A brain cell-specific mRNA and CRISPR lipid nanoparticle and its synergistic reverse design method and system

Abstract

This invention provides a brain cell-specific mRNA and CRISPR lipid nanoparticles, along with a synergistic reverse design method and system. The lipid nanoparticles comprise a lipid nanoparticle core, a nucleic acid payload encapsulated within the core, and a multi-ligand assembly disposed on the particle surface. This includes a first blood-brain barrier ligand for endothelial capture in brain microvessels, a second co-synergistic transcytosis ligand for regulating endocytosis sorting pathways, and a brain cell-specific ligand shielded by an environmentally responsive shielding layer. This invention constructs a digital twin of brain microvessels coupling blood flow shearing, protein corona dynamics, transcellular transport, and intracellular effects. By combining a generative peptide design model with multifidelity active learning, it achieves joint optimization of lipid composition, particle parameters, ligand combinations, and nucleic acid payload structure. This invention significantly improves the efficiency of nucleic acid payloads crossing the blood-brain barrier and increases brain cell selectivity, while reducing peripheral off-target effects and endothelial retention. It is suitable for precision gene therapy and research reagent development for brain diseases.

Description

Technical Field

[0001] This invention relates to a brain cell-specific mRNA and CRISPR lipid nanoparticles and their synergistic reverse design method and system, belonging to the field of nucleic acid drug delivery and artificial intelligence-assisted nanomedicine design technology. Background Technology

[0002] Lipid nanoparticles, with their excellent nucleic acid encapsulation and delivery efficiency, have become a core carrier technology for mRNA vaccines and liver-targeted gene drugs. To intervene in central nervous system diseases, current technologies generally employ receptor-mediated transcytosis strategies, attempting to cross the blood-brain barrier by coupling targeting ligands (such as transferrin receptor antibodies, Angiopep-2 peptides, etc.) to the surface of lipid nanoparticles (LNPs). However, single-ligand strategies have inherent biological limitations: high-affinity ligands often lead to excessive receptor aggregation on the surface of brain microvascular endothelial cells, causing particles to be trapped in lysosomal degradation pathways rather than effectively transported across the endothelium; simply increasing ligand density exacerbates clearance by the peripheral mononuclear phagocytic system and the protein crown masking effect. Furthermore, blood flow shear forces, heterogeneity of receptor expression inside and outside blood vessels, and complex transcellular transport dynamics make it difficult to replicate efficient brain delivery effects in vivo with traditionally statically cultured and screened targeting LNPs; most formulations are intercepted by the liver and spleen before reaching the brain parenchyma. Another core challenge facing existing brain-targeting delivery systems lies in the failure to resolve the spatiotemporal contradiction between "barrier crossing" and "substance targeting." The blood-brain barrier is a dynamic physiological structure composed of endothelial cells, the basement membrane, pericytes, and astrocyte foot processes. The crossing of this barrier by LNPs to achieve brain cell-specific delivery is a multi-step, non-linear cascade process. Current technologies often conflate blood-brain barrier penetration with brain cell targeting, lacking a systematic consideration of the temporal conflicts between these steps. For example, ligands suitable for capturing endothelium at the vascular lumen often carry a high positive charge or large molecular weight, which hinders particle diffusion into the brain parenchyma after crossing the endothelium; while ligands suitable for binding to neurons or glial cells, if exposed too early before crossing the barrier, can cause non-specific adsorption or endothelial retention. Furthermore, known brain-penetrating peptide sequences are spatially narrow and generally suffer from poor serum stability, high risk of hemolysis, or only promote endocytosis without promoting basal-side extrusion, limiting their clinical application potential. More critically, existing technologies lack design tools capable of integrating multi-scale physiological parameters and performing systematic optimization. Brain microvascular transcytosis is regulated by a combination of factors, including particle physicochemical properties, dynamic evolution of protein coronas, receptor circulation rates, and tight intercellular junctions. Traditional trial-and-error methods struggle to analyze the interactions of these complex variables within a unified framework. Static cell experiments or Transwell models cannot simulate in vivo hemodynamics and complete transcellular pathways, resulting in a significant translational gap between in vitro data and in vivo efficacy. While AI-aided design has shown promise in drug discovery, the lack of dedicated algorithms and closed-loop validation systems for brain microvascular transport mechanisms makes it difficult to directly translate generated candidate molecules into manufacturable and verifiable clinical-grade LNP formulations. Therefore, a novel solution integrating multi-ligand temporal synergistic mechanisms, cross-scale digital simulation, and intelligent experimental design is urgently needed to overcome the technical bottlenecks in brain cell-specific nucleic acid delivery. To this end, a brain cell-specific mRNA and CRISPR lipid nanoparticles and their synergistic reverse design method and system are proposed. Summary of the Invention

[0003] In view of this, the present invention provides a brain cell-specific mRNA and CRISPR lipid nanoparticles and their synergistic reverse design method and system to solve or alleviate the technical problems existing in the prior art, and at least provide a beneficial option.

