Fig. 1. Candidate regulators of a fractionally penetrant phenotype identified by stochastic frequency matching.

a Molecular design of NGFR-EGFR-FRB¹⁰ and NGFR-Erbb2–FKBP⁹ chimeras heterodimerized by AP21967 (AP). MCF10A-5E cells²⁰ were serially transduced–subcloned to generate B2B1 cells. b B2B1 cells were stained for Golgi polarity (GM130, green) and proliferation (phospho-Rb [pRb], magenta), counterstained for nuclei (DAPI, blue) and membrane-ECM (wheat germ agglutinin [WGA], white). One DCIS escapee (DE) in the plus-AP condition is outlined. Scale bar is 20 µm. c Fractional penetrance of the ErbB-induced DE phenotype in B2B1 cells. B2B1 cells were cultured in 3D for 9 days ±0.5 µM AP for the final 3 days. Data are shown as the arcsine transformed mean ± s.e. from n = 8 biological replicates where >200 outgrowths were scored per replicate. Differences were assessed by a two-sided t test after arcsine transformation. d DEs (arrow) are not apparent until after 48 h of ErbB heterodimerization. B2B1 cells were 3D cultured for 6 days, treated with 0.5 µM AP, and serially imaged by brightfield microscopy. Scale bar is 200 µm. e Approach for stochastic frequency matching. After 24 h of ErbB heterodimerization, 3D outgrowths (n = 20) were randomly sampled by laser-capture microdissection and assessed by stochastic profiling^20,24. Measured transcript fluctuations were compared to the DE frequency estimated in c. f Outgrowth-to-outgrowth transcript heterogeneities identified by stochastic profiling of B2B1 3D cultures ±0.5 µM AP for 24 h. Microarray probeset intensities were hierarchically clustered (Ward’s linkage, row standardization). g Approach for quantifying transcript fluctuations. For each gene, the minus-AP condition (top) was used to specify 10th and 90th percentiles that score the number of outgrowths above (middle) and below (bottom) the minus-AP control for each gene. h Stochastic frequency matching gleans targets from outgrowth-to-outgrowth heterogeneities associated with ErbB heterodimerization. Probesets are jittered about the expectation for a binomial distribution (20 trials, 35% success probability) versus their estimated up-or-down expression frequency as in g. Probesets within the interquartile range (IQR) of the inverse binomial distribution (B⁻¹_20,0.35) were scored as high-priority, whereas those within the 90% confidence interval (CI) were considered candidates. Source data are provided as a Source Data file.