1a). In this scheme, each eVLP production vector expresses both an eVLP variant and a barcoded sgRNA that uniquely identifies that eVLP variant (Fig. 1a). These barcoded eVLP production vectors are introduced into producer cells under conditions that maximize the likelihood that each producer cell receives only a single barcoded vector and, therefore, produces only a single eVLP variant–barcoded sgRNA combination. This strategy generates barcoded eVLP libraries in which each unique eVLP variant packages sgRNAs containing a unique corresponding barcode (Fig. 1a).
Fig. 1: Validation of the barcoded eVLP evolution system.
a, Overview of the barcoded eVLP evolution system. Each unique eVLP variant is linked to a unique barcoded sgRNA on the same eVLP production vector. eVLP production maintains barcode–variant correspondence and yields a barcoded eVLP library. In the resulting library, each eVLP variant packages RNPs containing barcoded sgRNAs that encode the identity of that particular eVLP variant. Barcodes that are enriched following a selection for desired properties identify eVLP variants that possess the desired properties. b, Schematic of the mock production selection experiment with barcode 1 linked to a functional Gag–ABE construct and barcode 2 linked to a nonfunctional ABE only (no Gag) construct. c, Frequencies of barcodes 1 or 2 detected in either the producer-cell gDNA or the eVLP-packaged sgRNAs. Bars reflect the mean of n = 3 biological replicates and dots represent individual replicate values. d, Schematic of the mock transduction selection experiment with barcode 1 linked to wild-type VSV-G envelope protein and barcode 2 linked to an impaired VSV-Gmut envelope protein. e, Frequencies of barcodes 1 or 2 detected in either the eVLP-packaged sgRNAs or the delivered sgRNAs. Bars reflect the mean of n = 3 biological replicates and dots represent individual replicate values.
Full size image
After subjecting a barcoded eVLP library to a selection for a desired property, eVLP variants surviving selection are identified by sequencing their sgRNAs and determining which barcodes are enriched in the postselection population compared to the input population (Fig. 1a). This scheme for evolving barcoded eVLPs in principle can be used to evolve different eVLP components—including capsid, envelope, cargo and other structural proteins—by placing the evolving component on the same vector as the barcoded sgRNA when constructing the library of eVLP production vectors (Fig. 1a). Additionally, this scheme is compatible with a wide range of selections for desired properties, including improved particle production, particle stability or transduction of a particular cell type in vitro or in vivo.
We first validated that barcoded sgRNAs are compatible with functional eVLP production. We inserted a 15-bp barcode sequence into the tetraloop of the sgRNA scaffold (Extended Data Fig. 1a), as previous studies showed that this location and length of insertion does not disrupt sgRNA function 49,50 . For all validation experiments, we used our previously developed v4 base editor (BE)-eVLPs 16 that package a highly active adenine base editor (ABE8e) RNP cargo 51 . Standard v4 BE-eVLPs are produced by cotransfecting four expression plasmids into producer cells (Extended Data Fig. 1b), encoding the expression of (1) the Gag–ABE fusion; (2) the sgRNA that directs on-target base editing in the transduced cells; (3) the Moloney murine leukemia virus (MMLV) Gag–Pro–Pol polyprotein, which contains the required viral protease and other structural components; and (4) the vesicular stomatitis virus G (VSV-G) envelope protein.
We produced v4 eVLPs containing canonical or tetraloop-barcoded sgRNAs with four arbitrarily chosen barcodes and compared their potencies by measuring base editing efficiencies at the BCL11A enhancer locus in eVLP-transduced HEK293T cells. We observed that barcoded eVLPs exhibited comparable potency to standard eVLPs and that eVLPs produced with distinct barcoded sgRNAs exhibited comparable potencies (Extended Data Fig. 1c,d). Because the evolution scheme requires that the barcoded sgRNA and evolving eVLP component are expressed from the same vector, we also confirmed that a single vector containing both an sgRNA expression cassette and a Gag–ABE fusion could support efficient eVLP production and cargo delivery (Extended Data Fig. 1c). Reverse transcription quantitative PCR (RT–qPCR) analysis confirmed that eVLPs lacking Gag–ABE package 216-fold fewer sgRNA molecules compared to canonical v4 eVLPs, suggesting that sgRNA packaging in the absence of Gag–ABE is negligible and, therefore, not likely to influence selection outcomes (Extended Data Fig. 1e). These results together indicate that barcoded BE-eVLPs can be produced in a manner that preserves standard BE-eVLP properties.
Lastly, we validated that barcoded eVLPs can be used to distinguish between eVLP variants with different functional properties. To do so, we performed two mock selections: a selection for cargo-loaded eVLP production and a selection for eVLP transduction of HEK293T cells. We performed the mock eVLP production selection using two different BE-eVLP cargo constructs: (1) a standard Gag–ABE cargo construct used in v4 eVLPs and (2) a nonfunctional cargo construct containing an ABE but no Gag fusion, which almost completely abolishes ABE cargo loading into eVLPs 16 . We paired each of these two cargo constructs with a unique barcoded sgRNA and used lentiviral integration to generate producer cells expressing either barcode 1 (corresponding to Gag–ABE) or barcode 2 (corresponding to ABE only) (Fig. 1b). We then initiated eVLP production from a 1:1 mixture of these producer cells. Because only the barcode 1 (Gag–ABE) producer cells and not barcode 2 (ABE only) producer cells can produce functional eVLPs containing substantial amounts of ABE RNP cargo, we anticipated that barcode 1 would be enriched in eVLPs, while barcode 2 would be depleted (Fig. 1b). Indeed, we observed that barcode 1 was strongly enriched 13-fold (93% of sequencing reads) compared to barcode 2 (7% of sequencing reads) in eVLP-packaged sgRNAs even though barcodes 1 and 2 were equally represented in the original producer cell mixture (Fig. 1c).
Next, we performed a mock eVLP transduction selection using two different eVLP envelope constructs: (1) a standard VSV-G envelope construct that enables transduction of most cell types, and (2) an impaired VSV-Gmut envelope construct that reduces but does not completely eliminate the ability of viral particles to engage target cell surface receptors and deliver their packaged cargo in the absence of additional targeting ligands 34,52,53 . We produced VSV-G-pseudotyped eVLPs packaging barcode 1 and VSV-Gmut-pseudotyped eVLPs packaging barcode 2 and then transduced HEK293T cells with a 1:1 mixture of these barcoded eVLPs (Fig. 1d). Because VSV-G-pseudotyped eVLPs should more efficiently transduce cells and deliver RNP cargo compared to VSV-Gmut-pseudotyped eVLPs, we anticipated that barcode 1 would be enriched in sgRNAs retrieved from eVLP-transduced cells compared to sgRNAs packaged in the input eVLP mixture while barcode 2 would be depleted (Fig. 1d). Indeed, we observed that barcode 1 was enriched 2.4-fold (71% of sequencing reads) compared to barcode 2 (29% of sequencing reads) in sgRNAs retrieved from eVLP-transduced cells even though barcodes 1 and 2 were equally represented in the original eVLP mixture (Fig. 1e).
These results of the mock eVLP production selection and mock eVLP transduction selection demonstrate that barcoded sgRNAs can be used to label different eVLP variants and that barcodes that are enriched following a selection identify variants with increased fitness. Collectively, these findings validate key aspects of the barcoded eVLP evolution system and establish a framework for using barcoded sgRNAs to identify eVLP variants with desired properties.
Mutating and selecting a barcoded eVLP capsid library
Next, we applied the barcoded eVLP evolution system to mutate and select eVLP capsids with improved properties. The capsid proteins that are used in v4 eVLPs are identical to the capsid proteins used in wild-type MMLV, which have evolved in nature to be optimal for packaging viral genomes 54,55 . Therefore, wild-type MMLV capsids are likely not optimal for packaging large, non-native protein cargos such as ABEs in eVLPs. We hypothesized that remodeling the internal eVLP capsid surface to optimize ABE RNP cargo packaging instead of viral genome packaging could substantially improve eVLP properties, including potency per particle, number of cargo molecules packaged per particle, overall particle yield or titer and particle stability.
