Single-cell isotope tracing - 13C-SpaceM

About this repository

This repository contains the code necessary to reproduce the figures from the manuscript 13C-SpaceM: Spatial single-cell isotope tracing reveals heterogeneity of de novo fatty acid synthesis in cancer

If you are interested in modeling your own bulk/MSI isotope tracing data with a binomial distribution, check out this easier example with more explanations: 13C-binomial

Installation

    git clone https://github.com/Buglakova/13C-SpaceM
    cd 13C-SpaceM
    conda env create -f environment.yml

Usage:

    conda activate sciso_env

Possible issues

It takes too long to solve environment

It's a widespread issue with using conda. Consider using mamba solver and refer to the official conda website.

Scanpy errors

This project uses Scanpy's pickled anndata objects to store the single-cell data. Sometimes different combinations of versions of pandas, anndata, pickle and scanpy can cause errors in reading the files or having imports work correctly. For possible troubleshooting take a look at the dump of the environment used for processing environment_dump.yml

Download data

Data for Hypoxia/Normoxia and ACLYkd experiments can be downloaded from Biostudies.

Data for mouse brain experiments can be downloaded from Metaspace project.

Single-cell workflow

SpaceM processing

Raw data for single-cell lipidomics consists of microscopy images and MALDI images. SpaceM software was used for correlating the modalities and producing single-cell data by overlaying mass spectrometry images with microscopy and attributing ablation marks with segmentation cell masks. This method is described in the corresponding paper: SpaceM reveals metabolic states of single cells.
SpaceM is a closed-sourced software, so for reproducibility we share the output of SpaceM.

Normally mass spec peak annotations downloaded from Metaspace are used in SpaceM pipeline, but Metaspace only generates isotopologue distributions based on natural isotope abundance, therefore here after generating all possible isotopologues for a given molecule, pyimzML is used to extract ion images and store in the format expected by SpaceM.

Therefore in this project we performed image registration and segmentation with SpaceM in a usual way until Metaspace annotation, then used extract_isotopologue_ion_images.py script to parse each imzML file and produce ion images for all isotopologues of the ions from molecule list.

    python sciso/extract_isotopologue_ion_images.py analysis/ion_list_fa.csv 20220322_AB_DKFZHypoxia/W2 

Example of extracted ion images and resulting objects is in analysis/00_extract_ion_images.ipynb. You don't need to download raw imzML files and rerun this step, as we share the output of SpaceM.

Single-cell analysis

Although SpaceM is used to register together pre-MALDI microscopy and MALDI ion images, signal normalization and single-cell analysis (aggregation of signal from each ablation mark overlapping with a given cell in a single cell's readout) steps are specific to this project, therefore they are implemented outside of the SpaceM pipeline.
SpaceM outputs results of analysis as a directory with the following structure:

    W1
    |-- config.json
    |-- analysis
    |   |-- ablation_mark_analysis
    |   |   |-- spatiomolecular_adata.h5ad
    |   |   |-- spatiomolecular_matrix.csv
    |   |-- cell_segmentation_external
    |   |   |-- cell.labels.tif
    |   |   |-- cell_segmentation.tif
    |   |-- ion_image_registration
    |   |   |-- ablation_mark.labels.tif
    |   |   |-- transformation.json
    |   |-- overlap_analysis2
    |   |   |-- ablation_mark.regions.csv
    |   |   |-- cell.regions.csv
    |   |   |-- overlap.labels.tif
    |   |   |-- overlap.regions.csv
    |   |-- single_cell_analysis
    |   |   |-- spatiomolecular_adata.h5ad
    |   |   |-- spatiomolecular_matrix.csv
    |   |-- transformation
    |   |   |-- ablation_mark_mask_pre_maldi_cropped.tif
    |   |   |-- cropped_pre_maldi_channels.json
    |   |   |-- pre_maldi_crop_frame.json
    |   |   |-- cropped_pre_maldi_channels
    |   |   |   |-- img_t1_z1_c0.tif
    |   |   |   |-- img_t1_z1_c1.tif
    |   |   |   |-- img_t1_z1_c2.tif
    |   |   |-- cropped_post_maldi_channels
    |   |   |   |-- img_t1_z1_c0.tif
    |   |   |   |-- img_t1_z1_c1.tif
    |
    |
    W2
    |--config.json
    |--analysis
    |  |--...
    ...

Run single-cell analysis for all wells on each slide:

    python sciso/sc_analysis.py $data_dir/20220322_AB_DKFZHypoxia/slide2/spacem_data $data_dir/20220322_AB_DKFZHypoxia/slide2/anndata
    python sciso/sc_analysis.py $data_dir/20220322_AB_DKFZHypoxia/slide3/spacem_data $data_dir/20220322_AB_DKFZHypoxia/slide3/anndata
    python sciso/sc_analysis.py $data_dir/20220411_AB_DKFZACLYac/slide1/spacem_data $data_dir/20220411_AB_DKFZACLYac/slide1/anndata
    python sciso/sc_analysis.py $data_dir/20220411_AB_DKFZACLYac/slide2/spacem_data $data_dir/20220411_AB_DKFZACLYac/slide2/anndata
    python sciso/sc_analysis.py $data_dir/20220411_AB_DKFZACLYac/slide3/spacem_data $data_dir/20220411_AB_DKFZACLYac/slide3/anndata

This script performs the following steps:

• Natural isotope abundance correction for each fatty acid in each ablation mark
• Averaging of signal in each cell by taking a median for each annotated m/z value
• Normalizing isotopologue distributions for each fatty acid for each cell

The results are saved in an anndata object joined for all wells on the slide. This analysis is done per slide to keep the file sizes manageable.

Quality control and data aggregation

After single-cell analysis data from all slides is aggregated together and quality control is performed, filtering out cells by size and number and total intensity of detected ions. Subsequently the cells are classified using normalized GFP intensity. Running all notebooks in analysis/hypoxia and analysis/ACLYkd produces anndata objects that contain all information per cell that is further used to generate manuscript plots. We share these files as well.

Reproducing manuscript figures

manuscript_figures contains notebooks that allow to reproduce plots for each figure panel based on the anndata objects produced in the previous step and SpaceM output (for registered microscopy images and segmentation). Figures 2-4 contain plots based on single-cell data.

Mouse brain tissue image analysis

For the tissue imaging analysis the analysis was done per pixel of the MALDI image instead of per cell. In each pixel we do the same analysis steps as for the single-cell data:

• Perform natural isotope abundance correction for each ion's isotopologue distribution
• Fit binomial distribution to each fatty acid's isotopologue distribution

To reproduce Figure 5 (panels c and d) and Figure 6 (panel a) do the following:

• Download imzml for the dataset 2023-09-25_ME_PattiLab_MouseBrain_4-4_010G_AIF_600-1600_30NCE_100-600_10umss_285x230_32at from the Metaspace project
• Run the script
  python analysis/brain/00_extract_ion_images.py $data_dir/raw/2023-09-25_ME_PattiLab_MouseBrain_4-4_AIF_600-1600_30NCE_100-600_10umss_285x230_32at.imzML analysis/brain/ion_list_fa.csv output_dir `