AlbaoRunx3Manuscript (In Revision)

(Manuscript in Revision!) Code and analysis assets for the manuscript investigating RUNX3-dependent regulation in CD8 T cells. The repository collects single-cell RNA-seq processing notebooks, ATAC-seq differential accessibility workflows, figure exports, and helper scripts used throughout the study.

Repository layout

• single_cell/ — Scanpy-based notebooks and outputs for preprocessing, clustering, differential expression, pathway enrichment, signature scoring, and RNA velocity analyses.
• atac/ — ATAC-seq peak filtering, differential accessibility, peak set enrichment analysis (PSEA), and motif analyses implemented in Jupyter notebooks and exported HTML reports.
• csv/ — Compressed intermediate tables for clustering, differential expression, GSEA outputs, and figure-ready summaries.
• figures/ — Generated figure assets grouped by panel.
• reanalysis/ — Supplemental GSEA notebook for comparative datasets.
• scripts/ — Utility R scripts for flow cytometry processing, SEA plotting, and notebook conversion.

Analysis overview

Single-cell RNA sequencing

• P14 CD8 T cells were hash tagged with BioLegend TotalSeq A antibodies (A0301–A0309) before pooling and loading 100,000 multiplexed cells across two 10x Genomics Single Cell 3' v3.1 GEMs.
• Hash tag oligo (HTO) libraries were amplified from cDNA with the BioLegend HTO additive primer, while gene expression (GEX) libraries followed the standard 10x protocol. Sequencing depth targeted ~35,000 reads per cell for GEX and ~300 reads per cell for HTO on an Illumina NextSeq 2000.
• CellRanger v7.1.0 with the 2020-A mm10 reference generated count matrices. Quality control in Scanpy v1.9.5 included SoupX ambient RNA correction, Scrublet and scDblFinder doublet removal, and filtering on mitochondrial content (>5%), UMI count (<3,000), and detected genes (<1,250). Cell cycle effects were regressed using S and G2/M gene lists, and Leiden clustering at resolution 1.0 (14–16 clusters) guided downstream analyses.
• Differential expression relied on the Scanpy Wilcoxon rank-sum test (v1.9.6), with GSEA run on Wilcoxon statistics via clusterProfiler v4.14.0 and DOSE v4.0.0 (R v4.4.2). Transcriptome correlations used Spearman coefficients on mean-normalized counts of the top 2,000 variable genes. RNA velocity inputs were produced with velocyto v0.17 and analyzed with scVelo v0.3.1.

ATAC-seq

• Nuclei from sorted P14 CD8 T cells were lysed and permeabilized for in situ Tn5 transposition (Nextera DNA Library Preparation kit), titrated to 1.25 μL enzyme per 5×10⁴ nuclei in a 50 μL reaction. Libraries were PCR-amplified to optimal cycles, quality-checked on an Agilent TapeStation (targeting a 2:1 170–280 bp to 280–390 bp fragment ratio), and sequenced on an Illumina NextSeq 2000 at ~0.9× whole-genome coverage (paired-end 50×2).
• Processing and peak calling used the nf-core/atacseq v2.1.2 pipeline (Nextflow 24.04.2) against mm10 in broad-peak mode. Public datasets (e.g., GSE111149, GSE88987, GSE144383, GSE131871, GSE213041) were merged with study data to define consensus peaks; peaks present in at least one replicate from three datasets yielded 47,192 shared peaks from an initial 285,385. Differential accessibility employed DESeq2 v1.46.0, PSEA used clusterProfiler/DOSE, and motif analysis relied on HOMER v5.1. Peak overlaps with ChIP-seq datasets (RUNX3, TBET, c-JUN) were computed with bedtools v2.31.1 (intersect -f 0.5 -F 0.5), with accessibility profiles generated via deepTools v3.5.6.

Gene and peak resources

• CD8 T cell-specific gene lists follow prior definitions, supplemented by re-analyses of public bulk RNA-seq datasets (e.g., Tcf7-reporter, ex-KLRG1, Tox genotypes, terminal-TEM). Reads were mapped with Salmon v1.10.2 (mm39, Ensembl 110) using dataset-specific k-mer parameters, with differential expression performed via DESeq2 v1.46.0 on TPM values.
• ATAC-seq peak lists were generated ad hoc from the consensus procedure above to support PSEA.

Computational reproducibility

Containerized tools were used end to end:

• Public images:
  • CellRanger v7.1.0 (docker.io/litd/docker-cellranger:v7.1.0) for alignment and count matrix generation.
  • bedtools v2.31.1 (docker.io/staphb/bedtools:2.31.1) for peak overlaps.
  • deepTools v3.5.6 (quay.io/biocontainers/deeptools:3.5.6--pyhdfd78af_0) for accessibility profile generation.
  • Salmon v1.10.2 (docker.io/combinelab/salmon:1.10.2) for transcript quantification.
• Custom DockerHub images:
  • HOMER v5.1 (docker.io/pipkinlab/homer:5.1) for motif analysis.
  • Scanpy v1.9.5 (docker.io/pipkinlab/scanpy:1.9.5) for core scRNA-seq QC and clustering.
  • Scanpy v1.9.6 with scVelo (docker.io/pipkinlab/scanpy:1.9.6) for differential expression and RNA velocity.
  • R 4.4.2 with DESeq2 v1.46.0 and plotting stack (docker.io/pipkinlab/r-deseq2-plotting:4.4.2) for ATAC-seq differential accessibility and visualization.

Gene and peak sets are available at ScrippsPipkinLab/CommonGeneSets. Code for this study is hosted in this repository.