To mutate and select eVLP capsids to become more optimal for packaging ABE RNP cargo, we first designed and constructed a barcoded eVLP capsid library. This library contained 3,762 single-residue mutants of the MMLV Gag protein capsid (amino acids 215–313 and 413–479) and nucleocapsid (amino acids 480–513) domains in the Gag–ABE cargo construct (Extended Data Fig. 2a). We implemented a library construction strategy that preserved the association between barcodes and mutants to enable decoding of selection outcomes (Extended Data Fig. 2b,c). Barcodes were chosen such that no two barcode sequences were within four mismatches of each other to minimize the likelihood of incorrect barcode classification because of sequencing errors during barcode retrieval or mutations during eVLP production.
We used this barcoded plasmid library to generate a library of barcoded eVLP producer cells (Fig. 2a). Lentiviral transduction of producer cells at a low multiplicity of infection (MOI) followed by expansion of the transduced cells maximized the fraction of producer cells that each received a single barcode–capsid variant pair (Supplementary Table 5). High-throughput sequencing analysis of genomic DNA (gDNA) isolated from the expanded producer cell library revealed that 99% of all barcode sequences were detected (Extended Data Fig. 3). These results demonstrate the successful generation of barcoded eVLP plasmid and producer cell libraries, laying the foundation for eVLP evolution campaigns.
Fig. 2: Barcoded eVLP capsid selections.
a, Schematic of the barcoded eVLP capsid library generation. Each unique capsid mutant was linked to a unique barcoded sgRNA on the same plasmid vector. These barcoded vectors were used to produce lentivirus, which was then used to generate a barcoded producer cell library in which each producer cell contained a single integration of a barcoded sgRNA and capsid mutant expression cassette. Following expansion of transduced cells, the barcoded producer cell library was transfected with the other plasmids necessary for eVLP production to generate a barcoded eVLP capsid library. b, Overview of selections for improved eVLP production and improved eVLP transduction. Barcodes enriched in eVLP-packaged sgRNAs relative to producer-cell gDNA identify capsid mutants that support improved eVLP production. Barcodes enriched in eVLP-transduced cells relative to eVLP-packaged sgRNAs identify capsid mutants that support improved eVLP transduction. c, Average barcode enrichment values for each capsid mutant in the production selection and transduction selection. Each capsid mutant is shown as a single dot whose x coordinate reflects the average production enrichment of the capsid mutant and y coordinate reflects the average transduction enrichment of the capsid mutant. The canonical capsid used in v4 eVLPs is shown as a red dot and the corresponding enrichment values associated with this dot are shown as dotted red lines. Capsid mutants selected for further characterization are shown as blue dots. Production and transduction enrichment values were calculated as the average of n = 2 replicates. Further details are provided in Extended Data Figs. 4 and 5 and Supplementary Table 3.
Full size image
Selections reveal how capsid mutants affect eVLP properties
We subjected the barcoded eVLP capsid library to two separate selections (Fig. 2b): (1) a selection for improved eVLP production from producer cells, and (2) a selection for improved eVLP transduction of HEK293T cells. To perform a selection for improved eVLP production, we initiated eVLP production from the barcoded producer cell library (Methods). We purified the resulting library of barcoded eVLP capsid variants, isolated the eVLP-packaged sgRNAs and sequenced the barcodes that were present after this production selection. For each barcode sequence in the library, we calculated the eVLP production enrichment by comparing the frequency of that barcode in eVLP-packaged sgRNAs to the frequency of that barcode in the producer-cell gDNA. In this production selection, barcodes that display greater enrichment than the canonical eVLP capsid barcode identify candidate capsid mutants that support improved production compared to the canonical capsid (Extended Data Fig. 4a). Enriched barcodes, for example, might indicate that those capsid mutants package more RNP cargo molecules per particle than the canonical capsid or are produced at a higher titer, either of which could explain why those particular sgRNAs were more abundant in the produced eVLPs relative to producer-cell gDNA.
Approximately 8% of all capsid mutants in the library exhibited an average production enrichment higher than that of the canonical eVLP capsid (Fig. 2c and Extended Data Fig. 4b). Because the complete MMLV capsid consists of a complex assembly of thousands of capsid subunits 54 , it is likely that the majority of capsid mutations disrupt the carefully orchestrated process of capsid assembly, explaining the rarity of mutants enriched beyond that of the canonical eVLP capsid in the eVLP production selection. The enrichment of a subpopulation of capsid mutants in the production selection beyond canonical eVLPs (Fig. 2c and Extended Data Fig. 4b), however, supported our hypothesis that the wild-type MMLV capsid is not optimal for RNP cargo packaging and that eVLP capsids can be mutated and selected in the laboratory to improve this property.
In addition to improving eVLP production, the eVLP evolution system can be used to improve transduction of eVLPs into target cells (Fig. 2b). We incubated HEK293T cells with the purified barcoded eVLP capsid library and isolated sgRNAs that were successfully transduced into target cells after 6 h. For each barcode sequence in the library, we calculated the eVLP transduction enrichment by comparing the frequency of that barcode in the transduced HEK293T cells to the frequency of that barcode in the eVLP-packaged sgRNAs before incubation with HEK293T cells. Barcodes that are enriched to a higher degree than the canonical v4 eVLP barcode identify capsid mutants that support improved transduction relative to the v4 eVLP capsid (Extended Data Fig. 5a). Enriched barcodes, for example, might reflect capsid mutants that transduce target cells more efficiently because they are more stable or enter target cells more efficiently. Notably, we observed that only 0.7% of all capsid mutants in the library exhibited an average transduction enrichment greater than that of the canonical v4 eVLP capsid (Fig. 2c and Extended Data Fig. 5b). These findings support a model in which capsid mutants are more likely to improve eVLP production or RNP cargo packaging but rarely improve particle stability, cell entry or other characteristics that influence transduction.
By integrating the results from both the production and the transduction selections, we generated a landscape that reveals how each capsid mutant influences these two properties of eVLPs (Fig. 2c). The vast majority of capsid mutants exhibited worse production and transduction efficiencies compared to the canonical v4 eVLP capsid. While a handful of mutants showed selection enrichments that suggest improvements in either production or transduction, virtually no mutants exhibited improvements in both properties, suggesting that eVLP production and transduction efficiencies are dictated by distinct and potentially competing mechanisms. Certain clusters of mutations consistently impacted eVLP production and transduction. For example, R440P or R443P improved transduction but negatively impacted production (Fig. 2c). Conversely, L478K, A479K or T480H improved production but modestly impaired transduction (Fig. 2c). These observations suggest that remodeling the internal charged surfaces of the eVLP capsid is a potential strategy for optimizing eVLP capsids to better package and deliver RNP cargo. Together, the results of the eVLP capsid selections demonstrate the utility of the barcoded eVLP system, reveal new insights into how different capsid mutations influence eVLP properties and nominate potentially improved capsid mutants that warrant further characterization.
Combinations of capsid mutations improve eVLP potency
On the basis of the results of the production and transduction selections, we identified a set of 36 capsid mutants for further characterization (blue dots in Fig. 2c). We chose these mutants on the basis of their positive enrichments in both replicates of the production or transduction selections, prioritizing mutants that improved one property without substantially impairing the other property (Fig. 2c). To perform a high-throughput assessment of the potency of multiple different variants simultaneously, we produced different eVLP variants through transient transfection of eVLP plasmids into producer cells in different wells of 96-well plates, transduced HEK293T cells with the same volume of each eVLP variant at a subsaturating dose (Methods) and determined each the potency of each variant by measuring adenine base editing efficiencies at the sgRNA-specified target BCL11A enhancer locus in the transduced cells. These eVLP production conditions impose an eVLP stoichiometry of 25:75 Gag–ABE:Gag–Pro–Pol, a stoichiometry that we previously determined to be optimal for eVLP potency 16 . This 25:75 stoichiometry likely differs from the stoichiometry imposed during the barcoded library selection conditions because eVLP production from singly integrated Gag–ABE producer cell lines likely results in a low Gag–ABE:Gag–Pro–Pol ratio. While we anticipated that this stoichiometry difference might lead to differences between an individual mutant’s performance in selections versus in potency assays, we chose to compare all mutants to canonical v4 eVLPs using optimal transfection-based production conditions to assess whether any of these mutants could outperform v4 eVLPs in this most relevant setting.
