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

• Address: https: //www.nature.com/articles/s41587-024-02528-1

• Project homepage:

• GitHub repository: https: //github.com/erturklab/SCP-Nano

• Citation:

Introduction | Quantifying nanocarrier distribution throughout the body at single-cell resolution

1. Background: Biomedical science requires drug delivery to be performed in a way that minimizes off-target effects. Nanocarriers, lipid nanoparticles (LNPs), can be used for the above objectives.

2. Gap & Motivation: Existing methods such as emission tomography (PET), computed tomography (CT), magnetic resonance imaging (MRI), and in vivo optical imaging lack the ability to identify targeted cells at the single-cell level and have limitations in detecting low-intensity off-target sites. In contrast, conventional histologic methods, while offering subcellular resolution and high sensitivity, are not applicable to whole-body analysis. A process to map and quantify the biodistribution of any fluorescently labeled nanocarrier throughout the mouse with single-cell resolution and high sensitivity.

3. Research and Contributions:
1. Developed a suite of processes called SCP-Nano that combines DISCO clearance, light-sheet imaging, and deep learning analysis to enable single-cell level distributions to map the distribution of nanobodies. (What Work Was Done).
2. SCP-Nano can quantify the distribution of nanomedicines throughout the body at very low doses (0.0005 mg/kg) and found that the mode of administration affects tissue targeting. (How well did the method work and what problem did it solve)
3. Slight off-target accumulation in the heart was detected and combined with proteomics to reveal underlying mechanisms, providing a possible explanation for some clinical side effects. Applied SCP-Nano to other nanocarriers (liposomes, polymers, DNA origami, AAV) and found it to be effective in different systems as well, identifying new target tissues (adipose tissue). (What are the results and do the findings matter)

Methods

Materials and methods for biochemistry experiments

Deep learning for LNP nanocarrier detection

Data labeling (and preprocessing)

• Nanoparticle-targeted cells were manually labeled in VR using SyGlass (v1.7.2).

• Data were obtained from head, heart, lung, liver, spleen and kidney, totaling 31 patches.

• Segmentation:
• Training/validation set: 21 patches totaling 13,927 particles.
• Test set: 10 patches covering all organs, totaling 6,424 particles.

• All patches were normalized to the 0-1 range before training.

Segmentation Models

• Several mainstream 3D segmentation models (VNet, U-Net++, Attention U-Net, UNETR, SwinUNETR, nnFormer, 3D U-Net) were compared.

• The best performing 3D U-Net (6-layer encoder, 5-layer decoder, leaky ReLU) is finally selected.

• The model is named SCP-Nano and the training details are as follows:
• Fifty-fold cross-validation;
• Patch size is 128×128×128;
• batch size = 2;
• Loss function: dice + Cross-Entropy;
• Optimizer: SGD, lr = 0.0001;
• Train 1000 rounds and keep the checkpoint with the lowest validation loss;
• Integrate the best model of five folds for inference.

Inference and analysis

• Raw scanned images are extremely large (up to 30,000 × 10,500 × 2,000 voxel), chunked (up to 500³) for post-inference, and finally stitched together for full-image segmentation results.

• The VOIs of six organs were manually labeled using VR: brain, heart, lung, liver, spleen, kidney.

• Combine the organ regions with the segmentation results for quantitative analysis:
• Connectivity domain analysis was performed using the cc3d library;
• Each nanoparticle point was weighted according to its brightness against the local background contrast;
• The sum of the contrast of all points reflects the total amount of LNP in the organ.

• Cell density maps were generated using a sliding window (16×16×4 voxel) and 3D Gaussian filtering to show the distribution of nanoparticles in vivo.

Results.

High-resolution whole-body biodistribution imaging (what about imaging)

AI-based quantification of cell-level LNP targeting (segmentation model effect)

SCP-Nano reveals LNP tropism in different pathways

SCP-Nano reveals potential off-target effects

• After injecting this mRNA-LNP into the body, a small amount of LNP was also detected in heart tissue (although not a lot, it was a "low-intensity accumulation").

• Then they did a proteomic analysis of the heart tissue (that is, they analyzed the changes in various proteins in the tissue).

• It turned out that some of the proteins associated with the immune system and blood vessel function were changed.

• This may explain some of the heart-related adverse reactions seen in the clinic after vaccination (such as myocarditis and the like).

Visualization of DNA origami targeting different cell types

• Fabrication of DNA origami nanoparticles: First, the researchers used DNA, a molecule, to design and assemble tiny nanostructures like origami. These DNA origami can be designed into different shapes as needed, like small boxes or rods, and are very small in size, making them suitable for use as carriers.

• Binding antibodies to DNA origami: In order for these nanoparticles to find and target specific cells in the body (e.g., immune cells), the researchers attach antibodies (e.g., the CX3CR1 antibody) to the surface of the DNA origami. The antibodies act as "keys" that help "lock" the DNA origami to specific cell types.

• Mouse experiments and imaging: In the experiments, the researchers injected the antibody-containing DNA origami nanoparticles into mice and used SCP-Nano analysis to observe how the nanoparticles were distributed in the mice and targeted to specific cells.

• Verification of targeting effect: The results show that SCP-Nano technology can effectively quantify and visualize nanocarriers (DNA origami), providing important technical support for future DNA origami-based precision medicine treatments.

Visualization and quantification of AAV distribution by SCP-Nano

Conclusion

Insights

1. For this kind of article in Nature, the results part is very important, for example, in this paper, there are several kinds of nano-antibodies, and then four different experiments were verified (EGFP mRNA, new crown mRNA vaccine, DNA origami, AVV), which is a good reflection of the workload.

2. The Introduction of this paper is very worthy of reference, and is also suitable for reference to study the writing structure of Nature subpublications:

Introduction part of the need to introduce: 1. research background (Background), to introduce the field of research, the development of the current situation is what, related to the introduction of important terms and concepts.In recent years, [field / issue] has attracted significant attention due to [reason]. ]. Numerous studies have attempted to [what has been done]. Gap & Motivation: However, despite these advances, [specific problem] remains unresolved.

Existing approaches typically suffer from [shortcomings/challenges]. Therefore, it is necessary to [your research motivation]. 4. 4. The research content and contribution of this paper (Objective & Contribution).

In this work, we propose a novel method to....

The main contributions of this paper are as follows.

3. Due to my own reading and working habits, I did not follow the structure of the original paper to write the notes, and omitted the part that favors biomedical experiments.