Bone-marrow-homing lipid nanoparticles for genome editing in diseased and malignant haematopoietic stem cells

Abstract

Therapeutic genome editing of haematopoietic stem cells (HSCs) would provide long-lasting treatments for multiple diseases. However, the in vivo delivery of genetic medicines to HSCs remains challenging, especially in diseased and malignant settings. Here we report on a series of bone-marrow-homing lipid nanoparticles that deliver mRNA to a broad group of at least 14 unique cell types in the bone marrow, including healthy and diseased HSCs, leukaemic stem cells, B cells, T cells, macrophages and leukaemia cells. CRISPR/Cas and base editing is achieved in a mouse model expressing human sickle cell disease phenotypes for potential foetal haemoglobin reactivation and conversion from sickle to non-sickle alleles. Bone-marrow-homing lipid nanoparticles were also able to achieve Cre-recombinase-mediated genetic deletion in bone-marrow-engrafted leukaemic stem cells and leukaemia cells. We show evidence that diverse cell types in the bone marrow niche can be edited using bone-marrow-homing lipid nanoparticles.

Haematopoietic stem cells (HSCs) differentiate into many types of immune cells and erythrocytes¹. Thus, genetic disorders in HSCs account for numerous haematopoietic diseases including sickle cell disease, β-thalassaemia, cancer and primary immune deficiencies^2–12. Although there have been reports of drug and nucleic acid delivery to the bone marrow (BM)^13–17, direct in vivo gene editor delivery to genetically disordered HSCs remains challenging due to the low frequency and quiescent nature of HSCs, the regulation of their niche^18–22 and the sensitivity of diseased and malignant stem cells to therapeutic intervention^23–26. While viral vectors have been safe and efficacious in rodents^27–31, the potential risks of leukaemogenesis arising from random insertion, off-target cytotoxicity and dose-limiting toxicities in humans could hinder further advancement^32–35. These concerns suggest a need to develop safe and effective synthetic delivery systems for genome editing in the BM.

Lipid nanoparticles (LNPs)—as a non-viral delivery technology—have demonstrated advantages of low immunogenicity and enabling re-dosing feasibility over viral vectors³⁶. This technology recently experienced further progress in the ability to enable liver and non-liver (extrahepatic) mRNA and genome editor delivery^37–45. For example, the development of selective organ targeting (SORT) LNPs, in which the addition of a defined amount of ionizable lipids, positively charged lipids or negatively charged lipids to the base-4-lipid LNP formulation, leads to exclusive mRNA delivery to the liver, lungs and spleen, respectively^37,41,44. The utilization of the SORT methodology to create LNPs for genetic medicine delivery into other therapeutically relevant extrahepatic tissues besides the lungs and spleen, however, has not yet been realized. Given the established mechanism of endogeneous targeting^38,39, we hypothesized that the incorporation of alternative biologically active molecular species could lead to novel delivery tropism.

Here we report the discovery of a series of BM-homing LNPs for the in vivo delivery of mRNA, genome and base editors to healthy, malignant and diseased HSCs. From a chemically diverse molecular library screen, we discovered that the inclusion of covalent-bond-forming lipid species in five-component LNPs created BM-homing LNPs that can transfect 14 unique cell types in the BM, including HSCs, progenitor cells, B cells, T cells, macrophages, monocytes, neutrophils and endothelial cells. Mass spectrometry proteomics analysis of serum proteins adsorbed on BM-homing LNPs revealed that BM delivery was dependent on the surface enrichment of apolipoprotein E (ApoE). More importantly, the co-delivery of Cas9 or adenine base editor (ABE) mRNA and guide RNA to a humanized mouse model for human sickle cell disease⁴⁶ resulted in the apparent gene editing of HSCs at two therapeutically relevant alleles^47,48. Cre-mediated gene deletion was successfully achieved on leukaemia and leukaemic stem cells in a mouse model for mixed-lineage leukaemia (MLL), an aggressive malignancy that is notoriously difficult to treat⁴⁹. We anticipate that this study may open up new avenues for LNP design and engineering for extrahepatic nucleic acid delivery and provide a clinical path to therapeutics for numerous haematopoietic diseases.

Discovery and development of BM-homing LNPs

Biologically active molecular species, including glycoproteins, neurotransmitters and vitamins, have been employed for targeted interactions by natural systems and artificial technologies^50–53. Starting with a chemically diverse pool of molecular species, we sought to identify those that confer a strong HSC affinity to LNPs. We incorporated 41 molecules, including 16 carbohydrates, 6 vitamins, 7 amino acids, 5 hormones, 5 neurotransmitters, 1 nucleotide and 1 covalent lipid (stearic acid N-hydroxysuccinimide ester (SA-NHS)) that has the potential to form covalent bonds with amino acid residues (Supplementary Fig. 1). We used a degradable dendrimer ionizable (pK_a < 8) cationic lipid 5A2-SC8 (ref. 54) formulation as the base-4-component formulation (hereafter referred to as LNP), which was originally optimized for the delivery of mRNA to the liver. The molar percentage of test molecules versus total lipids was fixed at 20% during the study, mirroring our prior approach to generate SORT LNPs³⁷, and mRNA-encoding firefly luciferase (Luc) was encapsulated to examine the delivery tropism of the produced protein. Although all the tested molecules showed various degrees of Luc mRNA delivery to the liver and spleen, surprisingly, we found that LNPs with SA-NHS incorporation demonstrated BM tropism, whereas all the remaining molecules did not.

