A multi-omics re-analysis of Fezf2-mediated cortical development using the public GSE153164 dataset (Di Bella et al. 2021, Nature). This project investigates the temporal and cell-type-specific mechanisms by which Fezf2 loss disrupts cortical development across the E10–P4 timeline, leveraging 20 scRNA-seq samples (wild-type time course; Fezf2-heterozygous and Fezf2-knockout at E13, E15 and P1) and 3 scATAC-seq samples (wild-type E13.5, E15.5, E18.5).
This is an independent re-analysis: all preprocessing, integration, cell-type annotation, dose-response modelling and regulatory analysis were re-run from the raw count matrices with an explicit, statistically conservative framework (described below).
- Integrated 121,889 cells (94.7% of 128,746 pre-QC cells) × 20,982 genes across 20 scRNA-seq samples, 13 developmental stages and 3 genotypes.
- Annotated 17 cell types from 21 Leiden clusters using marker-gene scoring.
- Median 2,141 genes/cell, 5,179 UMI/cell, 4.15% mitochondrial reads.
- At P1, Fezf2⁺ subcerebral projection neurons (SCPNs) are depleted in a dose-dependent manner: 2.21% of wild-type cells (mean of 2 replicates: rep1 3.17%, rep2 1.24%) vs 0.18% in heterozygotes (Female 0.186%, Male 0.164%) vs 0.15% in knockouts — a ~13–15-fold reduction.
- Fezf2 expression within SCPNs shows a dosage gradient (WT > Het > KO mean log1p), consistent with a cell-autonomous requirement.
- Most genotypes are represented by a single library per stage, so formal
per-gene p-values are not identifiable without pseudoreplication. We therefore:
- report effect sizes (log2 fold-changes) with bootstrap 95% CIs over cells;
- estimate a biological-noise null from the two P1 wild-type replicates (99th percentile |log2FC| = 0.342); and
- flag genes whose KO-vs-WT effect exceeds this null.
- Genes exceeding the null in KO: 947 (Maturing Projection Neurons), 1,119 (Corticothalamic Neurons), 838 (Callosal Projection Neurons).
- Because wild-type is only available at P1, the 3-genotype dose-response is valid only at P1; E13/E15 are reported separately as 2-genotype Het-vs-KO contrasts.
- KO-downregulated genes across all three projection-neuron types are strongly enriched for cytoplasmic translation / ribosome biogenesis (e.g. Corticothalamic Neurons: Cytoplasmic Translation adjusted p = 8.8×10⁻³⁰; Macromolecule Biosynthetic Process adjusted p = 3.1×10⁻²³), pointing to a neuronal maturation/biosynthetic defect.
- CollecTRI/decoupler regulon activity at P1. Top up in KO: Six4, Atf5, Wt1, Myt1. Top down in KO: Hdac7, Hlx, Kdm5a, Rest. Fezf2 regulon activity itself is near-zero change, consistent with Fezf2 acting as a dosage-sensitive repressor rather than through its own regulon abundance.
Source: GEO GSE153164 (Di Bella et al. 2021, Nature 595:554–559). SuperSeries of GSE153162 (scRNA-seq) and GSE153163 (scATAC-seq).
| Data type | Samples | Conditions |
|---|---|---|
| scRNA-seq | 20 | WT time course (E10–P4); Het (E13, E15, P1 F/M); KO (E13, E15, P1) |
| scATAC-seq | 3 | WT (E13.5, E15.5, E18.5) |
QC summary: 121,889 cells retained (94.7%) across 20,982 genes from 20 samples. Median 2,141 genes/cell, 5,179 UMI/cell, 4.15% mitochondrial reads.
The re-analysis is organised as sequential phases in the Python scverse ecosystem.