We began by introducing each of the 36 capsid mutants into the v4 Gag–ABE construct and used canonical versions of the other components of the v4 eVLP architecture (wild-type MMLV Gag–Pro–Pol, VSV-G and standard sgRNA) to produce the evolved eVLP variants. In this context, we observed that most of the mutations did not improve eVLP potency compared to v4 eVLPs (Extended Data Fig. 6a and Supplementary Fig. 1a), indicating that incorporating the capsid mutations into the Gag–ABE construct alone was not sufficient to improve potency.
Because the processed Gag protein expressed in the Gag–Pro–Pol construct, along with the processed Gag protein expressed in the Gag–ABE construct, are both important contributors to the overall eVLP capsid, we also incorporated the capsid mutations into the Gag–Pro–Pol construct used for eVLP production (Extended Data Fig. 6b). We first incorporated the Q226P mutation into the Gag–Pro–Pol construct (hereafter referred to as GagQ226P–Pro–Pol), because the Q226P mutation was the most strongly enriched mutation from the production selection that only modestly impaired transduction (Fig. 2c). Next, we assessed the potency of the same 36 capsid mutants in the Gag–ABE construct but now using the GagQ226P–Pro–Pol instead of the wild-type MMLV Gag–Pro–Pol. In this context, many of the tested capsid mutants exhibited 2–3-fold increases in BE delivery potency compared to v4 eVLPs (Fig. 3a and Supplementary Fig. 1b). In general, we observed that mutants that modestly improved production exhibited greater improvements in potency compared to mutants that substantially improved production (Extended Data Fig. 7), consistent with a model in which excessive improvements in eVLP production or cargo packaging might be detrimental to overall eVLP potency.
Fig. 3: Combinations of capsid mutations improve eVLP potency.
a, Fold change in eVLP potency compared to v4 eVLPs of each capsid mutant incorporated individually into the Gag–ABE construct and paired with GagQ226P–Pro–Pol. b, Fold change in eVLP potency relative to v4 eVLPs of each C-terminal capsid mutant with or without the Q226P mutant incorporated into either the Gag–ABE only, Gag–Pro–Pol only or both Gag–ABE and Gag–Pro–Pol. In a,b, bars reflect the mean of n = 3 biological replicates and dots represent individual replicate values. c, Comparison of v4 eVLPs and Gag–ABE mutants paired with the GagQ226P–Pro–Pol across a range of eVLP doses. Adenine base editing efficiencies at position A7 of the BCL11A enhancer site in HEK293T cells are shown. eVLPs were produced at a concentration of approximately 5 × 108 eVLPs per microlitre. d–i, Comparison of v4 and v5 eVLPs in mouse N2A cells across a range of eVLP doses. Adenine base editing efficiencies are shown at position A8 of the Angptl3 exon 7 splice acceptor site (d), position A4 of the Rosa26 site (e), position A9 of the Dnmt1 site (f), position A6 of the Pcsk9 exon 4 splice acceptor site (g), position A4 of the Pcsk9 exon 6 splice donor site (h) and position A8 of the Pcsk9 exon 8 splice acceptor site (i). In c–i, dots and error bars represent the mean ± s.e.m. of n = 3 biological replicates. Data were fit to four-parameter logistic curves using nonlinear regression.
Full size image
In light of the discovery that different Gag–ABE and Gag–Pro–Pol capsid mutants can synergize, we systematically evaluated the effects of incorporating different combinations of mutations into the Gag–ABE or Gag–Pro–Pol constructs. We selected five Gag–ABE mutants that exhibited the highest potency when paired with the GagQ226P–Pro–Pol: R501I, D502Q, A505W, C507F and C507V (Fig. 3a). We tested all possible combinations of each C-terminal mutant and Q226P mutant incorporated into the Gag–ABE only, Gag–Pro–Pol only or both Gag–ABE and Gag–Pro–Pol (Fig. 3b). Notably, four of the five C-terminal Gag–ABE mutants still performed best when paired with the GagQ226P–Pro–Pol instead of a matched Gag–Pro–Pol containing that same C-terminal mutant (Fig. 3b). These findings reveal the complex interplay between different capsid mutations and underscore the importance of assessing these mutations in several possible eVLP configurations.
We next evaluated the potency of these five eVLP variants (R501I, D502Q, A505W, C507F or C507V Gag–ABE mutants paired with GagQ226P–Pro–Pol) compared to v4 eVLPs across a range of doses in HEK293T cells (Fig. 3c). As expected, all eVLP variants exhibited improved base editing efficiencies at all doses tested compared to v4 eVLPs (Fig. 3c). In particular, the GagC507V–ABE + GagQ226P–Pro–Pol combination exhibited an average overall 3.7-fold improvement in potency (half-maximal effective concentration, EC50) (Fig. 3c). This substantial improvement in potency is comparable to what we observed between v2 and v1 eVLPs or v3 and v2 eVLPs in our previous study 16 . Therefore, we designated the GagC507V–ABE + GagQ226P–Pro–Pol combination as v5 BE-eVLPs (Fig. 3c).
We also investigated whether incorporating mutations identified in the BE-eVLP selections might also improve the delivery potency of eVLPs that package other gene editing cargos, such as Cas9 nuclease or PEs. We observed that Cas9-eVLPs containing GagC507V–Cas9 + GagQ226P–Pro–Pol exhibited an average twofold improvement in potency (EC50 value) compared to v4 Cas9-eVLPs at both the BCL11A enhancer site and the EMX1 site in HEK293T cells (Extended Data Fig. 8a,b). Therefore, we designated the GagC507V–Cas9 + GagQ226P–Pro–Pol combination as v5 Cas9-eVLPs. We also observed that v3 or v3b PE2-eVLPs 36 containing the Q226P mutation incorporated into the relevant Gag–Pro–Pol constructs exhibited comparable potency to canonical v3 or v3b PE2-eVLPs (Extended Data Fig. 8c,d). Together, these results suggest that mutations identified in the BE-eVLP selections can similarly improve the delivery potency of Cas9-eVLPs, which are highly similar in architecture to BE-eVLPs, but not PE-eVLPs, which contain multiple additional structural components and RNA-binding proteins specific to PE cargos that might behave differently when combined with the evolved capsid mutants. These findings also raise the possibility that PE-eVLPs or other eVLPs with distinct architectures might benefit from cargo-specific or architecture-specific production and transduction selections that are analogous to those we performed using BE-eVLPs in this study.
Lastly, we evaluated the potency of v5 BE-eVLPs compared to v4 BE-eVLPs across a range of doses and target genomic loci in mouse Neuro-2a (N2A) cells (Fig. 3d–i). In all tested cases, v5 BE-eVLPs exhibited improved potency compared to v4 BE-eVLPs. In general, the base editing efficiency at a given dose of v4 BE-eVLPs can be achieved using a 2–4-fold lower dose of v5 BE-eVLPs (Fig. 3c–i). We anticipate that v5 eVLPs will be especially useful for eVLP delivery applications that are limited by the maximum administrable dose. Collectively, these results demonstrate that the barcoded eVLP evolution system successfully generated improved eVLP capsid mutants that enabled the discovery of v5 BE-eVLPs with improved delivery potency compared to previous-best v4 BE-eVLPs.
v5 eVLPs improve base editing potency in primary human HSPCs
To further investigate the potential utility of v5 eVLPs, we compared the potencies of v4 and v5 BE-eVLPs in primary human hematopoietic stem and progenitor cells (HSPCs). Ex vivo gene editing of autologous HSPCs followed by transplantation has proven to be powerful approach for treating blood disorders, including sickle cell disease and β-thalassemia 56,57,58,59,60,61 . Nuclease-mediated disruption of the BCL11A erythroid-specific enhancer in HSPCs leads to the induction of fetal hemoglobin expression in erythrocytes, which is sufficient to rescue disease phenotypes associated with these blood disorders and is the first US Food and Drug Administration-approved gene editing drug 56,61,62 . Therapeutic induction of fetal hemoglobin has also been achieved using ex vivo cytosine or adenine base editing to install precise single-base conversions within the BCL11A erythroid-specific enhancer or the fetal hemoglobin (HBG) promoter in HSPCs 57,63,64 . Because base editing avoids negative consequences associated with nuclease-generated DNA double-strand breaks and uncontrolled mixtures of indel products 57 , base editing-mediated disruption of the BCL11A erythroid-specific enhancer in HSPCs could potentially offer advantages over nuclease-mediated disruption.