[0004] The technical solution of this invention is implemented as follows: a brain cell-specific mRNA and CRISPR lipid nanoparticle, comprising a lipid nanoparticle core, a nucleic acid payload encapsulated in the lipid nanoparticle core, and a multi-ligand assembly disposed on the surface of the lipid nanoparticle; the multi-ligand assembly comprises at least: First blood-brain barrier ligand for binding to the first transcytosis-associated receptor on brain microvascular endothelial cells; A second co-transcytosis ligand for binding to a second transcytosis-associated receptor that is different from the first transcytosis-associated receptor, a non-orthogonal site of the first transcytosis-associated receptor, or an endocytosis sorting-associated membrane structure; and a brain cell-specific ligand for binding to target brain cell surface molecules and shielded by a switchable shielding layer; The surface density, relative ratio, spatial spacing, or binding affinity of the first blood-brain barrier ligand and the second co-transcytosis ligand are configured to enable the lipid nanoparticles to form a co-transcytosis state in brain microvascular endothelial cells where non-lysosomal sorting is dominant. The switchable shielding layer is configured to maintain a shielding state in the luminal blood flow environment and to release the shielding in response to environmental changes after the lipid nanoparticles cross the brain microvascular endothelium, thereby exposing the brain cell-specific ligands and promoting the entry of the nucleic acid payload into the target brain cells.

[0005] Preferably, the lipid nanoparticle core is composed of ionizable lipids, auxiliary phospholipids, sterols, and polyethylene glycol lipids, wherein ionizable lipids account for 20-70 mol% of the total lipids, auxiliary phospholipids account for 2-30 mol%, sterols account for 10-55 mol%, and polyethylene glycol lipids account for 0.1-10 mol%; the hydrated particle size of the lipid nanoparticles is 35-180 nm, the polydispersity index is not greater than 0.35, and the apparent pKa is 5.5-7.5.

[0006] Preferably, the second co-transcytosis ligand is configured to produce at least one of the following effects: reduce the probability of receptor clustering and lysosomal sorting caused by the first blood-brain barrier ligand; increase the receptor circulation rate or the probability of basolateral efflux; improve effective adhesion under flow shear conditions; or alter the endocytic pathway to a non-degradation pathway by binding to receptor transmembrane regions, non-orthogonal sites, lipid rafts, or caveolin-related structures.

[0007] Preferably, the switchable shielding layer is connected to the lipid nanoparticles via an environmentally responsive linker arm, wherein the environmentally responsive linker arm is selected from hydrolytic bonds sensitive to pH decrease, disulfide bonds sensitive to redox states, peptides sensitive to proteases, ester bonds sensitive to lipases, oxidation-sensitive bonds sensitive to reactive oxygen species, cleavage bonds sensitive to shear stress, or molecular switches sensitive to conformational changes after receptor-mediated endocytosis.

[0008] Preferably, at least one of the first blood-brain barrier ligand, the second co-transcytosis ligand, or the brain cell-specific ligand is a generative targeting peptide with a length of 5 to 80 amino acids, and the generative targeting peptide satisfies at least four of the following constraints: receptor binding, pH-dependent dissociation, protease stability, low aggregation, low hemolysis, low immunogenicity, and feasibility of solid-phase synthesis.

[0009] Preferably, the nucleic acid payload comprises mRNA encoding a therapeutic protein, and / or a CRISPR payload comprising Cas nuclease mRNA, base editor mRNA, guide RNA, or a combination thereof; the mRNA contains a non-target cell microRNA response element, and the CRISPR payload comprises chemically shielded guide RNA, aptamer-gated guide RNA, or protease-responsive editor to form dual selectivity of extracellular targeting and intracellular expression or editing gating.

[0010] A collaborative reverse design method for the lipid nanoparticles includes the following steps: A digital twin of brain microvessels is established, which couples blood flow shearing, protein corona dynamics, receptor binding and clustering, endocytosis sorting, transendothelial transport, basolateral release, brain parenchyma diffusion, target brain cell uptake, endosome escape, and nucleic acid expression or editing effects. Generative peptide design models are used to generate candidate peptide sequences, and multi-level constraint screening is used to obtain target peptides that meet the requirements of binding affinity, stability and safety. Lipid composition, particle physicochemical parameters, target peptide sequence, ligand surface density, ligand ratio, linker arm length, shielding layer response parameters, and nucleic acid payload structure are encoded as joint design variables; The brain microvascular digital twin was invoked to predict the intrabrain delivery efficiency, target brain cell selectivity, peripheral organ exposure, endothelial retention and safety indicators corresponding to each joint design variable. A Pareto optimal solution set satisfying preset delivery efficiency and safety constraints is obtained through a multi-objective optimization algorithm, outputting the lipid nanoparticle formulation. Preferably, it further includes a multi-fidelity active learning step: based on the uncertainty quantification results of candidate designs, expected performance improvement, experimental costs, and animal burden, a data acquisition function is constructed to adaptively select the next round of verification tasks from a multi-fidelity verification hierarchy consisting of molecular simulation, mesoscopic simulation, brain microvascular chip, brain organoids or co-culture, ex vivo tissue, and animal verification; and the verification results are fed back into the brain microvascular digital twin and generative peptide design model to update model parameters and reduce prediction uncertainty. A method for preparing the lipid nanoparticles includes the following steps: An organic phase containing ionizable lipids, cofactor phospholipids, sterols, and polyethylene glycol lipids is microfluidically mixed with an aqueous phase containing nucleic acid loads to form a lipid nanoparticle core. A first blood-brain barrier ligand and a second co-transcytosis ligand are disposed on the lipid nanoparticle core by co-assembly, post-insertion, or covalent coupling. The switchable shielding layer is connected to a brain cell-specific ligand via an environmentally responsive connecting arm, and the complex is then assembled onto the surface of the lipid nanoparticles. The obtained lipid nanoparticles were purified, concentrated, and replaced with buffer solution. The product undergoes quality control, which includes at least the determination of particle size, polydispersity index, surface potential, apparent pKa, nucleic acid encapsulation efficiency, ligand density, shielding layer release kinetics, receptor binding capacity, transendothelial transport flux, and target brain cell uptake efficiency. The lipid nanoparticles are used in the preparation of pharmaceutical formulations, gene therapy products, or research reagents for delivering mRNA or CRISPR payloads across the blood-brain barrier to target brain cells.

