HGG-oncohistones analysis code

• This is the public repository accompanying the study K27M in canonical and noncanonical H3 variants occurs in distinct oligodendroglial cell lineages in brain midline gliomas, Jessa*, Mohammadnia*, Harutyunyan*, ..., Jabado, Kleinman, Nature Genetics, 2022
• Link to repository: https://github.com/fungenomics/HGG-oncohistones and rendered version: https://fungenomics.github.io/HGG-oncohistones/
• This repository is archived on Zenodo
• This repository contains primarily code, with some figure source data, see the Data availability section for links to data
• Jump to the Code to reproduce figures section for links to code and rendered HTMLs of each figure

Contents:

• Contents of the repository
  • Codebase overview
  • Codebase structure
• Materials for the manuscript
  • Links to code to reproduce figures/main results
  • Palettes & custom plotting utilities
  • Tables
• Processed single-cell data
  • scRNAseq
  • scATAC
  • scMultiome
  • Cell annotations matching the paper
  • Data integration
• Re-analysis of human fetal brain scRNAseq data
• Data availability
• Notes for reproducibility
• Citation

Contents of this repository

Codebase overview

• This repository is meant to enhance the Materials & Methods section by providing code for the custom analyses in the manuscript and the exact R dependencies, in order to improve reproducibility for the main results. However, it is not a fully executable workflow.
• In general, read alignment and cell calling for tumor and normal brain single-cell sequencing data has been performed using in-house pipelines, not included here. This repository mainly contains custom/downstream code.
• A first level of downstream analysis involves scripts applied in parallel to individual samples for specific data types. Copies of these scripts are provided in the code/scripts/ and R-4/code/scripts folders (the execution of these scripts is perofrmed in the data/scRNAseq, data/scATACseq, data/ChIPseq, etc folders, not included here).
• A second level of downstream analysis involves custom analyses, aggregating samples and data types, use to derive the main results included in the paper. These are provided in .Rmd files in code and R-4/code, with the associated .md and rendered HTML files. The rendered HTML files can be viewed at https://fungenomics.github.io/HGG-oncohistones/ under the "Code to reproduce key analyses" section.

Codebase structure

Brief explanation of the directory structure:

• renv --> renv-managed folder for R 3.6
• renv.lock --> lockfile containing all package versions for R 3.6 analysis
• code --> code for R 3.6 analysis, contains the .Rmd files that run the high-level analyses and produce figures included in the paper
  • functions --> contains .R files with custom functions used throughout the analysis
  • scripts --> contains .R and bash scripts for analyses that are repeated on individual samples, as well as helper scripts e.g. for creating references
  • infos_templates --> contains example config files for scripts in the scripts folder
  • seurat_v3_resources --> this is a copy of the folder referenced by SEURAT_V3_ASSETS in certain bash scripts, or file.path(params$assets, "resources") in the various preprocessing_*.Rmd scripts. It contains resources used for single-cell preprocessing.
• R-4 --> code for R 4.1 analysis (has a similar directory structure as the above main directory)
  • code --> contains .Rmd files, functions, and scripts for R 4.1 analysis
  • renv --> renv-managed folder for R 4.1
  • renv.lock --> lockfile containing all package version for R 4.1 analysis
• include --> contains templates, palettes, etc, for this repository
• rr_helpers.R --> contains helper functions for working with this GitHub repository template (rr)

Materials for the manuscript

Code to reproduce key analyses

Code to reproduce analyses is saved in code and R-4/code. (See here for why two different R versions are used.) When these analyses depend on inputs from pipelines, I've tried to note within the R Markdown documents where these scripts/pipelines are located.

This table contains pointers to code for the key analyses associated with each figure. The links in the Analysis column lead to rendered HTMLs.

Figure	Analysis	Path
Fig 1	Oncoprints summarizing tumor and cell line cohort	./code/00-oncoprints.Rmd
Ext Fig 1	Summary figures for extended mouse brain scRNAseq atlas	./code/05-mouse_atlas.Rmd
Fig 1, Ext Fig 2	cNMF analysis of variable gene programs	./R-4/code/01-cNMF_programs.Rmd
Fig 1	Cell type identity in tumors with automated consensus projections	./R-4/code/02-consensus_projections.Rmd
Ext Fig 2	Analysis of human fetal thalamus and hindbrain scRNAseq data	./code/01A-human_thalamus.{Rmd,html} and ./code/01B-human_hindbrain.{Rmd,html}
Ext Fig 2	Validation of cell type projections using human thalamic fetal brain reference	./R-4/code/02-consensus_projections.Rmd
Ext Fig 3	Characterization of malignant ependymal cells	./code/01C-ependymal_cells.Rmd
Fig 2, 4	Scatterplots for RNAseq/K27ac/K27me3 between H3K27M tumor subtypes	./code/02-bulk_comparisons.Rmd
Fig 2, Ext Fig 4	Systematic HOX analysis/quantification	./code/03A-HOX.Rmd
Fig 3	Analysis of thalamic patterning	./code/04-thalamus.Rmd
Fig 4-6, Ext Fig 4-6	Analysis of dorsal-ventral patterning and NKX6-1/PAX3 activation	./code/03B-NKX61_PAX3.Rmd
Fig 6	Analysis of ACVR1 cell lines	./code/06-ACVR1.Rmd
Fig 7, 8, Ext Fig 7	Analysis of histone marks in tumors & cell lines	./code/07A-histone_marks.Rmd
Ext Fig 8	Comparison of tumor epigenomes to scChIP of normal cell types	./R-4/code/03A-celltype_epigenomic_similarity.Rmd
Fig 8, Ext Fig 9	Heatmaps of H3K27me2/3 in CRISPR experiments	./code/07D-deeptools_*.sh and ./code/07E-deeptools_*.sh

