What this research found
Can a drug's mechanism be recovered from gene expression alone? Public microarray data from 16 MCF7 breast cancer samples, half treated with the chemotherapy agent doxorubicin, were run through differential expression testing, pathway enrichment, and a blind match against 17,389 known drug signatures. The three analyses converged on topoisomerase II inhibition, the established mechanism, with doxorubicin itself ranking second and the related anthracycline daunorubicin third in the signature search, alongside sharp suppression of cell cycle pathways.
- The response is cell-wide rather than narrow: 16,183 genes changed significantly at a false discovery rate below 0.05 with at least a two-fold change, roughly 65% of the detected transcriptome.
- Histone genes were the most strongly suppressed — HIST1H4B at log2 fold change -6.03, HIST1H4C at -5.98 (false discovery rate 3.5e-18), and HIST2H3A at -5.80 — along with the cell cycle regulators CDK1, CCNB1, and CDC20, all below -3.0. Because histone production is coupled to DNA replication, this pattern points to replication stress.
- Enrichment analysis flagged 89 significantly modulated pathways. The most suppressed were mitotic spindle (normalized enrichment score -2.94) and the G2-M checkpoint (-2.53), followed by DNA replication (-2.45) and E2F targets (-2.38), while ribosome (3.30), oxidative phosphorylation (3.18), and p53 signalling (2.15) were enriched.
- The blind signature search recovered the correct compound. Querying 1,000 of the most-changed genes against 17,389 drug profiles ranked doxorubicin second at a false discovery rate of 2.1e-99 and daunorubicin, another anthracycline topoisomerase II inhibitor, third at 1.1e-98; other topoisomerase inhibitors and DNA-damaging agents appeared in the top 50.
- All three independent lines of evidence point to the same cascade: topoisomerase II inhibition producing DNA double-strand breaks, p53 activation, G2-M arrest, and apoptosis — reconstructed here from expression data rather than assumed at the outset.
- Oxidative phosphorylation genes were among the most strongly induced (normalized enrichment score 3.18), consistent with the mitochondrial stress that the dataset's original authors were investigating as a route to doxorubicin resistance.
How it was done
A public microarray dataset of MCF7 breast cancer cells — 16 samples, 8 vehicle controls and 8 doxorubicin-treated — was retrieved from NCBI's Gene Expression Omnibus, quantile-normalised, and mapped from probe identifiers to gene symbols. Principal component analysis confirmed that the first component, carrying 67% of the variance, separated treated from control samples. Per-gene Welch's t-tests with Benjamini-Hochberg correction identified differentially expressed genes, and genes ranked by signed significance were fed into gene set enrichment analysis against the MSigDB Hallmark, KEGG, and Reactome collections using 1,000 permutations. The top 500 up-regulated and top 500 down-regulated genes were then queried against the Drug Signatures Database through Enrichr to find which known compounds produce the most similar transcriptional response. Outputs included a volcano plot, a top-50 gene heatmap, an enrichment summary, a connectivity ranking, an evidence dashboard, and a mechanism schematic.
Data sources
- NCBI Gene Expression Omnibus, GSE244574 — 16 MCF7 samples on the Affymetrix Human Gene 2.1 ST array
- MSigDB Hallmark gene sets, KEGG 2019 Human, and Reactome 2016 pathway collections
- Drug Signatures Database (DSigDB) — 17,389 drug-induced expression signatures, queried through Enrichr
- Original dataset study on mitophagy in doxorubicin-treated breast cancer cells (PubMed 38514650)
Limitations
The data come from one cell line at a single time point and a single doxorubicin concentration, so they cannot show how the response varies by breast cancer subtype or evolves over time. Signature matching is inference from expression data rather than direct biochemistry — no topoisomerase II activity assay was run — and doxorubicin's known reactive oxygen species activity did not emerge as a dominant pathway signal.
Figures from this analysis
Outputs produced
Selected GEO dataset metadata (GSE244574)
GEO dataset search script using NCBI E-utilities API
expression_matrix.csv
Gene-level expression matrix (16,173 genes × 16 samples, log2-transformed)
sample_metadata.csv
Sample annotations with condition labels (8 Control, 8 Treatment)
Main data download and QC script
Probe-to-gene mapping using GPL21185 SOFT annotations
Raw GEO series matrix file (4.2MB)
GPL21185.soft
Platform annotation file (2.3GB)
How this research was produced
K-Dense Web planned and ran this cancer biology investigation end to end — gathering the sources, carrying out the analysis, producing the figures, and drafting the report. The full session transcript, including every intermediate step, is available to view.