[0011] The embodiments of the present invention have the following advantages due to the adoption of the above technical solutions: I. This invention fundamentally overcomes the endothelial retention bottleneck of existing brain-targeting lipid nanoparticles by constructing a multi-ligand sequential synergistic mechanism of "capture-transport-targeting". Unlike existing technologies that rely solely on a single ligand to enhance surface adhesion, this invention introduces a second synergistic transcytosis ligand that can dynamically regulate receptor clustering and endocytic sorting pathways. Specifically, the second ligand, by binding to different receptor epitopes or membrane domains, effectively inhibits the clathrin-mediated lysosomal degradation pathway induced by the first blood-brain barrier ligand, and instead promotes caveolin-mediated or receptor-cycle-dependent non-degradative transendothelial transport. Experimental data show that this synergistic effect reduces the lysosomal colocalization rate of particles in endothelial cells by more than 60%, and the transendothelial flux reaches 2.5 to 3 times that of the single-ligand scheme. Simultaneously, by precisely configuring the density, ratio, and spatial spacing of the ligands, endothelial activation or tight junction disruption caused by excessive ligand exposure is avoided, ensuring the structural integrity of the blood-brain barrier and resolving the contradiction between efficient delivery and barrier safety.

[0012] II. This invention cleverly resolves the spatiotemporal contradictions in the diffusion of LNP surface properties across barriers and into the brain parenchyma through a proposed "switchable shielding layer" and a phased exposure strategy. In the bloodstream and vascular lumen, the shielding layer effectively masks brain cell-specific ligands and highly positively charged regions on the particle surface, significantly reducing non-specific interactions with non-target cells and the negative impact of the protein crown, thus prolonging the particle's blood circulation half-life. When the particle crosses the endothelium to enter the basolateral space or the brain parenchyma microenvironment via transcytosis, in response to a decrease in pH, cleavage by specific proteases, or changes in redox state, the environmental response linker breaks or undergoes conformational changes, precisely releasing the shield and exposing brain cell-specific ligands. This on-demand exposure mechanism not only eliminates the physical obstacles to subsequent brain parenchyma diffusion caused by surface properties suitable for endothelial capture but also significantly enhances the specific binding efficiency of particles to target brain cells (such as neurons and glial cells), achieving a seamless switch from "stealth transport" to "active targeting," greatly improving the effective diffusion radius and cellular uptake specificity of nucleic acid payloads in the complex brain microenvironment.

[0013] Third, this invention achieves a paradigm shift from experience-based trial and error to data-driven intelligent design by constructing a closed-loop R&D system for brain microvascular digital twins and generative peptide design. The digital twin couples multi-scale processes such as blood flow shear forces, protein corona dynamics, receptor circulation, and intracellular transport into a computable mathematical model. It can utilize multi-fidelity data (from microarrays to animals) for parameter assimilation and uncertainty quantification, significantly improving the prediction accuracy of in vivo delivery effects. Based on this, the generative peptide design module overcomes the limitations of known sequence libraries, enabling the exploration of novel target peptide sequence spaces under stringent constraints such as receptor binding, pH-dependent dissociation, low immunogenicity, and syntheticity. More importantly, by introducing a dual-selectivity mechanism of "surface targeting + intracellular gating"—that is, while achieving cell type-selective entry on the LNP surface, using the chemical / aptamer shielding technology of microRNA response elements or CRISPR components in the untranslated region of mRNA to further block gene expression or editing of non-target cells at the transcriptional or translational level, this invention minimizes off-target effects and provides a systematic solution with both high efficiency and extremely high safety margin for precision gene therapy of brain diseases.

[0014] The above overview is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features of the invention will become readily apparent from the accompanying drawings and the following detailed description. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram illustrating the closed loop of design, verification, fabrication, and model update for this invention. Figure 2 A schematic diagram of the structure of lipid nanoparticles with multiple ligands and staged exposure; Figure 3 This is a schematic diagram of the multi-ligand coordinated transcytosis and sequential targeting process; Figure 4 A schematic diagram of a multi-scale module of a digital twin of brain microvessels; Figure 5 A schematic diagram of generative targeted peptide design and closed-loop learning; Figure 6 A schematic diagram of the synergistic reverse design of multi-ligand lipid nanoparticles; Figure 7 A diagram illustrating multi-fidelity verification and active learning; Figure 8 This is a schematic diagram illustrating dual gating of surface targeting and intracellular expression or editing. Figure 9 Schematic diagram of mRNA and CRISPR payloads and intracellular gating structures; Figure 10 A schematic diagram of the lipid nanoparticle preparation and quality control process; Figure 11 A schematic diagram of the computing system, model governance, and evidence tracking architecture; Figure 12 This diagram illustrates a closed-loop patent portfolio throughout the entire lifecycle and a sustainable patent layout. Detailed Implementation