We next expanded the molecular diversity of covalent lipid species to identify additional molecules with higher BM transfection efficacy (Fig. 1a). To understand if BM tropism is specific to the amine-reactive NHS ester head group or is applicable to multiple reactivity profiles, we selected a collection of lipids with different reactive functional groups for in vivo evaluation (Supplementary Fig. 3). These lipids can be divided into four categories: solely reactive to amine; solely reactive to carboxylic acid; solely reactive to thiol; crosslinking reagent. Within each category, the molecules consisted of different head-group structures and/or hydrophobic domains to ensure chemical diversity in the study.

Fig. 1 |. Discovery and development of BM-homing LNPs.

a, Schematic of LNP preparation including covalent lipid species (covalent lipids and crosslinkers). b, Addition of a covalent lipid or crosslinker to the base-4-lipid LNP formulation leads to BM mRNA delivery and genome editing in a great breadth of unique BM cell types. c,d, Bioluminescence images of dissected femurs and summary of the average bioluminescence signal intensity on dissected femurs represented for covalent lipid (c) and crosslinker (d) molecular structures. The femurs were harvested from mice 6 h after the injection of BM-homing LNPs. Data are presented as mean ± standard deviation (s.d.) (n = 3 biologically independent samples).

In the amine-reactive group, we observed a high hit rate (defined as higher than 1 × 10⁶ photons s⁻¹ cm⁻² sr⁻¹ luminescence) of 83.3% from NHS ester lipids possessing various structures of hydrophobic domains (Fig. 1c). In the thiol-reactive group, neither acrylate-containing nor acrylamide-containing lipids displayed BM transfection activity (Fig. 1c). This is probably because the hydrophobic thiol-reactive C=C double bond is buried within the LNP and not exposed on the surface. In the carboxylic-reactive group, molecules containing the carbodiimide structure, a canonical moiety for activating carboxylic acid for crosslinking with amine, displayed a high hit rate with different hydrophobic domains varying the activity to some extent (Supplementary Fig. 3 and Fig. 1c).

The crosslinker category can be further divided into four groups based on their reactivities: amine to thiol crosslinker (AT), thiol to carboxylic crosslinker (TC), carboxylic to carboxylic crosslinker (CC) and amine to amine crosslinker (AA) (Supplementary Fig. 3). The AT group displayed a high hit rate, especially those with NHS moieties on one end (Fig. 1d). The TC and CC groups only contained one molecule, but both demonstrated high activity. AA11 displayed the highest Luc mRNA delivery efficacy among crosslinkers; surprisingly, AA3 with variation only on the sulfo-NHS ester over NHS ester showed only one-tenth efficacy (Fig. 1d). This can be attributed to the increased hydrophilicity of AA3, which resulted in poor presentation on the the hydrophobic LNP surface.

BM-homing LNPs deliver mRNA to HSCs with high efficacy

To determine which BM cell types are transfected by BM-homing LNPs, we utilized genetically engineered tdTomato (tdTom) reporter mice containing a LoxP-flanked stop cassette to prevent the expression of tdTom fluorescent protein⁵⁵. Cre recombinase can delete the stop cassette and activate the tdTom expression, thereby allowing the detection of gene-edited cells (Fig. 2a). We encapsulated Cre mRNA with BM-homing LNPs, injected into tdTom mice via intravenous (IV) administration, and harvested the femur bone 72 h after the injection (Fig. 2b). Fluorescence signals of the tested formulations appeared all over the bone and were much higher than that of phosphate-buffered saline (PBS)- or LNP-treated animals (Fig. 2c). tdTom+ cells were easily distinguished using confocal imaging of BM tissue sections (Fig. 2d).

Fig. 2 |. tdTom expression in BM cells was activated by Cre mRNA BM-homing LNP delivery.

a, Schematic showing how the delivery of Cre mRNA activates tdTom expression in tdTom transgenic mice via Cre-mediated genetic deletion of the stop cassette. b, BM tdTom fluorescence was detected/quantified 72 h after the IV injection of LNPs loaded with Cre mRNA. c, In vivo evaluation of 23 BM-homing formulations in tdTom reporter mice showing the fluorescence images of dissected femur bones. d, Confocal microscopy imaging on BM slices was used to confirm the tdTom activation in BM. Scale bar, 100 μm. e, Flow cytometry was used to quantify Cre mRNA delivery efficacy in various BM cell types. Gating strategy for determining each cell type, full name of the cell sub-populations and cell surface markers are listed in detail in Supplementary Figs. 5–8 and Supplementary Table 1.

