Nanocarrier imaging at single-cell resolution across entire mouse bodies with deep learning

Abstract

Efficient and accurate nanocarrier development for targeted drug delivery is hindered by a lack of methods to analyze its cell-level biodistribution across whole organisms. Here we present Single Cell Precision Nanocarrier Identification (SCP-Nano), an integrated experimental and deep learning pipeline to comprehensively quantify the targeting of nanocarriers throughout the whole mouse body at single-cell resolution. SCP-Nano reveals the tissue distribution patterns of lipid nanoparticles (LNPs) after different injection routes at doses as low as 0.0005 mg kg^-1-far below the detection limits of conventional whole body imaging techniques. We demonstrate that intramuscularly injected LNPs carrying SARS-CoV-2 spike mRNA reach heart tissue, leading to proteome changes, suggesting immune activation and blood vessel damage. SCP-Nano generalizes to various types of nanocarriers, including liposomes, polyplexes, DNA origami and adeno-associated viruses (AAVs), revealing that an AAV2 variant transduces adipocytes throughout the body. SCP-Nano enables comprehensive three-dimensional mapping of nanocarrier distribution throughout mouse bodies with high sensitivity and should accelerate the development of precise and safe nanocarrier-based therapeutics.

Fig. 1. Optimized DISCO clearing for imaging nanocarriers at low doses.

a, Scheme of SCP-Nano—a pipeline for mapping and quantifying the biodistribution of any fluorescently labeled nanocarrier throughout the entire mouse body with single-cell resolution and high sensitivity. b,c, Bioluminescence imaging (ventral) 6 h after intravenous injection of 0.5 mg kg⁻¹ (b) and 0.0005 mg kg⁻¹ (c) of luciferase mRNA-carrying LNPs. d–f, Whole body light sheet imaging of mice intravenously injected with 0.0005 mg kg⁻¹ Alexa Fluor 647–labeled EGFP mRNA-carrying LNPs and cleared with our refined DISCO clearing methods. This approach enables the visualization of mRNA delivery throughout the entire mouse body, including the liver (e) and spleen (f), at cellular resolution. g–m, Visualization of whole mouse body LNPs after intranasal delivery of 0.0005 mg kg⁻¹: maximum intensity projection (g) and single optical slice views (h–l); representative individual optical slices of the lung (m).

Fig. 2. SCP-Nano—a deep-learning-based pipeline to segment and analyze all targeted cells.

a, Flowchart of the SCP-Nano pipeline. b, Customized 3D U-Net architecture of the SCP-Nano deep learning segmentation model. c, Comparison of the F1 (instance Dice) scores of SCP-Nano segmentation model with other models. d, Comparison of Imaris and SCP-Nano prediction accuracy using liver images. e, Illustration of the entire nanoparticle prediction pipeline. After obtaining the whole body dataset via light sheet microscopy, we used VR glasses for organ annotation, followed by the application of our SCP-Nano analysis algorithm to quantify the LNP distribution in the whole body. Example images are from the lung. Compared to the ground truth data, our algorithm accurately detects all different sizes of delivery events in the lung. f, Raw data of LNP distribution in the liver and instance-separated multicolor segmentation obtained by SCP-Nano. Each color represents a separate delivery event as predicted by the model. g, Continuous segmented slice views demonstrate single-cell segmentation.

Fig. 3. SCP-Nano reveals differences in LNP biodistribution based on different application routes.

a, Density heatmaps of the distribution of LNP-delivered mRNA applied using different routes (0.0005 mg kg⁻¹ in each case). b, Raw projection images (left) and density heatmaps of selected organs. Arrows point to intra-organ delivery hotspots. c, Organ-level quantification of mRNA delivery events across key organs for different application routes using the SCP-Nano deep learning algorithm (n = 3 mice per group, mean ± s.d.). d, Quantification of mRNA delivery events in lymph nodes of intramuscularly injected mice. i.m., intramuscular; i.v., intravenous; LN, lymph node. Source data