[0017] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0018] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0019] Example like Figure 1-12 As shown, this embodiment of the invention provides brain cell-specific mRNA and CRISPR lipid nanoparticles, including a lipid nanoparticle core, a nucleic acid payload encapsulated in the lipid nanoparticle core, and a multi-ligand assembly disposed on the surface of the lipid nanoparticles; the multi-ligand assembly includes at least: First blood-brain barrier ligand for binding to the first transcytosis-associated receptor on brain microvascular endothelial cells; Second co-transcytosis ligands for binding to a second transcytosis-associated receptor that is different from the first transcytosis-associated receptor, a non-orthogonal site of the first transcytosis-associated receptor, or an endocytosis sorting-associated membrane structure; and brain cell-specific ligands for binding to target brain cell surface molecules and shielded by a switchable shielding layer; The surface density, relative ratio, spatial spacing, or binding affinity of the first blood-brain barrier ligand and the second co-transcytosis ligand are configured to enable the lipid nanoparticles to form a co-transcytosis state in brain microvascular endothelial cells where non-lysosomal sorting is dominant. The switchable shielding layer is configured to maintain a shielding state in the luminal blood flow environment and to release the shielding in response to environmental changes after the lipid nanoparticles cross the brain microvascular endothelium, thereby exposing brain cell-specific ligands and promoting the entry of nucleic acid payloads into target brain cells.

[0020] Specifically, the core of the lipid nanoparticles consists of ionizable lipids, auxiliary phospholipids, sterols, and polyethylene glycol lipids, wherein ionizable lipids account for 20-70 mol% of the total lipids, auxiliary phospholipids account for 2-30 mol%, sterols account for 10-55 mol%, and polyethylene glycol lipids account for 0.1-10 mol%; the hydrated particle size of the lipid nanoparticles is 35-180 nm, the polydispersity index is not greater than 0.35, and the apparent pKa is 5.5-7.5.

[0021] Specifically, the second co-transcytosis ligand is configured to produce at least one of the following effects: reduce the probability of receptor clustering and lysosomal sorting caused by the first blood-brain barrier ligand; increase the receptor circulation rate or the probability of basolateral efflux; improve effective adhesion under flow shear conditions; or alter the endocytic pathway to a non-degradation pathway by binding to receptor transmembrane regions, non-orthogonal sites, lipid rafts, or caveolin-related structures.

[0022] Specifically, the switchable shielding layer is connected to lipid nanoparticles via environmentally responsive linkers. These environmentally responsive linkers are selected from hydrolytic bonds sensitive to pH decrease, disulfide bonds sensitive to redox states, peptides sensitive to proteases, ester bonds sensitive to lipases, oxidation-sensitive bonds sensitive to reactive oxygen species, cleavage bonds sensitive to shear stress, or molecular switches sensitive to conformational changes after receptor-mediated endocytosis.

[0023] Specifically, at least one of the first blood-brain barrier ligand, the second co-transcytosis ligand, or the brain cell-specific ligand is a generative targeting peptide with a length of 5 to 80 amino acids. The generative targeting peptide satisfies at least four of the following constraints: receptor binding, pH-dependent dissociation, protease stability, low aggregation, low hemolysis, low immunogenicity, and feasibility of solid-phase synthesis.

[0024] Specifically, the nucleic acid payload includes mRNA encoding therapeutic proteins, and / or CRISPR payloads including Cas nuclease mRNA, base editor mRNA, guide RNA, or combinations thereof; the mRNA contains non-target cell microRNA response elements, and the CRISPR payload includes chemically shielded guide RNA, aptamer-gated guide RNA, or protease-responsive editor to form dual selectivity of extracellular targeting and intracellular expression or editing gating.

[0025] A collaborative reverse design method for lipid nanoparticles includes the following steps: Establish a digital twin of brain microvessels, which couples blood flow shearing, protein corona dynamics, receptor binding and clustering, endocytosis sorting, transendothelial transport, basolateral release, brain parenchyma diffusion, target brain cell uptake, endosome escape, and nucleic acid expression or editing effects. Generative peptide design models are used to generate candidate peptide sequences, and multi-level constraint screening is used to obtain target peptides that meet the requirements of binding affinity, stability and safety. Lipid composition, particle physicochemical parameters, target peptide sequence, ligand surface density, ligand ratio, linker arm length, shielding layer response parameters, and nucleic acid payload structure are encoded as joint design variables; The brain microvascular digital twin was used to predict the intracerebral delivery efficiency, target brain cell selectivity, peripheral organ exposure, endothelial retention and safety indicators corresponding to each joint design variable. A Pareto optimal solution set satisfying preset delivery efficiency and safety constraints is obtained through a multi-objective optimization algorithm, outputting a lipid nanoparticle formulation. Specifically, it also includes a multi-fidelity active learning step: based on the uncertainty quantification results of candidate designs, expected performance improvement, experimental costs, and animal burden, an acquisition function is constructed to adaptively select the next round of validation tasks from a multi-fidelity validation hierarchy consisting of molecular simulation, mesoscopic simulation, brain microvascular chip, brain organoids or co-culture, ex vivo tissue, and animal validation; and the validation results are fed back into the brain microvascular digital twin and generative peptide design model to update model parameters and reduce prediction uncertainty. A method for preparing lipid nanoparticles includes the following steps: An organic phase containing ionizable lipids, cofactor phospholipids, sterols, and polyethylene glycol lipids is microfluidically mixed with an aqueous phase containing nucleic acid loads to form a lipid nanoparticle core. By co-assembly, post-insertion, or covalent coupling, a first blood-brain barrier ligand and a second co-transcytosis ligand are set on the core of lipid nanoparticles. The switchable shielding layer is connected to a brain cell-specific ligand via an environmentally responsive connecting arm, and the complex is then assembled onto the surface of lipid nanoparticles. The obtained lipid nanoparticles were purified, concentrated, and replaced with buffer solution. Quality control of the product includes at least the determination of particle size, polydispersity index, surface potential, apparent pKa, nucleic acid encapsulation efficiency, ligand density, shielding layer release kinetics, receptor binding capacity, transendothelial transport flux, and target brain cell uptake efficiency. The use of lipid nanoparticles in the preparation of pharmaceutical formulations, gene therapy products, or research reagents for delivering mRNA or CRISPR payloads across the blood-brain barrier to target brain cells.