To evaluate eVLP-mediated base editing in HSPCs, we transduced healthy human donor HSPCs (Methods) with v4 or v5 BE-eVLPs packaging ABE8e and an sgRNA targeting the BCL11A erythroid-specific enhancer locus. We observed substantially higher target adenine base editing efficiencies from v5 BE-eVLPs compared to v4 BE-eVLPs across all doses tested (Fig. 4a–c and Extended Data Fig. 9) Overall, v5 BE-eVLPs exhibited an average 2.6-fold improvement in potency (EC50 value) in primary human HSPCs compared to v4 BE-eVLPs (P 30 and using seqkit 80 (version 2.0.0) to ensure that the reads contained the correct flanking sequences surrounding the 15-bp barcode sequence. The numbers of reads containing each unique barcode sequence were quantified using a custom Python script provided in Supplementary Note 1. For quantification purposes, to account for sequencing errors, any reads that contained a sequence that was within two mismatches of a particular barcode sequence in the library were marked as containing that particular barcode sequence. For calculating barcode frequency enrichments in one population relative to another population (for example, eVLP-packaged sgRNAs versus producer-cell gDNA), raw read counts were converted into reads per million (RPM) with a pseudocount of 1 added for each barcode and fold change values were calculated using the RPM values. Only barcodes that were found in>100 total reads in both preselection and postselection populations were analyzed. Production and transduction enrichment values for each capsid mutant are listed in Supplementary Table 3.
High-throughput capsid mutant potency assay
To assess the potency of individual capsid mutants independently in a high-throughput fashion, eVLPs were produced in 96-well plates. Gesicle cells were seeded in 96-well plates at a density of 20,000 cells per well. After 24 h, cells were transfected using the jetPRIME transfection reagent (Polyplus; 114-75) according to the manufacturer’s protocols. A mixture of plasmids expressing VSV-G (4.3 ng), evolved or wild-type MMLV Gag–Pro–Pol (36.3 ng), evolved Gag–ABE8e (12 ng) and an sgRNA targeting the BCL11A enhancer site (47.3 ng) was cotransfected per well. Edge wells were avoided. Then, 24 h after transfection, HEK293T cells were seeded for transduction in separate 96-well plates at a density of 16,000 cells per well. Next, 48 h after transfecting the Gesicle producer cells, the eVLP-containing supernatant was harvested and pipetted directly onto the seeded HEK293T cells without any additional concentration or purification. A volume of 10 µl of crude eVLP-containing supernatant was used to transduce each well of HEK293T cells. Then, 48 h after transduction, gDNA was extracted in 60 µl of lysis buffer as described above. gDNA was amplified and prepared for sequencing as described above to assess editing efficiency. The fold change in potency relative to v4 eVLPs was calculated by dividing the editing efficiency of the capsid variant by the editing efficiency of v4 eVLPs in the same experiment.
Culture and eVLP transduction of primary human HSPCs
CD34-enriched granulocyte colony-stimulating factor-mobilized peripheral blood stem cells were obtained from three deidentified healthy adult donors (Fred Hutchinson Cancer Center). Cells were collected under protocol 985.03, which was approved by the human subjects institutional review board at the Fred Hutchinson Cancer Center. All donors provided written consent.
HSPCs were maintained in stem cell medium: X-VIVO-10 medium (Lonza; 04-380Q) supplemented with 100 ng µl−1 human stem cell factor (R&D Systems; 255-SC/CF), 100 ng µl−1 human thrombopoietin (R&D Systems; 288-TP/CF) and 100 ng µl−1 human Flt3 ligand (R&D Systems; 308-FK/CF). HSPCs were thawed using a ThawSTAR CFT2 automated thawing system (Biolife Solutions) according to the manufacturer’s protocol. Immediately after thawing, cells were counted and seeded at a density of 125,000 cells per well in 50 µl of stem cell medium in a round-bottom 96-well plate. Before eVLP transduction, the number of eVLPs per unit volume was determined using the MuLV core antigen ELISA kit (Cell Biolabs; VPK-156) to quantify the number of p30 (capsid) molecules present in each eVLP preparation, assuming a copy number of 1,800 molecules of p30 per eVLP, as previously described 16 . Serial dilutions of ultracentrifuge-purified v4 or v5 BE-eVLPs in PBS were added to each well; each well received 2 µl of eVLPs in PBS. After adding eVLPs, the solution in each well was mixed gently but thoroughly by pipetting. eVLP-transduced HSPCs were immediately incubated at 37 °C for 4 h and then 75 µl of prewarmed stem cell medium was added to each well to achieve a cell density of ~1 × 106 cells per ml.
Next, 48 h after eVLP transduction, the total number of viable cells in a subset of treatment conditions was measured using a NucleoCounter NC-3000 (ChemoMetec) and Solution 13: AO•DAPI staining reagent (ChemoMetec) according to the manufacturer’s protocols using NucleoView NC-3000 software (version 2.1.25.12). On the basis of the measured cell numbers, 80 µl of prewarmed stem cell medium was added to each well to maintain a cell density of 1× 106–2 × 106 cells per ml. Then, 96 h after eVLP transduction, HSPCs were collected by centrifugation at 500g for 10 min. Cell pellets were lysed directly for gDNA extraction in 150 µl of lysis buffer as described above and isolated gDNA was sequenced as described above.
Western blot analysis of eVLP protein content
Western blots to analyze the percentage of cleaved ABE cargo in v4 versus v5 eVLPs were performed as described previously 16 . eVLPs were lysed in Laemmli sample buffer (50 mM Tris-HCl pH 7.0, 2% SDS, 10% (v/v) glycerol and 2 mM dithiothreitol) by heating at 95 °C for 15 min. Protein extracts were separated by electrophoresis at 150 V for 45 min on a NuPAGE 3–8% Tris-acetate gel (Thermo Fisher Scientific; EA03755BOX) in 1× NuPAGE Tris-acetate SDS running buffer (Thermo Fisher Scientific; LA0041). Transfer to a PVDF membrane was performed using an iBlot 2 gel transfer device (Thermo Fisher Scientific) at 20 V for 7 min. The membrane was blocked for 1 h at room temperature with rocking in blocking buffer: 1% BSA in TBST (150 mM NaCl, 0.5% Tween-20 and 50 mM Tris-HCl). After blocking, the membrane was incubated overnight at 4 °C with rocking with mouse anti-Cas9 (Cell Signaling Technology; 14697, 1:1,000 dilution). The membrane was washed three times with 1× TBST for 10 min each time at room temperature and then incubated with goat anti-mouse antibody (LI-COR IRDye 680RD; 926-68070, 1:10,000 dilution in blocking buffer) for 1 h at room temperature with rocking. The membrane was washed as before and imaged using an Odyssey imaging system (LI-COR). The relative amounts of cleaved ABE and full-length Gag–ABE were quantified by densitometry using ImageJ and the percentage of cleaved ABE relative to total (cleaved + full-length) ABE was calculated.
eVLP protein content quantification
eVLP protein content quantification was performed as described previously 16 . In brief, eVLPs were lysed in Laemmli sample buffer as described above. The concentration of ABE protein in ultracentrifuge-purified v4 or v5 eVLPs was quantified using the FastScan Cas9 (Streptococcus pyogenes) ELISA kit (Cell Signaling Technology; 29666C) according to the manufacturer’s protocols. Recombinant Cas9 (S. pyogenes) nuclease protein (New England Biolabs; M0386) was used to generate the standard curve for quantification. The concentration of MLV p30 protein in purified eVLPs was quantified using the MuLV core antigen ELISA kit (Cell Biolabs; VPK-156) according to the manufacturer’s protocols. The number of ABE protein molecules per eVLP was calculated by determining the ratio of Cas9 molecules to p30 molecules and assuming a copy number of 1,800 molecules of p30 per eVLP as previously described 16 .
eVLP sgRNA abundance quantification
RNA was extracted from eVLPs and reverse-transcribed as described above. qPCR analysis of the resulting cDNA was performed using a CFX96 Touch real-time PCR detection system (Bio-Rad) with SYBR green dye (Lonza; 50512). The amount of cDNA input was normalized to MLV p30 content and the relative sgRNA abundance per eVLP was calculated as the log2(fold change) (ΔC q ) relative to v4 eVLPs.