Phase 1 ──► Phase 2 ──► Phase 3 ──► Phase 4
Preprocess Temporal Dose- Regulon /
& integrate analysis response TF activity
- Phase 1 — Preprocessing & integration (
notebooks/01_preprocessing.ipynb): MAD-based QC, Scrublet + Solo doublet detection, Harmony + scVI integration, Leiden clustering. Output: integrated AnnData (121,889 × 20,982). - Phase 2 — Temporal analysis (
notebooks/02_temporal_analysis.ipynb): marker-based cell-type annotation (17 types), PAGA + diffusion pseudotime, WT-vs-KO differential expression at P1. - Phase 3 — Dose-response (
notebooks/03_dose_response.ipynb): WT→Het→KO dosage modelling at P1, per-cell-type dose-response tables, WT-replicate null, SCPN/CTHN compositional analysis, Het-vs-KO contrasts at E13/E15. - Phase 4 — Regulons / multi-omics (
notebooks/04_multiomics_grn.ipynb): CollecTRI TF-activity inference (decoupler) by genotype at P1.
Note: an earlier exploratory therapeutic-target module and a conceptual validation notebook were removed from this release. The therapeutic analysis was out of scope for the current manuscript, and the validation notebook contained illustrative (hard-coded) curves rather than data-derived results; neither met the reproducibility standard of this release.
fezf2-multiomics/
├── data/
│ ├── raw/ # 23 raw 10x .h5 matrices (GSE153164)
│ ├── filtered/ # Post-QC, pre-integration object + QC tables
│ ├── integrated/ # adata_integrated.h5ad (Harmony + scVI)
│ └── derived/ # adata_annotated.h5ad (17 cell types + pseudotime)
├── result/
│ ├── figures/
│ │ ├── main/ # Figures 1–5 (PNG + SVG)
│ │ └── supplement/ # Figures S1–S7 (PNG + SVG)
│ └── tables/
│ ├── main/ # Main-text tables (QC, composition, SCPN, dose, regulons)
│ └── supplement/ # Full DE, enrichment, per-cell-type dose, markers, QC
├── notebooks/ # Analysis notebooks (Phases 1–4 + figures)
├── docs/ # Setup, bioRxiv checklist, cover letter, reviewer notes
├── manuscript_fezf2_multiomics.md # Manuscript (Markdown)
├── manuscript_fezf2_multiomics.docx # Manuscript (Word, submission-ready)
├── environment.yml # Conda environment
├── requirements.txt # pip environment
├── CITATION.cff # Citation metadata
└── README.md
- Python ≥ 3.11
- RAM ≥ 64 GB recommended (integrated object ≈ 2 GB on disk; integration is memory-bound)
git clone <repo-url>
cd fezf2-multiomics
# Option A: conda
conda env create -f environment.yml
conda activate fezf2-multiomics
# Option B: pip
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txtExecute the notebooks in order (01 → 04). Raw count matrices live in data/raw/
(GEO GSE153164); each phase reads the AnnData objects written by the previous phase
(data/filtered → data/integrated → data/derived) and writes result tables and
figures to result/tables/ and result/figures/.
Figures in
result/figures/are provided as both PNG (preview) and SVG (editable vector, text preserved) for all main and supplementary figures.
| Category | Tools |
|---|---|
| Core framework | Scanpy, AnnData |
| Integration | Harmony, scVI |
| Doublets | Scrublet, Solo (scvi-tools) |
| Trajectories | PAGA, diffusion pseudotime |
| DE / statistics | effect sizes + bootstrap CIs, WT-replicate null, PyDESeq2 (where identifiable) |
| Enrichment | gseapy / Enrichr |
| Regulons | decoupler + CollecTRI |
| Visualisation | Matplotlib, Seaborn |
- Di Bella DJ, et al. (2021) Molecular logic of cellular diversification in the mouse cerebral cortex. Nature 595:554–559.
- Tsyporin J, et al. (2021) The transcription factor FEZF2 is a dosage-sensitive repressor of corticofugal neuron identity. Cell Reports.
- Aibar S, et al. (2017) SCENIC: single-cell regulatory network inference and clustering. Nature Methods 14:1083–1086.
- Squair JW, et al. (2021) Confronting false discoveries in single-cell differential expression. Nature Communications 12:5692.
MIT (see LICENSE). Intended for academic research use.
- Di Bella et al. and the Arlotta lab for generating and sharing GSE153164.
- The scverse community and the developers of Scanpy, scvi-tools, decoupler and the other tools used here.