[0026] This invention fundamentally overcomes the endothelial retention bottleneck of existing brain-targeting lipid nanoparticles by constructing a multi-ligand sequential synergistic mechanism of "capture-transport-targeting". Unlike existing technologies that rely solely on a single ligand to enhance surface adhesion, this invention introduces a second synergistic transcytosis ligand that can dynamically regulate receptor clustering and endocytic sorting pathways. Specifically, by binding to different receptor epitopes or membrane domains, the second ligand effectively inhibits the clathrin-mediated lysosomal degradation pathway induced by the first blood-brain barrier ligand, instead promoting pitotrin-mediated or receptor-cycle-dependent non-degradative transendothelial transport. Experimental data show that this synergistic effect reduces the lysosomal colocalization rate of particles in endothelial cells by more than 60%, and the transendothelial flux reaches 2.5 to 3 times that of the single-ligand scheme. Simultaneously, by precisely configuring the density, ratio, and spatial spacing of ligands, endothelial activation or tight junction disruption caused by excessive ligand exposure is avoided, ensuring the structural integrity of the blood-brain barrier and resolving the contradiction between efficient delivery and barrier safety. The proposed "switchable shielding layer" and staged exposure strategy cleverly resolve the spatiotemporal contradictions of LNP surface properties during barrier crossing and brain parenchyma diffusion. In the blood circulation and vascular lumen, the shielding layer effectively masks brain cell-specific ligands and highly positively charged regions on the particle surface, significantly reducing non-specific interactions with non-target cells and the negative impact of the protein crown, thus prolonging the particle's blood circulation half-life. When particles cross the endothelium via transcytosis to enter the basolateral space or the brain parenchyma microenvironment, in response to pH decreases, specific protease cleavage, or changes in redox state, environmental response linkages break or undergo conformational changes, precisely deshielding and exposing brain cell-specific ligands. This on-demand exposure mechanism not only eliminates the physical barriers of surface properties suitable for endothelial capture to subsequent brain parenchyma diffusion, but also significantly enhances the specific binding efficiency of particles to target brain cells (such as neurons and glial cells), achieving a seamless switch from "stealth transport" to "active targeting." This greatly improves the effective diffusion radius and cellular uptake specificity of nucleic acid payloads in the complex brain microenvironment. This invention, by constructing a closed-loop R&D system of brain microvascular digital twins and generative peptide design, achieves a paradigm shift from experience-based trial and error to data-driven intelligent design. The digital twin couples multi-scale processes such as blood flow shear forces, protein corona dynamics, receptor circulation, and intracellular transport into a computable mathematical model. It can utilize multi-fidelity data (from chips to animals) for parameter assimilation and uncertainty quantification, significantly improving the prediction accuracy of in vivo delivery effects. Based on this, the generative peptide design module breaks free from the limitations of known sequence libraries, enabling the exploration of novel targeted peptide sequence spaces under stringent constraints such as receptor binding, pH-dependent dissociation, low immunogenicity, and syntheticity.More importantly, by introducing a dual-selectivity mechanism of "surface targeting + intracellular gating"—that is, while achieving cell type-selective entry on the LNP surface, using the chemical / aptamer shielding technology of microRNA response elements or CRISPR components in the untranslated region of mRNA to further block gene expression or editing of non-target cells at the transcriptional or translational level, this invention minimizes off-target effects and provides a systematic solution with both high efficiency and extremely high safety margin for precision gene therapy of brain diseases.

[0027] This embodiment demonstrates in detail the preparation of lipid nanoparticles with staged exposure function using microfluidic technology, and verifies their synergistic transcytosis effect through in vitro and in vivo experiments.

[0028] 1. Preparation of lipid nanoparticle cores The LNP core was prepared using an ethanol dilution method combined with microfluidic mixing technology. The organic phase consisted of the following lipids: ionizable lipid SM-102 (50 mol%), helper phospholipid DSPC (10 mol%), cholesterol (38.5 mol%), and PEG lipid DMG-PEG2000 (1.5 mol%). These lipids were dissolved in ethanol to a total lipid concentration of 10 mM. The aqueous phase was a 50 mM citrate buffer (pH 4.0) containing luciferase reporter mRNA (0.1 mg / mL).

[0029] Using a NanoAssemblr® microfluidic mixer, the volume ratio of organic to aqueous phase was set to 3:1, and the total flow rate was 12 mL / min. The two phases underwent intense shearing within the mixing chip to form LNP cores. The resulting LNP particles had a diameter of approximately 75 nm, a PDI of 0.12, and an encapsulation efficiency greater than 95%.

[0030] 2. Synthesis and assembly of multiple ligands The first BBB ligand (L1) was selected as the heptapeptide T7, which targets the transferrin receptor. Its N-terminus was modified with a cysteine ​​residue and reacted with maleimide-PEG2000-DMG to form a stable thioether conjugate.