We quantified the delivery into specific cell types using flow cytometry with cells isolated from BM 72 h after Cre mRNA LNP IV administration into tdTom mice. All the tested formulations enabled Cre mRNA delivery to HSCs, progenitor cells, B cells, T cells, macrophages, monocytes, neutrophils and endothelial cells after a single injection at low dose (0.6 mg kg⁻¹) (Fig. 2e). Specifically, AA11 mediated the highest delivery efficacy in HSCs, leading to tdTom+ cells in 44.80% Lin⁻Sca-1⁺CD117⁺ (LSK) population, 40.20% long-term HSCs, 17.00–66.00% in different progenitor cells (Supplementary Fig. 5 and Supplementary Table 1), 7.19% B cells, 17.00% T cells, 32.10% macrophages, 7.55% monocytes, 22.90% neutrophils and 13.00% endothelial cells. In contrast, LNP and spleen SORT LNPs (20% molar percentage of 18PA (ref. 37)) formulations did not result in notable tdTom activation in any of the tested cell types, indicating the unique BM-transfecting capability of LNPs containing covalent lipid species (Supplementary Fig. 9). The high transfection efficacy in long-term HSCs suggests that BM-homing LNPs may be suitable for correcting inherited haematopoietic diseases. Moreover, the transfection efficacy was further enhanced by multiple doses (Supplementary Fig. 10).

Potential driving factors for BM tropism

The high delivery efficacy of BM-homing formulations to numerous BM cell types inspired us to investigate the mechanism of action. We first characterized the physical properties of BM-homing LNPs, including the hydrodynamic diameter, surface charge and mRNA encapsulation efficiency. All the formulations tested with the inclusion of molecules (Fig. 1) displayed 120–170 nm hydrodynamic diameter, <0.2 polydispersity index, near-neutral surface charge and >80% encapsulation efficiency with few outliers (Fig. 3a,b and Supplementary Table 3). These data were comparable with that of LNP, liver SORT LNPs, spleen SORT LNPs and lung SORT LNPs^37,38. Since covalent lipids and crosslinkers have the capability of forming strong interactions or even covalent bonds with certain serum proteins, we hypothesize that the composition of the protein coronas formed on the surface of BM-homing LNPs may differ from that of LNPs, thereby leading to BM delivery tropism.

Fig. 3 |. Factors that do and do not contribute to BM delivery tropism.

a,b, Hydrodynamic diameter and polydispersity index (PDI) (a) and ζ-potential (b) of selected BM-homing LNPs determined by dynamic light scattering (DLS). Data are presented as mean ± s.d. (n = 3 biologically independent samples). c, Protein composition of LNP-surface-adsorbed protein corona determined by unbiased mass spectrometry proteomics. d, Comparison of the luminescence signal intensity of selected BM-homing LNPs injected into wild-type (wt) C57BL/6 mice and ApoE knockout (ApoE^−/−) mice. Data are presented as mean ± s.d. (n = 3 biologically independent samples). *P < 0.05, **P < 0.01, ***P < 0.001 determined by a two-tailed t-test.

We selected nine BM-homing LNPs with high BM transfection efficacy and containing chemically diverse covalent lipid species for further study. We incubated these LNPs with mouse plasma, extracted bound proteins and analysed the proteins by sodium dodecyl sulfate–polyacrylamide gel electrophoresis. Unlike the spleen and lung SORT LNPs that display drastically different gel patterns, most BM-homing LNPs exhibited similar protein corona composition to that of LNP (Supplementary Fig. 14). We next employed unbiased mass spectrometry proteomics to identify and quantify which serum proteins bind most avidly to the BM-homing formulations. We found that ApoE was the top enriched serum protein in seven out of the nine BM-homing formulations (Fig. 3c and Supplementary Tables 4–13). To further study the impact of ApoE, we administered three BM-homing LNPs encapsulating Luc mRNA via IV administration to ApoE knockout (ApoE^−/−) mice and found that the luminescence signals were much lower than those in wild-type mice (Fig. 3d). Given that ApoE mediates liver delivery in multiple reports^38,39,56,57, this finding suggests the existence of a previously unknown role of ApoE in mediating LNP delivery into BM.

In vivo genome and base editing in diseased HSCs

Next, we investigated if BM-homing LNPs can achieve therapeutic genome editing in a mouse model for a human blood disease. We performed this study using Townes (HBB^S/S) mice in which the endogeneous adult α- and β-like globin genes are replaced by the orthologous human ones, with the HBB genes homozygous for the sickle cell disease mutation p.Glu6Val (ref. 46). We used a CRISPR/Cas9 gene-editing system to disrupt the BCL11A transcriptional repressor binding motif in the HBG1/HBG2 gene promoter⁴⁷, which has the potential to induce foetal haemoglobin (HbF) and reduce the morbidity and mortality of β-haemoglobinopathies (Fig. 4a). Homozygous sickle cell disease Townes (HBB^S/S) mice received two weekly IV injections of BM-homing LNPs containing 20 mol% palmitic hydrazide (C6) encapsulating Cas9 mRNA and targeting sgRNA (sgG34) (Fig. 4b). Next-generation sequencing (NGS) demonstrated that insertion and/or deletion mutations (indels) were present in 5.2% alleles in haematopoietic stem and progenitor cells (HSPCs) seven days after the final LNP administration (Fig. 4c,d).

