Fig 2: Perturb-seq functionally defines drug resistant cell states.a) Differential expression analysis of pathways from MAYA or PROGENy for HT-29 CBE and ABE perturb-seq screens. Heatmaps display log-fold changes for a given pathway-gRNA comparison, and statistical significance is denoted with a dot (significance at FDR < 0.1). b) Differential expression at the level of PROGENy pathway scores for drug addiction versus canonical drug resistance. For each gRNA the same number of iBARs was sampled to avoid biases resulting from an over-representation of individual gRNAs. c) Comparison of z-scores from proliferation read-out base editing screens to energy distance scores derived from perturb-seq screens. Variant classes based on the HT-29 proliferation screens in dabrafenib and cetuximab are indicated. Intermediate variants discussed in the text are labelled. d) Diffusion scores illustrate progressive levels of mutational impact for the CBE and ABE data set, with drug addiction variants having the highest scores and a range of different impact levels across the gRNAs conferring drug resistance. The intermediate variants KRAS E62K/E63K and KRAS K117R/E/D119G are highlighted. e) Volcano plots of significantly differentially expressed genes (vs NT control gRNA cells) from representative drug resistance gRNAs. B2M is downregulated by both variants. Significant down- and upregulation at FDR < 0.1 (Benjamini-Bogomolov correction) are indicated in blue and red respectively.

Fig 3: Perturb-seq functionally defines drug-resistant cell states.a, Schematic of perturb-seq screening to investigate the transcriptomic effects of variants conferring resistance to dabrafenib and cetuximab in HT-29 cells using base editing. DE, differential expression analysis. b, Uniform manifold approximation and projection colored by variant class and normalized energy distances (ed) between NT gRNA cells and drug-resistant cells in ABE HT-29 cells treated with the combination of dabrafenib (80 nM) and cetuximab (1 µg ml−1) for 16 h. c, Cell-cycle phase occupancy differences between cells with drug resistance conferring gRNAs and control gRNAs in the ABE perturb-seq dataset. P < 2.2 × 10−16 for ABE dataset, P < 1.5 × 10−14 for CBE dataset, chi-squared test, comparing control gRNAs with drug resistance gRNAs. d, Heatmap and hierarchical clustering of PROGENy pathway activity scores for each gRNA in the ABE dataset. e, Density plot of differences in PROGENy pathway scores between the variant groups for the ABE dataset. f, Volcano plot of differentially expressed genes between NT gRNA control cells and cells with the PI3K p110ɑ driver variant. Red, significantly downregulated transcripts (including B2M and HLA-A); blue, upregulated transcripts. g, Boxplot of PFS outcome score for each variant class, derived from CRC patients treated with BRAF, MEK and PD-1 inhibitor (PD-1i) combination therapy28. CBE and ABE perturb-seq scores are shown. ***P < 0.01, two-sided Wilcoxon rank-sum test compared with NT gRNAs. CBE; canonical drug resistance n = 7, driver n = 8, control n = 172, NT n = 39. ABE; drug addiction n = 11, canonical drug resistance n = 8, control n = 115, NT n = 39. Boxplots represent the median, IQR and whiskers are the lowest and highest values within 1.5× IQR. Control gRNAs are those that did not confer drug resistance in proliferation screens. See also Extended Data Figs. 7–9.