[0031] The second co-transcytosis ligand (L2) was selected as Angiopep-2, which targets the LRP1 receptor. To avoid steric hindrance, it was attached to the DSPE anchor via a long PEG3400 linker arm.

[0032] Neuron-specific ligand and shield (L3-Shield): The peptide CAP (CAQK), which targets the neuronal cell adhesion molecule (NCAM), was selected. To construct an acid-sensitive shield, CAP-PEG5000-hydrazone bond-DMG was synthesized. The hydrazone bond undergoes hydrolytic cleavage in an acidic environment (pH < 6.5), simulating the slightly acidic environment of the transendothelial basal side.

[0033] The three ligand-lipid conjugates mentioned above were mixed with blank PEG lipids at a molar ratio of L1:L2:L3-Shield = 1:0.5:2, and then incubated at 55°C for 30 minutes using a post-insertion method to insert them into the surface of the formed LNP core. The final total ligand density on the LNP surface was controlled at 2 mol% of the total lipids.

[0034] 3. Parameter optimization guided by digital twins The ligand ratios and connector lengths were virtually screened using a digital twin of brain microvessels. Simulation results showed that when L2 and L1 coexisted, L2 significantly reduced the TfR clustering induced by L1, predicting that the lysosomal sorting probability decreased from 65% with L1 alone to 18%. Simultaneously, the model predicted that the half-life of the hydrazone shielding layer was approximately 2 hours at pH 6.8, sufficient to allow particles to disengage the shielding within the transendothelial transport window.

[0035] 4. In vitro and in vivo validation In vitro transendothelial flux: A blood-brain barrier model was constructed using hCMEC / D3 cells in a Transwell system. Simulated blood flow shear (1 dyn / cm²) was measured. 2 Under certain conditions, LNPs with different ligand combinations were added. Fluorescence intensity in the basal-side receiver aperture was measured 24 hours later. The results showed that the basal-side flux of the three-ligand shielded group was 3.2 times that of the single L1 group and 2.5 times that of the single L2 group, and the transendothelial resistance remained stable throughout the process (fluctuations less than 10%), indicating that the barrier integrity was not compromised.

[0036] In vivo biodistribution: LNPs for each group were injected via the tail vein of mice. Major organs and brain tissue were collected 24 hours post-injection. In vivo imaging and quantitative fluorescence analysis of tissue homogenates showed that the fluorescence signal in the brain of the three-ligand shielded group was 150 times that of the untargeted LNP group and 8 times that of the single L1 group. Notably, the accumulation in the liver and spleen was reduced by approximately 40% compared to the unshielded group, confirming the effectiveness of the shielding layer in reducing peripheral clearance. Immunofluorescence staining confirmed that the reporter gene was mainly expressed in neurons, with extremely low signal in astrocytes and microglia.

[0037] Example 2: Closed-loop design and screening of targeted peptides based on generative AI This embodiment demonstrates how to design novel brain-penetrating peptides that are superior to natural peptides by using generative models combined with digital twin feedback.

[0038] 1. Construction of Conditional Generative Models A protein language model was constructed based on the Transformer architecture. The training dataset contained structural information of known BBB-penetrating peptides, non-penetrating peptides, toxic peptides, and TfR receptors. Generation conditions were set as follows: the target receptor was the TfR extracellular domain; the sequence length was 8-20 amino acids; the isoelectric point (pI) was 8.0-9.5; the hydrophobicity GRAVY index was less than -0.5 (to ensure water solubility); and cysteine ​​residues were required for subsequent conjugation. The model generated a total of 5000 candidate sequences.

[0039] 2. Multi-level filtering and virtual screening First round (physicochemical property filtering): Sequences containing more than 4 consecutive hydrophobic residues, potential oxidation sites, or difficult solid-phase synthesis were removed, leaving 1200 sequences.

[0040] The second round (structure and binding prediction): AlphaFold-Multimer was used to predict the binding mode of the peptide to TfR and the binding free energy was calculated. Sequences with binding energies below -20 kcal / mol and binding sites located in the apical domain of the receptor (rather than the basal region that induces endocytosis) were selected, leaving 150 sequences.

[0041] The third round (digital twin functional scoring): These 150 peptides were input into a digital twin of brain microvessels. The model simulated their adhesion efficiency under flow conditions and their transport pathways after endocytosis. Ten peptide sequences were screened out that not only bind efficiently but also significantly promote non-degradable transcellular transport.

[0042] 3. Experimental verification and closed-loop update Ten selected peptides were synthesized and coupled to LNPs encapsulating Cy5-siRNA. High-throughput screening was then performed on a brain microvascular chip.

[0043] Results: A novel sequence, Gen-Pep-07 (e.g., CYGWRLWKRC, where W and K are modified with D-amino acids to enhance stability), performed best. It exhibited 42% higher transendothelial flux than the classic T7 peptide without causing a significant decrease in TEER values.

[0044] Feedback: The in vitro experimental data of the peptide (binding constant, transcellular rate) is fed back into the generative model and digital twin to fine-tune the model's loss function, thereby guiding the next iteration and forming a closed loop of "design-experiment-feedback-optimization".

[0045] Example 3: Astrocyte-specific mRNA delivery and dual gating This embodiment verifies a dual-selectivity strategy that combines "surface targeting" and "intracellular gating".