Fig. 4 |. In vivo genome and base editing of β-globin-disorder-relevant genes in HBB^S/S Townes mice.

a, Extended β-globin locus, showing the target BCL11A binding motif in the promoters of the genes encoding γ-globin. BCL11A binding motif is represented by a red line and sgG34 is represented by a green line. b, HBB^S/S Townes mice received two weekly IV injections of BM-homing LNPs encapsulating Cas9 mRNA and sgG34 or ABE8e_NRCH mRNA and sgHBB (n = 3; the total RNA dose per injection was 3 mg kg⁻¹). BM samples were harvested and analysed seven days after the final injection. c, Insertions and deletions (indels) detected in CD117+ cells isolated from the BM of HBB^S/S Townes mice. Data are presented as mean ± s.d. (n = 3 biologically independent samples). ***P < 0.001 determined by a two-tailed t-test. d, Representative NGS reads of a sample treated with LNPs (encapsulating Cas9 mRNA and sgG34). e, Edited region of HBB with target A at protospacer position 7 shown in orange and bystander edit in blue (silent). sgHBB is represented by a purple line. f, A–G conversion detected in CD117+ cells isolated from the BM of HBB^S/S Townes mice. Data are presented as mean ± s.d. (n = 3 biologically independent samples). **P < 0.01 determined by a two-tailed t-test. g, Representative NGS reads of a sample treated with LNPs (encapsulating ABE8e_NRCH mRNA and sgHBB).

Although promising in preclinical and early clinical studies, treating β-globin disorders with nuclease genome-editing strategies is not without risk, as nucleases have been shown to induce DNA damage responses, cause loss of chromosome arms and lead to variability in HbF induction outcomes^58–62. Therefore, precision genome-editing strategies such as base editing and prime editing⁶³ may possess advantages over nuclease delivery. We used ABE8e_NRCH (refs. 64,65), an ABE designed to convert the sickle cell disease allele to the non-pathogenic Makassar allele (Fig. 4e). Similar to the CRISPR/Cas study, HBB^S/S mice received two weekly IV injections of BM-homing LNPs containing 20 mol% palmitic hydrazide (C6) encapsulating ABE8e_NRCH mRNA and sgRNA (sgHBB) targeting the sickle cell disease mutation. NGS demonstrated that 2.43% of the sickle cell disease alleles were converted into the Makassar allele along with 2.25% silent editing at the A9 position in HSPCs seven days after the final LNP administration (Fig. 4f,g). To further validate the therapeutic potential of base editing, we isolated CD117⁺ cells from the BM of Townes mice, base edited them by C6-LNPs co-delivery of ABE8e_NRCH mRNA and sgHBB, and treated differentiated erythrocytes under the hypoxic condition to study the phenotypic changes. Encouragingly, we observed that base editing on SCD CD117⁺ cells reduced sickling frequency in differentiated erythrocytes to 25% that of the untreated group, confirming that HBB^S to HBB^G conversion reduced sickling (Supplementary Fig. 15). It is noteworthy that this result is in line with a recent study using antibody-functionalized LNPs⁶⁶. Together, our results represent an exciting non-viral delivery approach for the in vivo genome editing of globin gene loci using therapeutic strategies to correct β-haemoglobinopathies.

In vivo mRNA delivery to leukaemic animals

Our studies established the capability of BM-homing LNPs to transfect HSCs for in vivo genome editing at disease-relevant genes. Next, we sought to investigate whether BM-homing LNPs could enable mRNA delivery to malignant cell types. We used an MLL-fusion-driven acute myeloid leukaemia (AML) model to evaluate the transfection efficacy of BM-homing LNPs to leukaemic cells. The t(9;11) (p22;q23) reciprocal translocation results in the expression of the MLL-AF9 fusion gene and myelo-monoblastic AML, which is typically associated with extramedullary tumour infiltration, resistance to chemotherapy, frequent relapses and poor survival^49,67,68.

To induce AML, an MLL-AF9-IRES-YFP-encoding plasmid^69–71 was used for the lentiviral transduction of Lin⁻ cells isolated from tdTom reporter mice (Fig. 5a). The transduced cells were then transplanted into lethally irradiated recipient mice and successful leukaemia development was confirmed by the detection of YFP⁺ cells in the peripheral blood three weeks after transplantation (Supplementary Fig. 17). Next, to confirm the integrity of loxP–tdTom reporter cassette and the capability of LNPs in transfecting malignant cells, we isolated BM-engrafted leukaemia cells, incubated with LNPs encapsulating Cre mRNA and determined the activation of tdTom expression 24 h later using confocal microscopy. We observed that 28% cells showed strong tdTom expression on low LNP dose treatment (100 ng mRNA per well) and the number increased to 63% on higher dose treatment (200 ng mRNA per well) (Fig. 5c,d). The Cre-mediated recombination was also confirmed by polymerase chain reaction (PCR), as only the treated samples showed the expected product size of 185 base pairs in the edited DNA, whereas the control sample yielded only one band of 1,056 base pairs on an agarose gel (Fig. 5e).