Cryo-EM sample preparation and imaging
Frozen eVLPs purified by ultracentrifugation were thawed on ice, centrifuged at 14,000g for 10 min and resuspended with an equivalent volume of cold PBS. Gentle pipet mixing was performed to ensure complete eVLP resuspension. Grid preparation for 300-mesh R2/1 Quantifoil Cu grids was conducted by glow discharge for 60 s at a set current of 25 mA using a K100X Glow Discharger (EM Sciences). Plunge-freezing was performed using a Vitrobot Mark IV (Thermo Fisher Scientific) set at force 4, 4 °C and 95% humidity; within this instrument, 3 μl of resuspended eVLP sample was drop-cast on a glow-discharged grid and then immediately blotted for 3 or 6 s before plunge-freezing in liquid ethane cooled by liquid nitrogen. Sample grids were then placed in liquid nitrogen for storage.
Cryo-EM was performed using a Talos Arctica G2 (Thermo Fisher Scientific) operating at 200 kV and with a Falcon 3 direct electron detector. Focusing was carried out on carbon film adjacent to a hole and four images at ×92,000 magnification were taken in each hole. Calibrated pixel sizes were 1.605 Å. All datasets were acquired automatically using EPU software (version 2.12.1.2782REL) (Thermo Fisher Scientific).
Cryo-EM image analysis
Upon visual inspection of cryo-EM images, eVLPs were classified into one of the following types: (1) enveloped immature capsid; (2) enveloped mature capsid; (3) enveloped undefined capsid; (4) nonenveloped immature capsid; or (5) nonenveloped mature capsid. Immature capsids were identified by their paracrystalline lattice shell, indicating the immature conformation of the capsid N-terminal and C-terminal domains 54 . Mature capsids were identified by an irregular polyhedral capsid shape and the lack of paracrystalline shell 54 . Nonenveloped capsids were identified by the absence of an envelope. Classification and sizing were performed manually in ImageJ (version 2.14.0, National Institutes of Health (NIH)). To minimize the potential for selection bias, every discernible eVLP in the set of cryo-EM images was classified and sized.
Statistical analysis
Data are presented as the mean and s.e.m. No statistical methods were used to predetermine sample size. Statistical analysis was performed using GraphPad Prism software. Sample sizes are described in the figure legends.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Data availability
The high-throughput sequencing data generated during this study were deposited to the National Center for Biotechnology Information (NCBI) Sequence Read Archive (SRA) database under accession PRJNA1034592. Raw data for individual mutants from the eVLP capsid library selections are provided in Supplementary Table 3. Sequences of eVLP protein components are listed in Supplementary Table 4. Key plasmids from this work were deposited to Addgene for distribution. Other plasmids are available from the corresponding authors on request. Source data are provided with this paper.
Code availability
The code used for analyzing base editing efficiencies is available from GitHub (https://github.com/pinellolab/CRISPResso2). Custom scripts generated in this study and used for analyzing eVLP selection data are provided in Supplementary Note 1.
References
Anzalone, A. V., Koblan, L. W. & Liu, D. R. Genome editing with CRISPR–Cas nucleases, base editors, transposases and prime editors. Nat. Biotechnol. 38, 824–844 (2020).
Article
CAS
PubMed
Google Scholar
Doudna, J. A. The promise and challenge of therapeutic genome editing. Nature 578, 229–236 (2020).
Article
CAS
PubMed
PubMed Central
Google Scholar
Raguram, A., Banskota, S. & Liu, D. R. Therapeutic in vivo delivery of gene editing agents. Cell 185, 2806–2827 (2022).
Article
CAS
PubMed
PubMed Central
Google Scholar
van Haasteren, J., Li, J., Scheideler, O. J., Murthy, N. & Schaffer, D. V. The delivery challenge: fulfilling the promise of therapeutic genome editing. Nat. Biotechnol. 38, 845–855 (2020).
Article
PubMed
Google Scholar
Wei, T. et al. Delivery of tissue-targeted scalpels: opportunities and challenges for in vivo CRISPR/Cas-based genome editing. ACS Nano 14, 9243–9262 (2020).
Article
CAS
PubMed
PubMed Central
Google Scholar
Davis, J. R. et al. Efficient prime editing in mouse brain, liver and heart with dual AAVs. Nat. Biotechnol. 42, 253–264 (2023).
Article
PubMed
PubMed Central
Google Scholar
Davis, J. R. et al. Efficient in vivo base editing via single adeno-associated viruses with size-optimized genomes encoding compact adenine base editors. Nat. Biomed. Eng. 6, 1272–1283 (2022).
Article
CAS
PubMed
PubMed Central
Google Scholar
Levy, J. M. et al. Cytosine and adenine base editing of the brain, liver, retina, heart and skeletal muscle of mice via adeno-associated viruses. Nat. Biomed. Eng. 4, 97–110 (2020).
Article
CAS
PubMed
PubMed Central
Google Scholar
Newby, G. A. & Liu, D. R. In vivo somatic cell base editing and prime editing. Mol. Ther. 29, 3107–3124 (2021).
Article
CAS
PubMed
PubMed Central
Google Scholar
Wang, D., Zhang, F. & Gao, G. CRISPR-based therapeutic genome editing: strategies and in vivo delivery by AAV vectors. Cell 181, 136–150 (2020).
Article
CAS
PubMed
PubMed Central
Google Scholar
Maeder, M. L. et al. Development of a gene-editing approach to restore vision loss in Leber congenital amaurosis type 10. Nat. Med. 25, 229–233 (2019).
Article
CAS
PubMed
Google Scholar
Li, T. et al. CRISPR/Cas9 therapeutics: progress and prospects. Signal Transduct. Target Ther. 8, 36 (2023).
Article
PubMed
PubMed Central
Google Scholar
Wu, Z., Yang, H. & Colosi, P. Effect of genome size on AAV vector packaging. Mol. Ther. 18, 80–86 (2010).
Article
CAS
PubMed
Google Scholar
Chandler, R. J., Sands, M. S. & Venditti, C. P. Recombinant adeno-associated viral integration and genotoxicity: insights from animal models. Hum. Gene Ther. 28, 314–322 (2017).
Article
CAS
PubMed
PubMed Central
Google Scholar
Greig, J. A. et al. Integrated vector genomes may contribute to long-term expression in primate liver after AAV administration. Nat. Biotechnol. 42, 1232–1242 (2024).
Article
CAS
PubMed
Google Scholar
Banskota, S. et al. Engineered virus-like particles for efficient in vivo delivery of therapeutic proteins. Cell 185, 250–265 (2022).
Article
CAS
PubMed
PubMed Central
Google Scholar
Loughrey, D. & Dahlman, J. E. Non-liver mRNA delivery. Acc. Chem. Res. 55, 13–23 (2022).
Article
CAS
PubMed
Google Scholar
Paunovska, K., Loughrey, D. & Dahlman, J. E. Drug delivery systems for RNA therapeutics. Nat. Rev. Genet. 23, 265–280 (2022).
Article
CAS
PubMed
PubMed Central
Google Scholar
Breda, L. et al. In vivo hematopoietic stem cell modification by mRNA delivery. Science 381, 436–443 (2023).
Article
CAS
PubMed
PubMed Central
Google Scholar
Cheng, Q. et al. Selective organ targeting (SORT) nanoparticles for tissue-specific mRNA delivery and CRISPR–Cas gene editing. Nat. Nanotechnol. 15, 313–320 (2020).
Article
CAS
PubMed
PubMed Central
Google Scholar
Dilliard, S. A. & Siegwart, D. J. Passive, active and endogenous organ-targeted lipid and polymer nanoparticles for delivery of genetic drugs. Nat. Rev. Mater. 8, 282–300 (2023).
Article
CAS
PubMed
PubMed Central
Google Scholar
Sun, Y. et al. In vivo editing of lung stem cells for durable gene correction in mice. Science 384, 1196–1202 (2024).