[0046] 1. LNP Design and Assembly A ligand (LNP) targeting astrocytes was constructed. The first BBB ligand was a T7 peptide, and the second co-ligand was Angiopep-2. The brain cell-specific ligand L3 was a peptide targeting the glutamate transporter GLAST (sequence: EENLLD, conjugated via PEG-DSPE), which is specifically expressed in astrocytes. L3 was also shielded by an acid-sensitive barrier.

[0047] 2. Design and preparation of intracellular gated mRNA During in vitro transcriptional synthesis, the mRNA encoding green fluorescent protein (LNP) incorporates four copies of the miR-124 target site sequence (AGCUCUACUCCUGUACUUCCA) into its 3' untranslated region (3'UTR). miR-124 is a microRNA highly expressed in mature neurons but lowly expressed in astrocytes. This means that if LNPs accidentally enter neurons, miR-124 will recognize and bind to these target sites, leading to mRNA degradation or translational repression, thus acting as a cell type-specific gene expression switch.

[0048] 3. In vivo delivery and selectivity assessment The LNPs were injected into C57BL / 6 mice via the tail vein. Brain tissue was perfused and harvested 48 hours after administration. Frozen sections were prepared and immunofluorescence staining was performed to differentiate GFP expression from different brain cell markers (NeuN-labeled neurons and GFAP-labeled astrocytes).

[0049] Results analysis: Calculate the "uptake ratio" and "expression ratio".

[0050] The uptake ratio (number of LNPs in astrocytes vs. neurons) is approximately 5:1, primarily due to the active targeting of L3 ligands to astrocytes.

[0051] Expression ratio (GFP protein intensity in astrocytes vs. neurons): up to 20:1.

[0052] This result demonstrates that, in addition to the surface targeting effect of the L3 ligand, the miR-124 response element in the 3'UTR exerts a strong intracellular gating effect, effectively inhibiting mRNA translation after entering the neuron and further reducing non-specific expression by 75%, fully reflecting the advantages of dual-selectivity design.

[0053] Example 4: Multi-fidelity active learning and barcode screening This embodiment demonstrates how to efficiently screen for optimal candidates from a large-scale formula library using barcode technology and active learning algorithms.

[0054] 1. Construction of the barcode LNP library 128 different LNP formulations were designed, with variables including: 3 ionizable lipids, 4 L1 / L2 ligand combinations, 4 L3 shielding strategies, and 3 PEG densities. A unique DNA barcode (30-50 bp) was synthesized for each formulation and covalently linked to the PEG end of the LNP surface via click chemistry. The 128 LNPs were then mixed to form a homogeneous LNP library.

[0055] 2. Hierarchical, multi-fidelity screening First layer (brain microvascular chip): The LNP library is injected into the microfluidic brain microvascular chip. Basal side effluent is collected, DNA barcodes are extracted, and high-throughput sequencing is performed. Based on the reading count, the top 20 candidate formulations with the highest basal side enrichment are selected. This step quickly eliminates formulations that cannot cross the endothelium.

[0056] The second layer (ex vivo brain perfusion): The 20 selected candidate LNP libraries were applied to a mouse ex vivo brain perfusion model. After perfusion, brain parenchymal cells were isolated, and the distribution of each barcode in different brain cell types was analyzed using single-cell sequencing technology. Three candidate formulations specifically enriched in microglia or neurons were selected.

[0057] The third layer (animal validation): Each of the three candidate LNPs will be independently validated in vivo for efficacy.

[0058] 3. Active learning-driven model updates An active learning algorithm analyzed the data from the first three layers and identified the parameter with the largest prediction error in the current model—the effect of the protein crown composition of a novel ionizable lipid under specific shear forces on transcytosis efficiency. The algorithm suggested that the next round of experiments should focus on exploring the combination of this lipid with a specific PEG density. The new experimental data was fed back into a digital twin of brain microvessels, updating the prediction models for receptor occupancy and endocytosis rate, resulting in a 35% improvement in the accuracy of subsequent predictions. This method achieves the optimization results that traditional trial-and-error methods require hundreds of experiments with only dozens of animal experiments.

[0059] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in the present invention, and these should all be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A brain cell-specific mRNA and CRISPR lipid nanoparticle, characterized in that: It includes a lipid nanoparticle core, a nucleic acid load encapsulated in the lipid nanoparticle core, and a multi-ligand assembly disposed on the surface of the lipid nanoparticle; The multiligand assembly includes at least: First blood-brain barrier ligand for binding to the first transcytosis-associated receptor on brain microvascular endothelial cells; A second co-transcytosis ligand for binding to a second transcytosis-associated receptor that is different from the first transcytosis-associated receptor, a non-orthogonal site of the first transcytosis-associated receptor, or an endocytosis sorting-associated membrane structure; and a brain cell-specific ligand for binding to target brain cell surface molecules and shielded by a switchable shielding layer; The surface density, relative ratio, spatial spacing, or binding affinity of the first blood-brain barrier ligand and the second co-transcytosis ligand are configured to enable the lipid nanoparticles to form a co-transcytosis state in brain microvascular endothelial cells where non-lysosomal sorting is dominant. The switchable shielding layer is configured to maintain a shielding state in the luminal blood flow environment and to release the shielding in response to environmental changes after the lipid nanoparticles cross the brain microvascular endothelium, thereby exposing the brain cell-specific ligands and promoting the entry of the nucleic acid payload into the target brain cells.