Fig. 5 |. tdTom expression in MLL-AF9-driven AML model was activated by BM-homing LNP-mediated editing.

a, The MLL-AF9-IRES-YFP gene was installed into the genome of Lin⁻ cells extracted from the foetal liver of tdTom reporter mice and the cells were incubated with LNPs containing Cre mRNA in a 24-well plate for 24 h. b, Delivery of Cre mRNA activates tdTom expression in tdTom transgenic mice via the Cre-mediated genetic deletion of the stop cassette. c, Summary of the percentage of tdTom+ cells measured from the confocal microscopy images of the edited MLL-AF9-IRES-YFP Lin⁻ cells. Data are presented as mean ± s.d. (n = 3 biologically independent samples). d, Representative confocal microscopy images of the control and LNP-treated cells. Scale bar, 100 μm. e, DNA agarose gel of the PCR amplicon performed from the genomic DNA extracted from the control and LNP-treated cells with primers flanking the Ai14 locus (n = 3 biologically independent samples). f, C57BL/6 recipient mice were lethally irradiated and received IV transplantation of MLL-AF9-transfected Lin⁻ cells. Three weeks after the transplantation, the BM and spleen were extracted from the primary transplant recipient animal and the isolated cells were transplanted into a secondary transplant recipient to establish the model for LNP study. Three weeks after the secondary transplantation, Cre mRNA BM-homing LNPs were injected via IV administration. Leukaemic cells were harvested and analysed 72 h after LNP injection. g, Percentage of tdTom+ cells after LNP and BM-homing LNP injection on bone-engrafted leukaemia animals. BM-isolated leukaemic cells, leukaemic stem cells residing in BM (BM LSCs) and spleen-isolated leukaemic cells are characterized by flow cytometry. Data are presented as mean ± s.d. (n = 3 biologically independent samples).

Next, we sought to determine if BM-homing LNPs could transfect this leukaemic model in vivo. To mimic the clinical aggressiveness of MLL-AF9-driven AML, all the model mice were generated from secondary transplantation (Fig. 5f), in which mice barely survive for more than 60 days on transplantation and the median lifespan of the model mice is more than a month shorter than those generated from primary transplantation⁷². Cre mRNA encapsulating BM-homing LNPs were injected via IV administration. The BM and spleen were harvested 72 h post-injection for flow cytometry analysis (Fig. 5f). A single, low-dose (0.6 mg kg⁻¹) injection of tested BM-homing LNPs led to tdTom expression activation in 13–18% BM leukaemic stem cells (defined as the YFP⁺CD11b⁺CD117⁺ population) (Fig. 5g), the cell type that is believed to account for drug resistance and leukaemia relapse⁷³, and in 2–4% BM- or spleen-residing leukaemic cells. Importantly, the injection of control Cre mRNA LNPs (no covalent lipid added) at the same dose led to less than 0.3% tdTom+ cells in all the determined cell types (Fig. 5g), which further indicates the unique BM-homing capability of covalent lipid species even under malignant settings. Overall, these results demonstrate the potential application of therapeutic genetic medicine delivery to treat aggressive leukaemia through non-invasive, less-toxic LNP strategy.

Discussion

In this paper, we report a series of BM-homing LNPs that enabled mRNA, CRISPR/Cas9 and base editor delivery to a breadth of at least 14 cell types in BM, especially HSCs, in healthy reporter mice, mouse models expressing sickle cell disease phenotypes and an aggressive acute myeloid leukaemia model. We unexpectedly discovered that covalent lipid species enable BM delivery tropism when incorporated into base-4-lipid LNP formulations. The delivery efficacy can be related to the structure and activity of both head-group and hydrophobic domains. We also demonstrate the therapeutic potential of BM-homing LNPs to achieve genome editing in a β-globin disorder model and an aggressive AML model. Our results on the HBB^S/S Townes model represent an exciting non-viral strategy for the in vivo gene therapy of β-globin disorder diseases. It is worth noting that there is a lack of therapeutic studies of β-globin disorders and MLL-AF9-driven AML by the in vivo administration of genetic medicine, and due to the severity of the sickle cell disease phenotype (severe anaemia and splenomegaly) and the aggressiveness of the MLL-AF9-driven AML model (mice barely survive for more than 60 days after transplantation), we focused here on demonstrating proof-of-concept genome editing. Since LNPs are a modular platform that can be repeatedly administered and the broad applicability of the BM-homing capability of covalent lipid species on the most clinically advanced four-component LNP systems, it holds a vast opportunity to reach therapeutically relevant editing levels following the further optimization of formulation engineering, dosing regimen and stabilizing nucleic acid modifications. Since BM-homing LNPs can transfect and edit new cell types in the BM niche, including HSCs and cancer cells, we anticipate further expansion and utilization of BM-homing LNPs for the treatment of haematopoietic diseases.