Palettes & custom plotting utilities

Most color palettes (e.g. for tumor groups, genotypes, locations, cell types, HOX genes, etc) and ggplot2 theme elements (theme_min(), no_legend(), rotate_x(), etc) are defined in include/style.R.

Tables

Supplementary tables (included with the manuscript) and processed data tables (on Zenodo) were assembled from the following input/output/figure source data files. (Only tables produced with the code included here are listed below.)

Supplementary table	Path
6	./output/05/TABLE_mouse_sample_info.tsv
7	./output/05/TABLE_mouse_cluster_info.tsv
8	./R-4/output/02/TABLE_cNMF_programs_per_sample.tsv
9	./R-4/output/02/cNMF_metaprogram_signatures.malignant_filt.tsv
10	./R-4/output/02/TABLE_reference_cnmf_program_overlaps.tsv
11	./output/01A/TABLE_thalamus_QC.tsv and ./output/01B/TABLE_hindbrain_QC.tsv
12	./output/01A/info_clusters3.tsv and ./output/01B/info_clusters3.tsv
13	./output/03A/TABLE_HOX_expression_per_transcript.tsv
14	./output/03A/TABLE_HOX_H3K27ac_H3K27me3_per_transcript.tsv
16	./figures/03B/enhancer_diff-1.source_data.tsv

Processed data table	Path
1a	./output/02/TABLE_bulk_counts.tsv
1b	./output/02/TABLE_dge_H3.1_vs_H3.3.tsv
1c	./output/02/TABLE_dge_thal_vs_pons.tsv
2a	./output/07A/TABLE_K27me3_CGIs.tsv
2b	./output/07A/TABLE_K27me2_100kb_bins.tsv
3a	./output/02/TABLE_promoter_H3K27ac_H3K27me3_per_sample.tsv

Processed single-cell data

This section describes the scripts used for preprocessing of single-cell data from this project. That includes: sn/scRNAseq, scATACseq, and scMultiome (joint RNA & ATAC in the same cells). This document refers to sn and scRNAseq generally as 'scRNAseq'. Please see the sample metadata for the technology used to profile each sample. Please see the Methods section of the manuscript for more details on the single-cell profiling.

Single-cell RNAseq data

The pipeline for scRNAseq processing applied per-sample is summarized in this schematic. In general, scripts contain the code to run the analysis and config files contain the parameters or setting specific to a certain iteration of the analysis.

Following Cellranger, the scRNAseq samples have all been processed with the lab's preprocessing workflow (./code/scripts/scRNAseq_preprocessing.Rmd). Each sample is then subject to several downstream analyses as described in the schematic above, with the associated scripts indicated.

Single-cell ATACseq data

The pipeline for scATACseq processing applied per-sample is summarized in this schematic:

Following Cellranger, preprocessing of the scATAC data is done with a script that builds off the scRNAseq workflow, at ./R-4/code/scripts/preprocessing_scATAC.Rmd. This workflow is run in the scATAC pipeline at ./R-4/data/scATACseq/pipeline_10X_ATAC, with one folder per sample. Each sample is then subject to several downstream analyses as described in the schematic above, run in that sample's folder, with the associated scripts.

Single-cell Multiome data

The pipeline for scMultiome processing applied per-sample is summarized in this schematic:

Following Cellranger, preprocessing of the scMultiome data is done with a script that builds off the scRNAseq workflow, at ./R-4/code/scripts/preprocessing_scMultiome.Rmd. This workflow is run in the scMultiome pipeline at ./R-4/data/scMultiome/pipeline_10X_Multiome, with one folder per sample. Each sample is then subject to several downstream analyses as described in the schematic above, run in that sample's folder, with the associated scripts.