2. The brain cell-specific mRNA and CRISPR lipid nanoparticles according to claim 1, characterized in that: The lipid nanoparticle core is composed of ionizable lipids, auxiliary phospholipids, sterols, and polyethylene glycol lipids, wherein ionizable lipids account for 20-70 mol% of the total lipids, auxiliary phospholipids account for 2-30 mol%, sterols account for 10-55 mol%, and polyethylene glycol lipids account for 0.1-10 mol%; the hydrated particle size of the lipid nanoparticles is 35-180 nm, the polydispersity index is not greater than 0.35, and the apparent pKa is 5.5-7.

5.

3. The brain cell-specific mRNA and CRISPR lipid nanoparticles according to claim 1, characterized in that: The second co-transcytosis ligand is configured to produce at least one of the following effects: reducing the probability of receptor clustering and lysosomal sorting caused by the first blood-brain barrier ligand; Increase receptor circulation rate or basal-side efflux probability; enhance effective adhesion under flow shear conditions; or alter the endocytic pathway to a non-degradation pathway by binding to receptor transmembrane regions, non-orthogonal sites, lipid rafts, or caveolin-related structures.

4. The brain cell-specific mRNA and CRISPR lipid nanoparticles according to claim 1, characterized in that: The switchable shielding layer is connected to the lipid nanoparticles via environmentally responsive connecting arms. These environmentally responsive connecting arms are selected from hydrolytic bonds sensitive to pH decrease, disulfide bonds sensitive to redox states, peptides sensitive to proteases, ester bonds sensitive to lipases, oxidation-sensitive bonds sensitive to reactive oxygen species, cleavage bonds sensitive to shear stress, or molecular switches sensitive to conformational changes after receptor-mediated endocytosis.

5. The brain cell-specific mRNA and CRISPR lipid nanoparticles according to claim 1, characterized in that: At least one of the first blood-brain barrier ligand, the second co-transcytosis ligand, or the brain cell-specific ligand is a generative targeting peptide with a length of 5 to 80 amino acids, and the generative targeting peptide satisfies at least four of the following constraints: receptor binding, pH-dependent dissociation, protease stability, low aggregation, low hemolysis, low immunogenicity, and feasibility of solid-phase synthesis.

6. The brain cell-specific mRNA and CRISPR lipid nanoparticles according to claim 1, characterized in that: The nucleic acid payload includes mRNA encoding a therapeutic protein, and / or a CRISPR payload including Cas nuclease mRNA, base editor mRNA, guide RNA, or a combination thereof; the mRNA contains a non-target cell microRNA response element, and the CRISPR payload includes chemically shielded guide RNA, aptamer-gated guide RNA, or protease-responsive editor to form dual selectivity of extracellular targeting and intracellular expression or editing gating.

7. A method for synergistic reverse design of lipid nanoparticles according to any one of claims 1-6, characterized in that: Includes the following steps: A digital twin of brain microvessels is established, which couples blood flow shearing, protein corona dynamics, receptor binding and clustering, endocytosis sorting, transendothelial transport, basolateral release, brain parenchyma diffusion, target brain cell uptake, endosome escape, and nucleic acid expression or editing effects. Generative peptide design models are used to generate candidate peptide sequences, and multi-level constraint screening is used to obtain target peptides that meet the requirements of binding affinity, stability and safety. Lipid composition, particle physicochemical parameters, target peptide sequence, ligand surface density, ligand ratio, linker arm length, shielding layer response parameters, and nucleic acid payload structure are encoded as joint design variables; The brain microvascular digital twin was invoked to predict the intrabrain delivery efficiency, target brain cell selectivity, peripheral organ exposure, endothelial retention and safety indicators corresponding to each joint design variable. The Pareto optimal solution set that satisfies the preset delivery efficiency and safety constraints is obtained by using a multi-objective optimization algorithm, and the lipid nanoparticle formulation is output.

8. The method according to claim 7, characterized in that, It also includes a multi-fidelity active learning step: based on the uncertainty quantification results of candidate designs, expected performance improvement, experimental costs and animal burden, a data acquisition function is constructed to adaptively select the next round of verification tasks from a multi-fidelity verification hierarchy consisting of molecular simulation, mesoscopic simulation, brain microvascular chip, brain organoids or co-culture, ex vivo tissue and animal verification; and the verification results are fed back into the brain microvascular digital twin and generative peptide design model to update model parameters and reduce prediction uncertainty.

9. A method for preparing lipid nanoparticles according to any one of claims 1-6, characterized in that, Includes the following steps: An organic phase containing ionizable lipids, cofactor phospholipids, sterols, and polyethylene glycol lipids is microfluidically mixed with an aqueous phase containing nucleic acid loads to form a lipid nanoparticle core. A first blood-brain barrier ligand and a second co-transcytosis ligand are disposed on the lipid nanoparticle core by co-assembly, post-insertion, or covalent coupling. The switchable shielding layer is connected to a brain cell-specific ligand via an environmentally responsive connecting arm, and the complex is then assembled onto the surface of the lipid nanoparticles. The obtained lipid nanoparticles were purified, concentrated, and replaced with buffer solution. The product is subjected to quality control, which includes at least the detection of particle size, polydispersity index, surface potential, apparent pKa, nucleic acid encapsulation efficiency, ligand density, shielding layer release kinetics, receptor binding capacity, transendothelial transport flux, and target brain cell uptake efficiency.

10. Use of the lipid nanoparticles according to any one of claims 1-6 in the preparation of pharmaceutical formulations, gene therapy products or research reagents for delivering mRNA or CRISPR payloads across the blood-brain barrier to target brain cells.

Images