Fig. 4. SCP-Nano reveals protein expression from LNP-delivered mRNA and LNP off-targeting.

a. Whole body projection view of mRNA and EGFP protein expression 72 h after intramuscular injection of 0.0005 mg kg⁻¹ EGFP mRNA-LNPs. b–d, Detailed views of the spleen (b), liver (c) and heart (d). e, Quantitative evaluation of the SCP-Nano segmentation model (fine-tuned on EGFP data) for detecting protein expression (FN, false negatives; FP, false positives; TP, true positives, compared to manual annotation). f. Body-wide distribution of SARS-CoV-2 spike S1 protein derived from LNP-delivered mRNA administered intramuscularly at 72 h after injection (f). Spike proteins were detected in the heart (f′). g, Confocal images of heart tissue sections stained for endothelial cells in capillaries using podocalyxin antibody (red), arteries using αSMA antibody (green) and spike S1 protein using a spike nanobody (yellow). h, PCA of mass-spectrometry-based proteomics data of different groups: spike LNP, EGFP LNP, no-cargo LNP and PBS. i, Top-level pathways in Reactome database differentially expressed between the two control groups (no-cargo LNP/PBS) and the combined spike LNP and EGFP LNP groups (n = 9, mean ± s.d.; one-way ANOVA). j,k, Same analysis for proteins upregulated in no-cargo LNP in comparison to the PBS (j) and in the spike mRNA in comparison to no-cargo LNP (k) (n = 9, mean ± s.d.; one-way ANOVA). l, Analysis of vascular health using typical protein markers (Supplementary Table 3) for the three different groups. NS P > 0.05, **P < 0.01 (n = 9; one-way ANOVA). i.m., intramuscular; PC, principal component; NS, not significant.

Fig. 5. SCP-Nano reveals targeted delivery of DNA origami.

a–f, Cell-level resolution biodistribution of non-targeted and CX3CR1 antibody immune-cell-targeted (illustrated on the left) origami throughout the entire mouse body (c,d) and in detailed views of the liver (e,f; illustrating single-cell resolution) 20 min after intravenous injection of 50.725 mg kg⁻¹ DNA origami. g, Intra-tissue distribution of the immune cell and non-targeting origami (red) and co-staining with target cell marker CX3CR1 (left, yellow) and general immune cell marker CD68 (right, yellow) in confocal images of individual liver slices. h, Quantification of the co-localization of untargeted and CX3CR1-targeted origamis with CX3CR1⁺ and CD68⁺ cells (n = 3, mean ± s.d.). i, The prediction accuracy of DNA origami detection by SCP-Nano algorithm in different organs. j, Density map of CX3CR1 immune-cell-targeted origami distribution throughout the entire mouse body. k, SCP-Nano-based quantification of the biodistribution of CX3CR1 immune-cell-targeted origami in different organs. Source data

Fig. 6. SCP-Nano quantifies targeting specificity of two types of AVVs.

a–e, Biodistribution of EGFP expressed after gene delivery with PHPeB-AAV throughout the body 2 weeks after injection. Whole body view (a) and subregions showing PHPeB distribution in various areas, such as the brain (b, with density maps of cell bodies labeled by PHPeB), hippocampus (c), spinal cord (d) and the inguinal and lumbar lymph nodes (e). f–j, Distribution of EGFP after gene delivery with Retro-AAVs throughout the body 2 weeks after injection: whole body view (f) and subregions, including adipose tissue near the hindlimb (g), close to the ureter (h), beneath the abdominal wall (i) and at the neck (j). k, SCP-Nano quantification of the number of cells labeled by PHPeB-AAV and Retro-AAV-derived EGFP with high accuracy. l, Intra-brain differences in PHP.eB-AAV-based gene delivery. Blue indicates lower-than-average and red indicates higher-than-average EFP⁺ cell density. Color scales represent log₂-transformed cell density relative to whole brain average, plotted using DELiVR on the Allen Brain Atlas reference Atlas with BrainRender. Color coding is per Allen Brain Atlas CCF3 regions.