Article
CAS
PubMed
Google Scholar
Campbell, L. A. et al. Gesicle-mediated delivery of CRISPR/Cas9 ribonucleoprotein complex for inactivating the HIV provirus. Mol. Ther. 27, 151–163 (2019).
Article
CAS
PubMed
Google Scholar
Choi, J. G. et al. Lentivirus pre-packed with Cas9 protein for safer gene editing. Gene Ther. 23, 627–633 (2016).
Article
CAS
PubMed
Google Scholar
Gee, P. et al. Extracellular nanovesicles for packaging of CRISPR–Cas9 protein and sgRNA to induce therapeutic exon skipping. Nat. Commun. 11, 1334 (2020).
Article
CAS
PubMed
PubMed Central
Google Scholar
Hamilton, J. R. et al. Targeted delivery of CRISPR–Cas9 and transgenes enables complex immune cell engineering. Cell Rep. 35, 109207 (2021).
Article
CAS
PubMed
PubMed Central
Google Scholar
Indikova, I. & Indik, S. Highly efficient ‘hit-and-run’ genome editing with unconcentrated lentivectors carrying Vpr.Prot.Cas9 protein produced from RRE-containing transcripts. Nucleic Acids Res. 48, 8178–8187 (2020).
Article
CAS
PubMed
PubMed Central
Google Scholar
Lu, Z. et al. Lentiviral capsid-mediated Streptococcus pyogenes Cas9 ribonucleoprotein delivery for efficient and safe multiplex genome editing. CRISPR J. 4, 914–928 (2021).
CAS
PubMed
Google Scholar
Lyu, P., Javidi-Parsijani, P., Atala, A. & Lu, B. Delivering Cas9/sgRNA ribonucleoprotein (RNP) by lentiviral capsid-based bionanoparticles for efficient ‘hit-and-run’ genome editing. Nucleic Acids Res. 47, e99 (2019).
Article
PubMed
PubMed Central
Google Scholar
Lyu, P. et al. Adenine base editor ribonucleoproteins delivered by lentivirus-like particles show high on-target base editing and undetectable RNA off-target activities. CRISPR J. 4, 69–81 (2021).
Article
CAS
PubMed
Google Scholar
Lyu, P., Wang, L. & Lu, B. Virus-like particle mediated CRISPR/Cas9 delivery for efficient and safe genome editing. Life (Basel) 10, 366 (2020).
CAS
PubMed
Google Scholar
Mangeot, P. E. et al. Genome editing in primary cells and in vivo using viral-derived nanoblades loaded with Cas9–sgRNA ribonucleoproteins. Nat. Commun. 10, 45 (2019).
Article
CAS
PubMed
PubMed Central
Google Scholar
Yao, X. et al. Engineered extracellular vesicles as versatile ribonucleoprotein delivery vehicles for efficient and safe CRISPR genome editing. J. Extracell. Vesicles 10, e12076 (2021).
Article
CAS
PubMed
PubMed Central
Google Scholar
Strebinger, D. et al. Cell type-specific delivery by modular envelope design. Nat. Commun. 14, 5141 (2023).
Article
CAS
PubMed
PubMed Central
Google Scholar
Haldrup, J. et al. Engineered lentivirus-derived nanoparticles (LVNPs) for delivery of CRISPR/Cas ribonucleoprotein complexes supporting base editing, prime editing and in vivo gene modification. Nucleic Acids Res. 51, 10059–10074 (2023).
Article
CAS
PubMed
PubMed Central
Google Scholar
An, M. et al. Engineered virus-like particles for transient delivery of prime editor ribonucleoprotein complexes in vivo. Nat. Biotechnol. 42, 1526–1537 (2024).
Article
CAS
PubMed
PubMed Central
Google Scholar
Ogden, P. J., Kelsic, E. D., Sinai, S. & Church, G. M. Comprehensive AAV capsid fitness landscape reveals a viral gene and enables machine-guided design. Science 366, 1139–1143 (2019).
Article
CAS
PubMed
PubMed Central
Google Scholar
Byrne, L. C. et al. In vivo-directed evolution of adeno-associated virus in the primate retina. JCI Insight 5, e135112 (2020).
Article
PubMed
PubMed Central
Google Scholar
Dalkara, D. et al. In vivo-directed evolution of a new adeno-associated virus for therapeutic outer retinal gene delivery from the vitreous. Sci. Transl. Med. 5, 189ra176 (2013).
Article
Google Scholar
Deverman, B. E. et al. Cre-dependent selection yields AAV variants for widespread gene transfer to the adult brain. Nat. Biotechnol. 34, 204–209 (2016).
Article
CAS
PubMed
PubMed Central
Google Scholar
Maheshri, N., Koerber, J. T., Kaspar, B. K. & Schaffer, D. V. Directed evolution of adeno-associated virus yields enhanced gene delivery vectors. Nat. Biotechnol. 24, 198–204 (2006).
Article
CAS
PubMed
Google Scholar
Tabebordbar, M. et al. Directed evolution of a family of AAV capsid variants enabling potent muscle-directed gene delivery across species. Cell 184, 4919–4938 (2021).
Article
CAS
PubMed
PubMed Central
Google Scholar
Hwang, B. Y. & Schaffer, D. V. Engineering a serum-resistant and thermostable vesicular stomatitis virus G glycoprotein for pseudotyping retroviral and lentiviral vectors. Gene Ther. 20, 807–815 (2013).
Article
CAS
PubMed
PubMed Central
Google Scholar
Vu, H. N., Ramsey, J. D. & Pack, D. W. Engineering of a stable retroviral gene delivery vector by directed evolution. Mol. Ther. 16, 308–314 (2008).
Article
CAS
PubMed
Google Scholar
Chan, K. Y. et al. Engineered AAVs for efficient noninvasive gene delivery to the central and peripheral nervous systems. Nat. Neurosci. 20, 1172–1179 (2017).
Article
CAS
PubMed
PubMed Central
Google Scholar
Bartel, M. A., Weinstein, J. R. & Schaffer, D. V. Directed evolution of novel adeno-associated viruses for therapeutic gene delivery. Gene Ther. 19, 694–700 (2012).
Article
CAS
PubMed
Google Scholar
Wang, Y. et al. Directed evolution: methodologies and applications. Chem. Rev. 121, 12384–12444 (2021).
Article
CAS
PubMed
Google Scholar
Packer, M. S. & Liu, D. R. Methods for the directed evolution of proteins. Nat. Rev. Genet. 16, 379–394 (2015).
Article
CAS
PubMed
Google Scholar
Shechner, D. M., Hacisuleyman, E., Younger, S. T. & Rinn, J. L. Multiplexable, locus-specific targeting of long RNAs with CRISPR-Display. Nat. Methods 12, 664–670 (2015).
Article
CAS
PubMed
PubMed Central
Google Scholar
Zhu, S. et al. Guide RNAs with embedded barcodes boost CRISPR-pooled screens. Genome Biol. 20, 20 (2019).
Article
PubMed
PubMed Central
Google Scholar
Richter, M. F. et al. Phage-assisted evolution of an adenine base editor with improved Cas domain compatibility and activity. Nat. Biotechnol. 38, 883–891 (2020).
Article
CAS
PubMed
PubMed Central
Google Scholar
Dobson, C. S. et al. Antigen identification and high-throughput interaction mapping by reprogramming viral entry. Nat. Methods 19, 449–460 (2022).
Article
CAS
PubMed
PubMed Central
Google Scholar
Hamilton, J. R. et al. In vivo human T cell engineering with enveloped delivery vehicles. Nat. Biotechnol. https://doi.org/10.1038/s41587-023-02085-z (2024).
Article
PubMed
Google Scholar
Qu, K. et al. Structure and architecture of immature and mature murine leukemia virus capsids. Proc. Natl Acad. Sci. USA 115, E11751–E11760 (2018).
Article
CAS
PubMed
PubMed Central
Google Scholar
Basyuk, E., Boulon, S., Skou Pedersen, F., Bertrand, E. & Vestergaard Rasmussen, S. The packaging signal of MLV is an integrated module that mediates intracellular transport of genomic RNAs. J. Mol. Biol. 354, 330–339 (2005).