Methods

Nanoparticle formulation

RNA-loaded LNP formulations were formed using the ethanol dilution method based on or modified from previous publications^{37,38,41,43–45}. For all the ex vivo assays, the characterization of physiochemical properties and in vivo Luc mRNA delivery studies, the ratio of the lipid to total RNA weight is fixed at 40/1. For all the in vivo Cre mRNA delivery studies, the ratio of the lipid to total RNA weight was fixed at 20/1. For all the Cas9 mRNA/sgG34 and ABE8e_NRCH mRNA/sgHBB delivery studies, the ratio of the lipid to total RNA weight was fixed at 10/1. In this work, we incorporated covalent lipid species or crosslinkers as a supplemental lipid to conventional base-4-lipid formulations to achieve BM delivery tropism. The base-4-lipid formulation contains 5A2-SC8 (as the ionizable lipid), DOPE (as the phospholipid), cholesterol and DMG-PEG2000 at a fixed molar ratio of 15:15:30:3. We incorporated 20 mol% of covalent lipid/crosslinker as a proportion of the total lipids in the study. To prepare the LNPs, a mixture of 23.8 mol% 5A2-SC8, 23.8 mol% DOPE, 47.6 mol% cholesterol and 4.8 mol% DMG-PEG2K was dissolved in ethanol and then mixed with mRNA and diluted in 10 mM, pH 4.0 citrate buffer at a volume ratio of 3:1 (mRNA:lipids). To prepare BM-homing LNPs, a mixture of 19 mol% 5A2-SC8, 19 mol% DOPE, 38 mol% cholesterol, 4 mol% DMG-PEG2K and 20 mol% covalent lipid species (covalent lipid or crosslinker) was dissolved in ethanol and then mixed with mRNA and diluted in 10 mM, pH 4.0 citrate buffer at a volume ratio of 3:1 (mRNA:lipids). After 10 min of mixing the mRNA and lipid solutions, the LNP formulations were diluted with 1× PBS to 0.5 ng μl⁻¹ mRNA for in vitro assays and characterization of physicochemical properties. For in vivo experiments, the formulations were dialysed (Pur-A-Lyzer Midi Dialysis Kits, Sigma-Aldrich; MWCO 3.5 kDa) against 1× PBS for 2 h. Afterwards, the LNPs were diluted with PBS to a final volume of 200 μl per mouse for IV injections.

Characterization of mRNA encapsulating nanoparticles

Hydrodynamic diameter, polydispersity index and ζ-potential were measured using dynamic light scattering (Malvern MicroV model; He–Ne laser, λ = 632 nm) based on or modified from a previous publication³⁸. The encapsulation efficiency of mRNA in each LNP was quantified by measuring the mRNA binding following the Quant-iT RiboGreen assay protocols.

mRNA synthesis

Luc, Cre recombinase, Cy5-labelled Cre recombinase, Cas9 and ABE8e_NRCH mRNAs were produced using in vitro transcription based on or modified from a previous publication³⁷. Briefly, the coding fragments of each protein were prepared using a PCR program (Supplementary Table 15). Then, these fragments were cloned into pCS2+MT vectors with optimized 5′(3′)-untranslated regions and poly-A sequences. In vitro transcription reactions were performed following standard protocols but with N1-methylpseudouridine-5′-triphosphate replacing the typical uridine triphosphate for Luc, Cre recombinase, Cas9 and ABE8e_NRCH, as well as with Cy5-uridine triphosphate (APExBIO, B8333) replacing 25% of typical uridine triphosphate for Cy5-labelled Cre recombinase. Finally, the mRNA was capped (Cap-1) using the ScriptCap system (CellScript). The coding sequences for these proteins are detailed in the Supplementary Materials and Methods.

Animal experiments

All the animal experiments were approved by the Institution Animal Care and Use Committees of The University of Texas Southwestern Medical Center and were consistent with local, state and federal regulations as applicable. C57BL/6 mice were obtained from Charles River. B6.Cg-Gt(ROSA)26Sor^{tm14(CAG-tdTom)Hze}/J mice (also known as Ai14 or Ai14(RCL-tdT) mice) were obtained from the Jackson Laboratory (007914) and bred to maintain the homozygous expression of the Cre reporter allele. B6;129-Hbb^{tm2(HBG1,HBB*)Tow}/Hbb^{tm3(HBG1,HBB)Tow} Hba^tm1(HBA)Tow/J mice (also known as the Townes model) were obtained from the Jackson Laboratory (013071). B6.129P2-Apoe^tm1Unc/J mice were acquired from the Jackson Laboratory (002052).

In vivo Luc mRNA delivery

C57BL/6 mice with weights of 18–20 g were injected via IV administration with various Luc mRNA formulations: n = 3 per group. Then, 6 h post-injection, mice were injected with d-luciferin (150 mg kg⁻¹, intraperitoneal). The femur bone was dissected and imaged and the luminescence intensity was quantified using Living Image 4.1 software (PerkinElmer).

Gene editing (Cre mRNA) in tdTom reporter mice

Cre mRNA formulations were prepared as described above and IV injections were performed (0.6 mg kg⁻¹ Cre mRNA). After 72 h, mice (n = 3 per group for selected formulations) were euthanized and femurs were dissected and imaged using an IVIS Lumina system (PerkinElmer).

