a, Representative IHC for Hes1 and cleaved caspase-3 (CC3) in serial TKO tumour sections. Inset: higher magnification of a positive control for CC3 (tumour from a mouse treated with chemotherapy). b, Predicted numbers and ratios of GFPneg and GFPhigh tumour cells if the two populations divide independently of each other, and GFPneg cells cycle approximately three times faster than GFPhigh cells. c, Representative IHC in serial sections from TKO Hes1GFP/+ tumours initiated by Adeno-CGRP-Cre. d, Quantification of the frequency (Freq.) of Hes1pos cells in TKO hyperplasias (n = 23, from five mice) and tumours (n = 50, from seven mice) induced by Adeno-CGRP-Cre. Scores: 0 (0%), 1 (1–20%), 2 (20–60%), 3 (>60%). e, Flow cytometry analysis showing the percentage of GFPhigh cells in pooled tumours from TKO Hes1GFP/+ mice (n = 4) infected with Ad-CGRP-Cre. f, Images of freshly isolated GFPneg cells grown on dishes coated with Dll4 or PBS control (representative of n = 3 biological replicates). g, h, Freshly isolated GFPneg cells that remained GFPneg after culture on Dll4-coated dishes were re-plated on dishes coated with Dll4 ligand (+Dll4) or PBS control (-Dll4). g, Flow cytometry and images (representative of n = 2 biological replicates). h, GFPneg and GFPhigh cells that formed after this second round of Dll4 stimulation were sorted and analysed by immunoblot. Control: GFPhigh cell line. i, Relative number of GFPhigh cells formed from freshly isolated GFPneg cells grown on Dll4 after 2 weeks of tarextumab treatment (n = 3 biological replicates each). j, Single-cell qRT–PCR (n = 45 each) of H29, H82, and H889 human SCLC cell lines. Heatmap was generated by unsupervised clustering of each cell line. Dark blue regions indicate undetectable expression. k, qRT–PCR for HES1 after 72 h of culture with or without Dll4. Data are normalized to GAPDH (n = 3 biological replicates with n = 3 technical replicates each). l, m, GFPhigh cell lines were treated with DMSO or DBZ while grown with or without Dll4 or co-cultured with three individual NE cell lines in the absence of Dll4. GFP expression was analysed by flow cytometry after 72 h. l, Flow cytometry of GFPhigh cell line 1 (representative of n = 3 biological replicates; GFP intensity quantified in Fig. 2e). m, Quantification of GFP intensity in GFPhigh cell line 2; relative median GFP intensity normalized to the ‘DMSO-Dll4’ condition (n = 3 biological replicates). n, o, Representative images (n) and qRT–PCR (o) of GFPhigh cell lines cultured in the presence of absence of Dll4 for more than a month (n = 3 biological replicates with n = 3 technical replicates). p, Freshly isolated GFPneg cells that became GFPhigh after culture on Dll4 were re-plated on dishes coated with Dll4 ligand or PBS control and analysed by immunoblot after a month. GFPneg cell lysate: positive control for Ascl1. *P < 0.05; **P < 0.01. Statistical significance was determined by two-tailed paired Student’s t-test. Data are represented as mean ± s.d. Scale bars, 50 μm.