Article
CAS
PubMed
Google Scholar
Frangoul, H. et al. CRISPR–Cas9 gene editing for sickle cell disease and β-thalassemia. N. Engl. J. Med. 384, 252–260 (2021).
Article
CAS
PubMed
Google Scholar
Mayuranathan, T. et al. Potent and uniform fetal hemoglobin induction via base editing. Nat. Genet. 55, 1210–1220 (2023).
Article
CAS
PubMed
PubMed Central
Google Scholar
Everette, K. A. et al. Ex vivo prime editing of patient haematopoietic stem cells rescues sickle-cell disease phenotypes after engraftment in mice. Nat. Biomed. Eng. 7, 616–628 (2023).
Article
CAS
PubMed
PubMed Central
Google Scholar
Newby, G. A. et al. Base editing of haematopoietic stem cells rescues sickle cell disease in mice. Nature 595, 295–302 (2021).
Article
CAS
PubMed
PubMed Central
Google Scholar
Ferrari, S. et al. Efficient gene editing of human long-term hematopoietic stem cells validated by clonal tracking. Nat. Biotechnol. 38, 1298–1308 (2020).
Article
CAS
PubMed
PubMed Central
Google Scholar
Frangoul, H. et al. Exagamglogene autotemcel for severe sickle cell disease. N. Engl. J. Med. 390, 1649–1662 (2024).
Article
CAS
PubMed
Google Scholar
Harris, E. Sickle cell disease approvals include first CRISPR gene editing therapy. JAMA 331, 280 (2024).
PubMed
Google Scholar
Liao, J. et al. Therapeutic adenine base editing of human hematopoietic stem cells. Nat. Commun. 14, 207 (2023).
Article
CAS
PubMed
PubMed Central
Google Scholar
Zeng, J. et al. Therapeutic base editing of human hematopoietic stem cells. Nat. Med. 26, 535–541 (2020).
Article
CAS
PubMed
PubMed Central
Google Scholar
Gorelick, R. J. et al. Characterization of the block in replication of nucleocapsid protein zinc finger mutants from moloney murine leukemia virus. J. Virol. 73, 8185–8195 (1999).
Article
CAS
PubMed
PubMed Central
Google Scholar
D’Souza, V. & Summers, M. F. Structural basis for packaging the dimeric genome of Moloney murine leukaemia virus. Nature 431, 586–590 (2004).
Article
PubMed
Google Scholar
Yu, Q. & Darlix, J. L. The zinc finger of nucleocapsid protein of Friend murine leukemia virus is critical for proviral DNA synthesis in vivo. J. Virol. 70, 5791–5798 (1996).
Article
CAS
PubMed
PubMed Central
Google Scholar
Yeager, M., Wilson-Kubalek, E. M., Weiner, S. G., Brown, P. O. & Rein, A. Supramolecular organization of immature and mature murine leukemia virus revealed by electron cryo-microscopy: implications for retroviral assembly mechanisms. Proc. Natl Acad. Sci. USA 95, 7299–7304 (1998).
Article
CAS
PubMed
PubMed Central
Google Scholar
Mattei, S., Schur, F. K. & Briggs, J. A. Retrovirus maturation—an extraordinary structural transformation. Curr. Opin. Virol. 18, 27–35 (2016).
Article
CAS
PubMed
Google Scholar
Adamson, B., Norman, T. M., Jost, M. & Weissman, J. S. Approaches to maximize sgRNA-barcode coupling in Perturb-seq screens. Preprint at bioRxiv https://doi.org/10.1101/298349 (2018).
Feldman, D., Singh, A., Garrity, A. J. & Blainey, P. C. Lentiviral co-packaging mitigates the effects of intermolecular recombination and multiple integrations in pooled genetic screens. Preprint at bioRxiv https://doi.org/10.1101/262121 (2018).
Xu, C. et al. piggyBac mediates efficient in vivo CRISPR library screening for tumorigenesis in mice. Proc. Natl Acad. Sci. USA 114, 722–727 (2017).
Article
CAS
PubMed
PubMed Central
Google Scholar
Anzalone, A. V. et al. Programmable deletion, replacement, integration and inversion of large DNA sequences with twin prime editing. Nat. Biotechnol. 40, 731–740 (2022).
Article
CAS
PubMed
Google Scholar
Pandey, S. et al. Efficient site-specific integration of large genes in mammalian cells via continuously evolved recombinases and prime editing. Nat. Biomed. Eng. https://doi.org/10.1038/s41551-024-01227-1 (2024).
Article
PubMed
Google Scholar
Yu, B. et al. Engineered cell entry links receptor biology with single-cell genomics. Cell 185, 4904–4920 (2022).
Article
CAS
PubMed
Google Scholar
Dahlman, J. E. et al. Barcoded nanoparticles for high throughput in vivo discovery of targeted therapeutics. Proc. Natl Acad. Sci. USA 114, 2060–2065 (2017).
Article
CAS
PubMed
PubMed Central
Google Scholar
Sago, C. D. et al. High-throughput in vivo screen of functional mRNA delivery identifies nanoparticles for endothelial cell gene editing. Proc. Natl Acad. Sci. USA 115, E9944–E9952 (2018).
Article
CAS
PubMed
PubMed Central
Google Scholar
Clement, K. et al. CRISPResso2 provides accurate and rapid genome editing sequence analysis. Nat. Biotechnol. 37, 224–226 (2019).
Article
CAS
PubMed
PubMed Central
Google Scholar
Chen, S., Zhou, Y., Chen, Y. & Gu, J. fastp: an ultra-fast all-in-one FASTQ preprocessor. Bioinformatics 34, i884–i890 (2018).
Article
PubMed
PubMed Central
Google Scholar
Shen, W., Le, S., Li, Y. & Hu, F. SeqKit: a cross-platform and ultrafast toolkit for FASTA/Q file manipulation. PLoS ONE 11, e0163962 (2016).
Article
PubMed
PubMed Central
Google Scholar
Download references
Acknowledgements
This work was supported by the NIH (UG3AI150551, U01AI142756, R35GM118062, RM1HG009490, and R01HL136135), Bill and Melinda Gates Foundation and Howard Hughes Medical Institute (D.R.L.). Primary human HSPCs were obtained from the Fred Hutchinson Cancer Center, with support from a Cooperative Centers of Excellence in Hematology National Institute of Diabetes and Digestive and Kidney Diseases grant (U54DK106829). Cryo-EM specimens were prepared and data were collected at the Cryo-EM Facility at MIT.nano, including use of the Talos Arctica gifted by the Arnold and Mabel Beckman Foundation. A.R. was supported by a National Science Foundation Graduate Research Fellowship. P.Z.C. is supported by Schmidt Science Fellows, in partnership with the Rhodes Trust, and a Natural Sciences and Engineering Research Council Postdoctoral Fellowship. We thank K. Mears and K. Chao for technical assistance and P. Chen, J. Davis, J. Doman, J. Queenan and P. Randolph for helpful discussions.
Author information
Author notes
Aditya Raguram
Present address: Whitehead Institute for Biomedical Research, Cambridge, MA, USA
Authors and Affiliations
Merkin Institute of Transformative Technologies in Healthcare, Broad Institute of MIT and Harvard, Cambridge, MA, USA
Aditya Raguram, Meirui An, Paul Z. Chen & David R. Liu
Department of Chemistry and Chemical Biology, Harvard University, Cambridge, MA, USA
Aditya Raguram, Meirui An, Paul Z. Chen & David R. Liu
Howard Hughes Medical Institute, Harvard University, Cambridge, MA, USA
Aditya Raguram, Meirui An, Paul Z. Chen & David R. Liu
Koch Institute for Integrative Cancer Research, Massachusetts Institute of Technology, Cambridge, MA, USA
Paul Z. Chen
Contributions
A.R. and D.R.L. conceptualized the project. A.R. designed the research, performed experiments and analyzed data. M.A. assisted with eVLP production and potency analyses. P.Z.C. performed cryo-EM analyses. A.R. and D.R.L. wrote the manuscript with input from all authors.
