This project is the code used for the analysis of the paper : Distinct forms of structural plasticity of adult-born mouse interneuron spines induced by different odor learning paradigms. Available here : https://www.biorxiv.org/content/10.1101/2023.07.20.549882v1 The reconstruction pipeline as well as the analysis pipeline was used to study dendritic spines but feel free to test it on your data. Please, if you use this code, cite us :

 Distinct forms of structural plasticity of adult-born interneuron spines induced by different odor learning paradigms
Aymeric Ferreira, Vlad-Stefan Constantinescu, Sarah Malvaut, Armen Saghatelyan, Simon V. Hardy
bioRxiv 2023.07.20.549882; doi: https://doi.org/10.1101/2023.07.20.549882 

Table of Contents

1. Install
2. Deconvolution
3. Segmentation
4. 3D reconstruction
5. Extraction of the spines
6. Analysis of the spines
7. Analysis pipeline
8. Workflow
9. Notes

Install

This package may require the installation of the package PyImageJ for the deconvolution process. If needed, follow the instructions on the official website : https://github.com/imagej/pyimagej

If you prefer to skip deconvolution, plan to use ImageJ directly, or have installed PyImageJ, you can install the remaining requirements via pip. For guidance on installing Python, refer to our previous project: https://github.com/SaghatelyanLab/clusterAnalysis

pip install -r requirements.txt

Deconvolution

This package allows for the deconvolution of images. We have included the FIJI version with the Iterative Deconvolve 3D plugin pre-installed. To use it, open your image in FIJI and run the plugin. Alternatively, the deconvolve_folder method in deconvolution_segmentation.py can be used, but it requires correctly binding Maven and JAVA to PyImageJ. For more information, refer to the https://py.imagej.net/en/latest/Troubleshooting.html#jgo-jgo-executablenotfound-mvn-not-found-on-path

Segmentation

Place your images in the Images folder and run deconvolution_segmentation.py. Adjust the parameters in the file as needed. The best parameters for single spine images were found to be mu = 1, lambda1 = 1, lambda2 = 4. For full confocal images, mu = 1, lambda1 = 1, lambda2 = 9 yielded the best results.

3D reconstruction

The 3D_reconstruction.py script reconstructs 3D images from the Segmented image folder, creating several meshes in the Mesh folder. The optimal level_threshold value depends on your image's signal-to-noise ratio and the strength of the spine neck signal. For our images, parameters between 10 and 20 provided the best results.

Extraction of the spines

To extract the dendritic spines from the dendrite we used meshlab. You can download it here : https://www.meshlab.net/#download The extraction is performed manually.

Analysis of the spines

After extraction, place your spines in the Spines folder and run metrics.py. The script calculates and outputs metrics for each spine in a CSV file. These metrics include Length, Volume, Surface, Hull Volume, Hull Ratio, Average Distance, CVD, and Open Angle.

• Length: the length of the spine
• Volume : the volume of the spine
• Surface : the surface of the spine
• Hull volume : the volume of the convex hull of the spine
• Hull ratio : the ratio between the volume of the spine - the volume of the convex hull by the Volume of the spine
• Average distance : the average distance between the spine and the convex hull
• CVD : the coefficient of variation of the distance between the edges of the spine and the spine base center
• Open angle : the angle between the spine normal and the different vertices of the spine

Users can easily add metrics if needed by adding them in the function compute_trimesh_metrics in the file metric.py.

Analysis pipeline

The analysis pipeline is similar to our previously published work : https://github.com/SaghatelyanLab/clusterAnalysis We added scripts (kde.py and piecharts.py) for generating Kernel Density Estimation (KDE) graphs, pie charts, and associated statistical tests.

Workflow

Add your images in Images/ folder. The parameters of segmentation mu, lambda1, lambda2 and iterations can be adjusted to increase segmentation quality. After defining the parameters, run :

python deconvolution_segmentation.py

Then, adjust the parameters z_spacing, pixel_size_x and pixel_size_y according to your image stack and perform the reconstruction :

python reconstruction.py

Then, extract your spines manually using Meshlab and place them in Spines/ folder. Finally, run :

python metrics.py

The metrics will be saved in a CSV file in the Spines/ folder.

After downloading the analysis pipeline, modify the DataSetpath in the general_analysis_parameters.json and perform the analysis using :

python analyse.py

Notes :

• PyimageJ has specific system requirements, including the installation of conda/mamba, and it currently supports only up to Python 3.8. For users looking to perform deconvolution tasks, an alternative approach is to utilize ImageJ with the Iterative Deconvolve 3D plugin, available here : https://imagej.net/plugins/iterative-deconvolve-3d . This plugin is compatible with versions of ImageJ up to 1.52p, a working version of FIJI is included in this repo.
• Additionally, it's important to note that the original morphsnakes.py script has compatibility issues with newer versions of numpy (specifically versions above 1.22), leading to deprecation warnings. The warning in question, VisibleDeprecationWarning, is triggered by creating an ndarray from sequences of varying lengths or shapes. Currently, to avoid these issues, it is recommended to use numpy versions below 1.23. However, we have proposed a modification to morphsnakes.py to address this warning and enhance compatibility with newer numpy versions.