Cell isolation and staining for flow cytometry

To isolate the BM cells for the analysis of tdTom+ cells and genomic DNA extraction, femur and tibia bones were collected and BM cells were isolated via centrifugation (1,000g, 2 min, 4 °C). Peripheral blood was collected in EDTA-coated tubes to prevent clotting. Spleen was collected in cold PBS. Isolated cells passed through a 100 μm cell strainer (BD Biosciences, 352360) to obtain a single-cell suspension and were treated with 1× red blood cell lysis buffer (BioLegend, 420301) for 5 min (BM and spleen samples) or 10 min (peripheral blood samples) on ice. The red blood cell lysis buffer was neutralized by adding twice the volume of a cell-staining buffer (BioLegend, 420201). Then, 1 × 10⁶ cells were stained with fluorescent labelled anti-mouse antibodies at 1:100 dilution for 20–30 min on ice. Live Dead Aqua (Thermo Fisher, L34957) was used to distinguish live cells for all the samples. The BM-HSPC panel was stained with FITC lineage cocktail (BioLegend, 133301), PerCP-Cy5.5 Sca-1 antibody (Thermo Fisher, 45–5981-80), Alexa Fluor 700 CD117 antibody (Thermo Fisher, 56–1172-80), PE/Cy7 CD34 antibody (BioLegend, 128617), APC CD135 antibody (BioLegend, 135309) and Brilliant Violet 711 CD16/32 antibody (BioLegend, 101337). BM-B cell panel was stained with FITC CD3 antibody (BioLegend, 100203), FITC Ly-6G antibody (BioLegend, 108405), FITC CD11b antibody (BioLegend, 101205), FITC TER-119 antibody (BioLegend, 116205), APC B220 antibody (BioLegend, 103211) and PerCP CD19 antibody (BioLegend, 115531). BM-T cell panel was stained with FITC CD3 antibody, Alexa Fluor 700 CD8a antibody (Thermo Fisher, 56–0081-80) and PerCP CD4 antibody (BioLegend, 100431). BM-macrophage, BM-monocyte and BM-neutrophil panels were stained with APC Ly-6G antibody (BioLegend, 127613), FITC CD11b antibody and Alexa Fluor 594 F4/80 antibody (BioLegend, 123140), respectively. The leukaemic cell panel was stained with APC-Cy7 CD11b antibody (BioLegend, 101225) and Alexa Fluor 700 CD117 antibody. Cells were washed twice with a cell-staining buffer to remove excess antibodies and resuspended in 500 μl cell-staining buffer. The cells were kept on ice until analysed by LSRFortessa flow cytometer (BD Biosciences).

For genomic DNA extraction, isolated BM cells were treated with the red blood cell lysis buffer for 5 min on ice and then incubated with Biotin CD117 antibody (BioLegend, 135129) at 1:100 dilution for 20–30 min on ice. Cells were washed twice with PBS and incubated on an EasySep magnetic stand (STEMCELL Technologies, 18000) for 5 min. Free cell suspension was removed and the attached cells were washed with PBS for three times. Whole BM cells and CD117+ cells were resuspended in 50 μl of 1× passive lysis buffer (Promega) together with 2 μl of proteinase K (Thermo Fisher). Afterwards, a lysis PCR program (65 °C for 15 min, 95 °C for 10 min) was run to obtain cell lysates. The targeted genomic loci were then amplified using the following PCR amplification program (95 °C for 3 min; 95 °C for 30 s, 61 °C for 20 s and 72 °C for 30 s for 30 cycles; 72 °C for 2 min and then maintained at 4 °C). Cell lysates were used as DNA templates.

Isolation of plasma proteins adsorbed on LNPs

A procedure published in a previous work was adapted for the current study³⁸. LNPs were prepared according to the previously described method and were diluted to a final lipid concentration of 1 g l⁻¹ with 1× PBS. Mouse plasma was added to each LNP solution at a 1:1 volume ratio and incubated for 15 min at 37 °C. A 0.7 M sucrose solution was prepared by dissolving solid sucrose in MilliQ water. The LNP/plasma mixture was loaded onto a 0.7 M sucrose cushion of equal volume to the mixture and centrifuged at 15,300g and 4 °C for 1 h. The supernatant was removed, and the pellet was washed with 1× PBS. Next, the pellet was centrifuged at 15,300g and 4 °C for 5 min, and the supernatant was removed. Washing was performed twice more for a total of three washes. Following the final wash, the pellet was resuspended in 2 wt% sodium dodecyl sulfate. Excess lipids were removed from each sample by following the protocol provided with the ReadyPrep 2-D Cleanup (Bio-Rad). The resulting pellet from the cleanup step was resuspended in 2× Laemmli buffer.

Sodium dodecyl sulfate–polyacrylamide gel electrophoresis characterization of plasma proteins

A procedure published in a previous work was adapted for the current study³⁸. Plasma proteins isolated from the surface of LNPs were loaded onto a 4–20% Mini-PROTEAN TGX Precast Protein Gel at a volume of 10 μl and separated at 200 V. The proteins were visualized by staining the gel for 1 h with SimplyBlue SafeStain. The gel was destained using deionized water overnight and imaged with a LI-COR scanner the following day.