Corresponding authors
Ethics declarations
Competing interests
A.R. and D.R.L. have filed patent applications on this work through the Broad Institute. D.R.L. is a consultant and/or equity owner for Prime Medicine, Beam Therapeutics, Pairwise Plants, Chroma Medicine and Nvelop Therapeutics, companies that use or deliver genome editing or epigenome engineering agents. The remaining authors declare no competing interests.
Peer review
Peer review information
Nature Biotechnology thanks David Schaffer, James Davies and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Extended data Extended Data Fig. 1 Barcoded sgRNAs are compatible with eVLPs.
a, Schematic of the 15-bp barcode sequence (in red) inserted into the tetraloop of the SpCas9 sgRNA scaffold. b, Overview of the standard eVLP production workflow. c, Adenine base editing efficiencies at position A7 of the BCL11A enhancer site in HEK293T cells transduced by eVLPs produced using a four-plasmid transfection or three-plasmid (combined sgRNA+gag–ABE vector) transfection and either standard or barcoded sgRNAs as shown. d, Adenine base editing efficiencies at position A7 of the BCL11A enhancer site in HEK293T cells transduced by eVLPs produced using one of four arbitrarily selected barcoded sgRNAs. e, Fold change in eVLP-packaged sgRNA abundance measured by RT-qPCR using sgRNA-specific primers, normalized relative to sgRNA abundance in v4 eVLPs. c–e, Bars reflect the mean of n = 3 biological replicates, and dots represent individual replicate values.
Extended Data Fig. 2 Barcoded eVLP capsid library construction.
a, Schematic of the region of gag that was mutated in the eVLP capsid library. The mutated regions, underlined in purple, spanned 198 total residues of the capsid and nucleocapsid domains of gag. Each residue was mutated to each of the 19 possible non-wild-type residues at each position, resulting in a total library size of 3,762 single-residue mutants plus the canonical v4 eVLP. b, Overview of the first step of the barcoded capsid library cloning procedure. Each capsid mutant was synthesized with a 15-bp barcode on the same short oligonucleotide and cloned into an appropriate vector backbone by Gibson assembly. c, Overview of the second step of the barcoded capsid library cloning procedure. The Gibson assembly products from the first cloning step were subjected to a Golden Gate assembly reaction to install the intervening promoters and gene sequences in between the barcoded sgRNA and capsid mutant sequences. These two cloning steps were repeated independently four times to generate four sub-libraries, with each sub-library containing all capsid mutants within a 150 bp region of gag.
Extended Data Fig. 3 Barcode frequency distribution in producer-cell gDNA.
Distribution of barcode frequencies detected in producer-cell gDNA. The theoretical median of this frequency distribution, assuming perfectly equal representation of all library members, is depicted by a dotted grey line at (x=frac{1}{3763}approx 0.000266). The observed median (0.000225) is depicted by a dotted red line.
Extended Data Fig. 4 Selecting eVLP capsid mutants with improved production.
a, Schematic of the production selection. The production enrichment value for each barcode was calculated by dividing the frequency of that barcode in the eVLP-packaged sgRNAs by the frequency of that barcode in the producer-cell gDNA. Enriched barcodes identify capsid mutants that support improved production. b, Production enrichment values for all assessed capsid mutants. Each capsid mutant is shown as a single dot whose x-coordinate reflects that mutant’s enrichment in production selection replicate 1 and y-coordinate reflects that mutant’s enrichment in production selection replicate 2. The canonical capsid used in v4 eVLPs is shown as a red dot, and the corresponding enrichment values associated with this dot are shown as dotted red lines.
Extended Data Fig. 5 Selecting eVLP capsid mutants with improved transduction.
a, Schematic of the transduction selection. The transduction enrichment value for each barcode was calculated by dividing the frequency of that barcode in the successfully delivered sgRNAs by the frequency of that barcode in the eVLP-packaged sgRNAs. Enriched barcodes identify capsid mutants that support improved transduction. b, Transduction enrichment values for all assessed capsid mutants. Each capsid mutant is shown as a single dot whose x-coordinate reflects that mutant’s enrichment in transduction selection replicate 1 and y-coordinate reflects that mutant’s enrichment in transduction selection replicate 2. The canonical capsid used in v4 eVLPs is shown as a red dot, and the corresponding enrichment values associated with this dot are shown as dotted red lines.
Extended Data Fig. 6 Incorporating capsid mutations into gag–ABE or gag-pro-pol.
a, Fold change in eVLP potency relative to v4 eVLPs of each capsid mutation incorporated individually into the gag–ABE construct and paired with the wild-type MMLV gag-pro-pol. Bars reflect the mean of n = 3 biological replicates, and dots represent individual replicate values. b, Schematic of incorporating different combinations of capsid mutations into either the gag–ABE or gag-pro-pol or both.
Extended Data Fig. 7 Characterizing the production enrichment, transduction enrichment, and fold improvement in potency of eVLPs containing capsid mutations.
Each capsid mutant is shown as a single dot whose x-coordinate reflects that capsid mutant’s average production enrichment and y-coordinate reflects that capsid mutant’s average transduction enrichment (data from Fig. 2c). The color of each dot reflects the fold improvement in potency relative to v4 eVLPs (data from Fig. 3a).
Extended Data Fig. 8 Incorporating capsid mutations into Cas9- or PE-eVLPs.
a–b, Comparison of v4 Cas9-eVLPs and v5 Cas9-eVLPs (gagQ226P-pro-pol/gagC507V–Cas9) across a range of eVLP doses. Indel frequencies at BCL11A enhancer site (a) or the EMX1 site (b) in HEK293T cells are shown. c, Comparison of v3 PE2-eVLPs and v3 PE2-eVLPs containing the Q226P mutation incorporated into gag-pro-pol components across a range of eVLP doses. +1 T•A-to-A•T editing efficiencies (without indels) at the HEK3 site in HEK293T cells are shown. d, Comparison of v3b PE2-eVLPs and v3b PE2-eVLPs containing the Q226P mutation incorporated into gag-pro-pol components across a range of eVLP doses. +1 T•A-to-A•T editing efficiencies (without indels) at the HEK3 site in HEK293T cells are shown. a–d, Dots and error bars represent mean ± s.e.m. of n = 3 biological replicates. Data were fit to four-parameter logistic curves using nonlinear regression.
Extended Data Fig. 9 Comparison of v4 and v5 BE-eVLPs in primary human HSPCs across a range of eVLP doses with individual replicate values.
a, Adenine base editing efficiencies at position A4 of the BCL11A enhancer site. b, Adenine base editing efficiencies at position A7 of the BCL11A enhancer site. c, Adenine base editing efficiencies at both positions A4 and A7 of the BCL11A enhancer site. a–c, Dots represent individual replicate values for n = 3 biological replicates. Data were fit to four-parameter logistic curves using nonlinear regression. See also Fig. 4.
Extended Data Fig. 10 Characterization of v5 eVLPs.
a, Western blot analysis of lysed v4 and v5 eVLPs using an anti-Cas9 antibody. The full-length non-cleaved gag–ABE fusion is ~247 kD and the cleaved ABE is ~184 kD. Three additional cleavage products, which are generated by protease cleavage at one of three cleavage sites internal to gag (see Extended Data Fig. 2a) and still contain the ABE, were detected at intermediate molecular weights. b, Comparison of the potency of v5 eVLPs with barcoded or standard sgRNAs across a range of eVLP doses. Adenine base editing efficiencies at position A7 of the BCL11A enhancer site in HEK293T cells are shown. eVLPs were produced at a concentration of approximately 5e8 eVLPs/µL. Dots and error bars represent mean ± s.e.m. of n = 3 biological replicates. Data were fit to four-parameter logistic curves using nonlinear regression.
Source data
Supplementary information Source data Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Reprints and permissions
About this article
Cite this article
Raguram, A., An, M., Chen, P.Z. et al. Directed evolution of engineered virus-like particles with improved production and transduction efficiencies.
Nat Biotechnol (2024). https://doi.org/10.1038/s41587-024-02467-x
Download citation
Received: 14 June 2024
Accepted: 10 October 2024
Published: 13 November 2024
DOI : https://doi.org/10.1038/s41587-024-02467-x
Information contained on this page is provided by an independent third-party content provider. This website makes no warranties or representations in connection therewith. If you are affiliated with this page and would like it removed please contact editor @payson.business