Cell annotations matching the paper

For scRNAseq, scATACseq and scMultiome samples, the cell metadata provided with the paper contains several columns matching the analyses used in the paper:

• Cell_type_granular_mouse_correlations --> cell-type projection to the extended mouse atlas, based on the Spearman correlation, using the cluster label (REGION-TIMEPOINT_CLUSTER)
• Cell_type_mouse_correlations --> cell-type projection to the extended mouse atlas, based on the Spearman correlation, summarized to a broader cell class (ontology is described in Table S7)
• Cell_type_consensus_Jessa2022--> consensus cell-type projection to the extended mouse atlas, based on agreement between Spearman correlation and at least one other cell-type projection method. Cells without a consensus are classified as "Uncertain", see Methods for details
• Malignant_normal_consensus_Jessa2022 --> assignment as normal or malignant, used to decide whether cells should be included in downstream analyses

The cell annotations/metadata are included in processed data deposition on Zenodo and on GEO (GSE210568).

Data integration

As described in the Methods, we used the harmony package for integration of single-cell datasets.

• Samples within each subgroup were integrated, using "Sample" and "Technology" as the batch variables
• Only malignant cells were used, and only the RNA modality was used for integration
• The general script used for integration is at code/scripts/integrate_harmony.R, which expects a config file info.experiment.tsv to be present (example at code/infos_templates/harmony.info.experiment.tsv)
• Integrations were performed at R-4/data/integrations, with one directory for each group of samples being integrated, and the config files within

Human fetal brain scRNAseq data

Human fetal data brain data for the hindbrain and thalamus were obtained from two studies, Eze et al, Nature Neuroscience, 2021, and Bhaduri et al, Nature, 2021.

• The count matrices for each sample, as produced by Cellranger, were downloaded from the BICCN repository.
• The data was processed using the scRNAseq preprocessing workflow as described above for human tumor samples (./code/scripts/scRNAseq_preprocessing.Rmd)
• Thalamic samples were labelled using the published cluster labels, together with manual correction based on known cell type markers (performed in ./code/01A_2-human_thalamus.{Rmd,html})
• Hindbrain samples were labelled by projection to the mouse atlas (described in ./code/01B-human_hindbrain.{Rmd,html})

Data availability

• For scRNA-seq data for normal E10, E13, E16 and E18 murine samples, data has been deposited to GEO (GSE188625), whereas E12, E15, P0, P3 and P6 samples have been previously deposited to GEO (GSE133531) as part of Jessa et al, Nature Genetics, 2019.
• For tumor samples and patient-derived cell lines, please see the associated Zenodo record or for the processed data including counts matrices, cell annotations, and genome wide histone mark levels. The counts and cell annotations are also deposited to GEO deposition (GSE210568).
• Raw data have been deposited to GEO (GSE188625) and EGA (EGAS00001005773)

Notes for reproducibility

rr template & helpers

This repository uses the rr template, which contains a set of R markdown templates to help me ensure reproducibility. Secondly, this also provides a set of helper functions (located in rr_helpers.R and prefixed by rr_ in the function name) to help encourage documentation.

R and R package versions

The R libraries for this project are managed with the package renv. The R versions used are 3.6.1 and 4.1.2, and renv manages one library for each R version.

The renv package:

1. maintains two isolated project-specific libraries in the renv folder (for R 3.6.1) or R-4/renv folder (for R 4.1.2) - the libraries themselves are not on GitHub
2. stores packages according to version
3. records the R, Bioconductor, and package versions in the files renv.lock and R-4/renv.lock, which can be used to reproduce the R package environment.

The reason for using two different R versions is that certain analyses involving 10X Multiome data require versions of Seurat/Signac dependent on R > 4.

R Markdown

Each markdown/HTML file has a "Reproducibility report" at the bottom (example), indicating when the document was last rendered, the most recent git commit when it was rendered, the seed, and the R session info.

Testing

Lightweight testing is performed in certain cases (e.g. validating metadata) using the ensurer package, combined with the testrmd testing framework for R Markdown documents. Certain reusable ensurer contracts (reusable tests) are stored in ./code/functions/testing.R.

GitHub / version control

The following are tracked / available on GitHub:

• .Rmd files, containing the code, and .md and rendered HTML files, containing code and outputs
• Figure source data, under figures, when sufficiently small
• The brief desc files for outputs, under outputs
• The lockfile produced by the renv package

The following are not tracked / available on GitHub:

• Figures in png/pdf format, and some figure source data
• Raw data and analysis output / processed data files (but see the "Data availability" section for links to data depositions)
• The actual packages in the R library

Citation

If you use or modify code provided here, please cite this work as follows:

Selin Jessa, Steven Hébert, Samantha Worme, Hussein Lakkis, Maud Hulswit, Srinidhi Varadharajan, Nisha Kabir, and Claudia L. Kleinman. (2022). HGG-oncohistones analysis code. Zenodo. https://doi.org/10.5281/zenodo.6647837

Acknowledgements

• Thank you to the Kleinman Lab and particularly Bhavyaa Chandarana (@bhavyaac) for code review and feedback
• The "Reproducibility report" section of each R Markdown was inspired by workflowr