Preparation of plasma protein samples for mass spectrometry

A procedure published in a previous work was adapted for the current study³⁸. Plasma proteins isolated from the surface of the LNPs were loaded onto a 4–20% Mini-PROTEAN TGX Precast Protein Gel at a volume of 10 μl and run into the gel (1 cm) at 90 V. The gel was stained with SimplyBlue SafeStain for 1 h to fix and visualize the proteins. After destaining for 1 h, the protein bands were excised using a sterile razor blade and sliced into 1 mm³ cubes. The cubes were added to a 1.5 ml tube that had been rinsed with 1:1 MilliQ water:ethanol and stored at 4 °C until being submitted to the University of Texas Southwestern Proteomics Core for mass spectrometry analysis. The samples were submitted to the UT Southwestern Proteomics Core for analysis using a Thermo Q Exactive HF mass spectrometer to identify the protein corona constituents.

ApoE knockout mice experiments

LNPs were prepared according to a previously described method. B6.129P2-Apoetm1Unc/J mice with body weight of 18–20 g received an IV administration of BM-homing LNPs at a dosage of 0.1 mg kg⁻¹ Luc mRNA (n = 3). As a comparison, C57BL/6 mice weighing 18–20 g received the same IV dosage of BM-homing LNPs (n = 3). After 6 h, mice received an intraperitoneal injection of d-luciferin (150 mg kg⁻¹) and imaged by an IVIS Lumina system (PerkinElmer). Femur bone was dissected and imaged and the luminescence intensity was quantified using Living Image software (PerkinElmer).

Creation of MLL-AF9-driven leukaemia model

MLL-AF9-expressing retroviruses were produced in 293 T cells using an MSCV-MLL-AF9-IRES-YFP-encoding plasmid, as previously described^69–71, and used for infection with Lin⁻ foetal liver cells isolated from tdTom reporter mice after two rounds of spinoculation in the presence of 4 μg ml⁻¹ polybrene. The cells were then injected via IV administration into lethally irradiated C57BL6 mice and peripheral blood of injected mice were collected three weeks post-transplantation to check the YFP+ cell population and confirm the establishment of the leukaemia model. BM and spleen cells from the primary transplant recipients were further injected into lethally irradiated C57BL6 mice for the secondary transplantation. All the model mice used in this study were generated from secondary transplantation.

In vitro gene editing (Cre mRNA) on the leukaemia cells

On infection by MLL-AF9-expressing retroviruses, cells were seeded in a 24-well plate at a concentration of 1 × 10⁶ cells ml⁻¹. Four hours after seeding, LNPs (100 or 200 ng mRNA per well) were added and incubated for 24 h. Cells were then collected by centrifugation and mounted on a cover slip for confocal microscopy imaging or lysed for genomic DNA extraction for PCR.

Gene editing (Cre mRNA) in the MLL-AF9-driven leukaemia model

Once AML establishment was confirmed on the transplant recipients, Cre mRNA formulations were prepared as described above and IV injections were administered (0.6 mg kg⁻¹ Cre mRNA). After 72 h, mice (n = 3 per group) were euthanized. Femur and tibia bones and spleen were collected for flow cytometry analysis.

Gene editing (Cas9 mRNA/sgG34 and ABE8e_NRCH mRNA/sgHBB) in HBB^S/S mice

HBB^S/S female mice were injected with BM-homing formulations via IV administration for the co-delivery of Cas9 mRNA and modified sgG34 or ABE8e_NRCH mRNA and modified sgHBB (Supplementary Table 14) at a total dose of 3 mg kg⁻¹ (2/1, mRNA/sgRNA, wt/wt) (n = 3 per group). All the mice received two weekly injections. Seven days after the last injection, the liver, spleen, reproductive organ, femur and tibia bones were collected and the genomic DNA were collected following the procedure described above.

Targeted amplicon deep-sequencing analysis

Deep amplicon sequencing was used to measure the genome and base-editing efficiency in the genomic DNA described above. HBG1 loci or human β-globin (HBB) loci containing sickle cell mutation was PCR amplified with primers listed in Supplementary Table 16 with the addition of 8 bp barcodes on both ends. Then, 40 ng of genomic DNA was used for the PCR reaction using a Phusion U Green Multiple PCR Master Mix (Thermo Fisher, F564S) with an amplification program (98 °C for 3 min; 98 °C for 10 s, 61 °C for 30 s and 72 °C for 30 s for 30 cycles; 72 °C for 2 min). PCR products were purified with a PCR purification kit (QIAGEN, 28106) with 25 μl of DNase-free water and quantified by Qubit dsDNA high-sensitivity assay (Invitrogen, Q33231). A targeted amplicon deep-sequencing library was then prepared and later sequenced by Novogene using Illumina NovaSeq 6000. After demultiplexing, the amplicon sequencing data were analysed with CRISPResso2 (https://crispresso.pinellolab.partners.org/) to determine the editing efficiency⁷⁴.

Display items

The images of mice and syringes (Figs. 1b, 2a, 4b,d,e,g and 5a,f) were created with BioRender.com.

Data Availability

DNA sequencing files can be accessed at the National Center for Biotechnology Information Sequence Read Archive (NCBI SRA) with accession code PRJNA1082713. Source data are provided with this paper. All other data are available from the corresponding author upon reasonable request.